Release Notes
SDW AIKO
SDW AIKO is the core product of the SDW platform. Below is the complete version release history.
Current Version
v3.1.17 (Latest)
Release Date: October 23, 2026
Release Note
This release introduces the personal assistant tim and proactive duty: tim supports a custom avatar and nickname, DingTalk-style bubble conversations, visual card replies, and tracking BiBrain tasks assigned to you; "Scheduled Tasks" is officially renamed "Duty", allowing AI to draft rules from a single sentence that only take effect after your confirmation, with due reminders appearing in the bottom-right corner of the desktop. It also adds an Artifact Library for managing generated files, a conversation visibility switch to keep private conversations from being exposed, cloud experts connected to the BiBrain conversation tree, hierarchical task orchestration with dependency-chain highlighting, and major stability hardening for persistence and synchronization. Multiple issues are fixed, including errors from repeated sync clicks, white flashes in dark mode, and deleted goals coming back to life.
New Features
| Feature | Description |
|---|---|
| Personal assistant tim | Adds a personal assistant with a customizable avatar and nickname, DingTalk-style bubbles and IM-style scrolling, visual cards and decision-card references in replies; enable "Task Tracking" in settings to automatically follow BiBrain tasks assigned to you |
| Proactive duty | Renames "Scheduled Tasks" to "Duty" and changes the page to a left-list / right-details layout; AI can draft duty rules from one sentence, which take effect only after your confirmation, with follow-up questions supported; each duty can start a new conversation on every wake-up or continue the previous one; due reminders appear in a bottom-right desktop popup that opens the duty directly |
| Artifact Library | Adds an Artifact Library to centrally manage files generated by conversations and projects; the pinned search bar supports global search, and artifacts can jump back to their source conversation with the relevant sidebar location highlighted |
| Conversation visibility | Adds a visibility switch for private conversations in the chat header; visibility is recorded when archived and uniformly filtered on read, so private conversations are not shared with the team by default |
| Cloud experts connected to BiBrain | Renames the Experts section to ProPilot; cloud expert conversations connect to BiBrain as the sixth conversation label, with a new "Experts" group in the conversation tree; experts support MCP connectors, Skills and batch folder uploads |
| Hierarchical task orchestration | Tasks support moving, breaking down, creating containers and arranging dependency edges; the new-task dialog fills in defaults (Friday due date, estimate conversion and creator as default owner); selecting a task in the relationship graph highlights the full dependency chain, and the "Start" pill shows the goal name and opens its goal drawer |
| Search in model selector | Adds a search box to the model dropdown so names can be filtered directly when there are many models |
| Remember last-used model | Adds an application-level preference that remembers the last model and reasoning level, carrying them into newly created conversations |
| "Next sentence" suggestions in the input box | Suggests the next sentence in the input box after a task finishes; press Tab to accept it and continue without reorganizing your wording |
| File browser edit / preview toggle | Adds an edit / preview mode switch to the file browser, so viewing and editing files no longer requires switching back and forth |
| Large-file and long-session rendering optimization | Large Markdown files now render progressively; long-session switching uses windowed rendering and less disk I/O for a significant speedup |
Bug Fixes
| No. | Problem Description | Solution |
|---|---|---|
| 1 | Clicking "Sync Now" repeatedly reports "The staging area is empty" | Repeated sync clicks no longer falsely report an empty staging area |
| 2 | Deleted goals reappear later | Fixes deleted goals being resurrected by backfill; they no longer reappear after deletion |
| 3 | Two white flashes before entering BiBrain in dark mode | The startup flash and initial window background now follow the current theme instead of flashing white before turning dark |
| 4 | Data from the previous team remains visible after switching teams | Reloads the whole window when switching teams so the previous team's in-memory state is not retained |
| 5 | Large parts of the interface remain in Chinese after switching to Japanese / English | Fixes BiBrain localization not taking effect |
| 6 | The assignee dropdown lists yourself twice in the new-goal dialog | The assignee dropdown no longer duplicates the current user |
| 7 | Assignee and due date lock each other during task editing | Removes the mutual lock and preserves completed assignee information while editing |
| 8 | An overly long goal name pushes over the status pill in "My Tasks" | Long goal names no longer squeeze the status label |
| 9 | A 403 error occasionally appears when opening a goal page | A goal not yet synced to the server is no longer shown as an error in a normal window |
| 10 | Bulk deletion in shared spaces has no protection | Deletions are no longer staged automatically; operations over the threshold require a second confirmation and can be restored from the commit graph |
| 11 | Sync gets stuck during commit | Long path parameters are processed in batches so sync no longer stops midway |
| 12 | Manual "Sync Now" misses conversations with open tabs | Fixes sync skipping active conversations |
| 13 | Long sessions return 413 and pile up in sync | Fixes the concurrent race in session persistence and oversized upload bodies |
| 14 | Horizontal content in DOCX preview cannot scroll | Fixes scrolling in horizontally overflowing content |
| 15 | The directory tree becomes blank or duplicated after refresh | Fixes empty-refresh redraws, refresh stacking and clearing the tree on timeout |
| 16 | Some corrupted images are retried repeatedly | Images are validated by file header; corrupted images show an error directly instead of being retried |
| 17 | Images are stretched after visualization zoom | Uses proportional scaling instead of stretching charts to fill the area |
| 18 | A goal can be marked complete even when its tasks are unfinished | Completing a goal is now constrained by task progress; it cannot be selected until all tasks are complete |
v3.1.15
Release Date: October 10, 2026
Release Note
This release further refines the BiBrain collaboration experience: task lists support one-click batch dispatch, and shared spaces support importing an entire folder directly. File preview is greatly enhanced — very large files are loaded in chunks and Markdown / HTML rendering is no longer truncated; AI chart cards gain zoom and full-screen viewing; the resource marketplace adds SQA Agent and CodeReview Agent preset tags; and the default Auto model can be turned off. Multiple issues are also fixed, including cross-team data leakage, incorrect overdue detection for tasks, and blurry text selection in chart cards.
New Features
| Feature | Description |
|---|---|
| Batch dispatch from task list | Supports selecting multiple tasks at once for batch dispatch, and fixes the assignee filter being swallowed by concurrency |
| Import folder into shared space | The space tree supports importing an entire folder directly, with no need to drag files in one by one |
| Enhanced large file preview | Very large files support chunked loading preview; Markdown / HTML preview now reads the full content, so rich rendering is no longer broken by truncation |
| Zoom and full screen for AI chart cards | Visualization chart cards inside the conversation gain a zoom button and full-screen viewing, so every detail is visible |
| New preset tags in resource marketplace | Adds two preset tags, SQA Agent and CodeReview Agent, kept in sync with the marketplace filter vocabulary |
| Option to turn off the default Auto model | A new switch in settings can turn off the default Auto model and switch to manually specifying the model in use |
Bug Fixes
| No. | Problem Description | Solution |
|---|---|---|
| 1 | Completed tasks still show as overdue | Completed tasks now show only the due date itself and are no longer counted as overdue |
| 2 | "My Tasks" shows tasks from another team after switching teams | The read path adds a team binding check, so data no longer leaks across teams after switching |
| 3 | "My Tasks" list and badge miss parent tasks | Adds the leaf-task criterion, so parent tasks are no longer omitted and the list rows and badge share the same source |
| 4 | Selected text in chart cards is blurry and smeared | The chart component injects a text selection style, restoring clear text when selecting |
| 5 | Messages can still be sent when all models are disabled | Intercepts the send and prompts the user to enable a model first, avoiding messages with no response |
| 6 | No save prompt after editing fields in task details | Adds a save prompt, and the list row updates on the spot |
| 7 | Sessions occasionally freeze | Rejects corrupted thinking signatures, preventing sessions from freezing permanently |
| 8 | After clicking, the mirror cannot be released and the conversation cannot be interacted with | Fixes the failure to release the mirror by mouse click, restoring conversation interaction |
v3.1.14
Release Date: September 30, 2026
Release Note
This release introduces all-new in-conversation AI visualization: when AI generates charts, dashboards or web components, they render live right inside the conversation — what you see is what you get, with no need to copy code elsewhere and run it separately. The BiBrain workspace also receives a unified overhaul: a new source control view and file tree operations support switching repository branches and reverting or resetting commits; new cross-team "My Workbench" and "Member Distribution" views are added, "My Tasks" becomes a standalone page with multi-dimension grouping; task status transitions can be customized per team; session replay is upgraded to high-fidelity rendering and the conflict resolution experience is further improved. Multiple stability issues including sync hangs, draft loss and double counting in statistics are fixed.
New Features
| Feature | Description |
|---|---|
| In-conversation AI visualization | Upgrades AI output from "text" to a "visible deliverable" — charts, dashboards and web components render live inside the conversation, what you see is what you get, with no need to copy code out and run it separately; data results are presented on the spot, easy to grasp at a glance and interact with at any time |
| Source control view | Adds a source control view that centralizes changes, commits, sub-repositories and a commit graph with branch lines; the commit button supports "Commit Only / Commit and Push" |
| Git repository and branch management | The Git view can switch between the team repository and each sub-repository, and supports switching / creating branches, copying commit hashes, and reverting / resetting commits |
| File tree enhancements | The file tree supports creating, renaming, drag-to-move, deleting and a context menu, and the entire working copy can be opened in VS Code |
| Cross-team workbench | Adds a "My Workbench" entry that aggregates personal tasks across all teams, grouped by team and showing counts of incomplete, overdue, due within 3 days and blocked items, with jumps to goals, boards or relationship graphs |
| Member distribution view | Adds a "Member Distribution" tab in the workspace showing stacked task status and overdue counts by owner, expandable for details with jumps to the corresponding board or goal |
| "My Tasks" as a standalone page | The cross-goal personal task list becomes a standalone page supporting three dimensions — "Owned by me / Created by me / Delivered by me" — groupable and collapsible by overdue, to-do and completed |
| Configurable task state machine | Teams can customize task status transition rules, different goals can advance along their own processes, and boards, statistics and KPIs adapt automatically |
| Sidebar session history shows only your own by default | The session history tree shows only your own sessions by default; the title row can expand to view all team sessions with one click |
| "Start" anchor and dependency check in the task dependency graph | The dependency graph adds a "▶ Start" anchor pointing to all initial tasks; when a predecessor is incomplete the "Execute" button is automatically disabled and indicates exactly which predecessor is blocking |
| Navigation experience improvements | Activity bar icons support drag-to-reorder and show / hide via the context menu; the last visited page tab, selected goal and view are remembered automatically, so there is no need to search again after switching |
| High-fidelity session replay rendering | Archived session replay is upgraded to a full conversation form that restores the reasoning process, tool calls and usage statistics instead of showing only a summary |
| Change highlighting in file evolution history | When playing back file evolution history, content added or rewritten in each version is highlighted, making it intuitive to see what changed |
| Improved conflict resolution experience | Conflicting files support accepting "mine / theirs" directly by clicking the code area or a button, and automatically enter the merge guide after resolution |
| Manual override for model image capability | Whether a model supports image input can be specified manually instead of relying solely on automatic system detection |
| Reviewer can be assigned for task review | A reviewer can be assigned when a task enters review; the default remains the goal owner |
Bug Fixes
| No. | Problem Description | Solution |
|---|---|---|
| 1 | Unsent draft in the input box is lost after switching away and back | Fixes the draft save timing: drafts are now saved synchronously on unmount and persisted with debouncing after typing pauses, so they are no longer lost on page switches |
| 2 | Engine "unexpected stop" kills live processes and retries report no engine | Process liveness detection now uses a system-level exit probe, write timeouts first confirm whether the process really exited, and a missing slot is rebuilt automatically on retry |
| 3 | Shared-space sync hangs and the top bar shows "Syncing…" forever | Adds a timeout fallback to sync initialization, with independent timeouts and exponential backoff for each clone / fetch / push step, eliminating permanent hangs |
| 4 | Sync encounters a leftover rebase and pops a 0-file conflict dialog that never closes | Leftover rebases are detected and advanced automatically before syncing; the resolution dialog only appears for real conflicts, ending the infinite loop |
| 5 | Notification popups stack up with duplicate identical text | Notification deduplication now compares title + body, so identical notifications fired consecutively at session end no longer stack |
| 6 | Clicking a card in the relationship graph forcibly resets canvas zoom to 1x | Fixes the view zoom being mistakenly adjusted after the drawer opens, so opening the drawer by clicking a card no longer forces a zoom level change |
| 7 | Branch lines break and change color at merges in the Git view commit graph | Fixes the "pass-through" column calculation during lane merging, so branch lines are continuous end to end and keep the same color throughout |
| 8 | Goal and task counts are double counted | Statistics now count only leaf tasks; parent containers are no longer counted repeatedly, and boards, badges and member distribution are corrected accordingly |
| 9 | Sidebar task execution sessions show a snowflake ID instead of the task title | The sidebar now reads titles from the local task index, so task names display correctly even after a cold start |
| 10 | Scheduled task run history stays at the first run time | Run history now shows the last actual trigger time, and the current session row adds a "last triggered" hint |
| 11 | The whole main view flickers and re-pages when switching models in a project session | The engine restart flag is no longer misused for gating decisions, eliminating the full-view page flicker when switching models |
| 12 | Push rejected by the server with non-fast-forward | The remote is now merged unconditionally before pushing, avoiding rejection caused by pushing without pulling local changes first |
| 13 | Completed tasks still show as overdue | Completed tasks now show only the due date itself and no longer compute overdue against the current time |
| 14 | The conflict dialog does not appear when a rebase has real conflicts | Fixes the pre-check intercepting early when a rebase has unresolved conflicts, so the conflict resolution dialog is now triggered correctly |
Code Cleanup
| Description |
|---|
| Localizes fonts and removes runtime Google Fonts requests, so client startup no longer makes external network access |
v3.1.12
Release Date: September 20, 2026
Release Note
This release redesigns the model selector and adds an Auto mode; the BiBrain window is significantly more stable and recovers after re-login without having to be closed; a new AI Change Review lets you inspect file diffs in the sidebar after AI edits and choose to accept or roll them back; shared-space synchronization is improved and Marketplace publishing now supports preset tags; multiple stability and interaction issues are fixed.
New Features
| Feature | Description |
|---|---|
| Redesigned model selector | Adds an Auto mode that stays pinned at the top and can be selected just like a regular model |
| AI Change Review | After AI modifies files, a new "Review" tab appears in the chat sidebar collecting all file diffs pending confirmation; accept or roll back changes one by one or in bulk, so every change is confirmed by you |
| Preset tags for Marketplace publishing | When publishing a Skill or extension, the confirmation dialog lets you select preset category tags (sharing the same vocabulary list as the Marketplace filter bar); after publishing, items can be found by filtering on those tags in the Marketplace |
| Personal profile as a standalone page | The personal profile is separated from "Team Settings" and its entry moves to the avatar menu in the top bar, so it can be edited at any time without joining a team |
| Pending-push commit count shown in shared-space sync | The preview stage of the sync dialog now reports when the working tree is clean but N local commits have not been pushed yet, instead of showing the misleading "no uncommitted changes" |
| On-demand sub-repository sync in shared space | Sub-repositories are excluded from sync by default and are committed and pushed only after being manually selected; the dialog shows each sub-repository's path and current branch |
Bug Fixes
| No. | Problem Description | Solution |
|---|---|---|
| 1 | Spinner stops immediately after sending a message and the message vanishes | Fixes an engine concurrency race: the busy flag is now set before the message is written, eliminating probabilistic message loss caused by the engine being mistakenly restarted mid-write |
| 2 | Model resets to Auto after sending a message | Auto state is now stored per session, so remounting when navigating to the session view no longer overrides the selected model |
| 3 | Model button in the toolbar occasionally disappears | A disabled placeholder button is rendered while the model list is loading, keeping the toolbar layout stable |
| 4 | Input box is empty after jumping to chat from "Use" in the Marketplace | Fixes slash commands being cleared due to an inconsistent draft ownership key |
| 5 | Preview keeps spinning when clicking a task in BiBrain "My Tasks" | In cross-goal task list views, preview, status transitions and field editing all take the goal ID from the list item instead of relying on a global goalId that no longer exists |
| 6 | Enter from a Chinese IME on Mac accidentally triggers goal / task creation | Goal creation and task creation forms now distinguish IME confirmation Enter from a regular Enter, so they no longer submit prematurely |
| 7 | Shared-space sync pops up an empty-file conflict dialog because of .gitignore | .gitignore blocking errors are caught during staging, and stageable paths are filtered and retried automatically, so false conflicts are no longer reported |
| 8 | BiBrain window stays stuck on "Please log in first" or "Engine offline" after re-login | After a login state change, the window automatically restores the team list, shared-space connection and model list, without closing and reopening it |
| 9 | Conversations from multiple rounds of the same task bleed into each other | Each execution round uses an independent session and no longer shares context with the previous round |
| 10 | Operation toasts never appear inside the BiBrain window | Fixes the Toast component not being rendered, so feedback such as fork / archive failures now displays correctly |
| 11 | Own archived sessions show "read-only replay" and cannot be continued | Fixes a stale session-list snapshot that misjudged whether a session exists locally, so just-archived sessions are no longer treated as read-only |
| 12 | Creating or clicking a shared-space conversation navigates to the shared-space page | Conversations now open as session tabs within the current page instead of replacing the page being viewed |
| 13 | Clicking "Sync Changes" reports a "staging area is empty" error | The working tree is actually checked before syncing and uncommitted changes are staged automatically, so skipping the preview step no longer causes an error |
| 14 | Profile editor for new members is fully greyed out and cannot be typed into | Input fields are always editable; typing directly creates a "team-specific override", overridden fields are highlighted separately and can be restored with one click |
| 15 | Shared space still shows "Ready" after the team repository directory is deleted | When reading the team repository, the directory's existence is verified automatically; if missing, the status is corrected from "Ready" back to "Not initialized" and an initialization entry is provided |
| 16 | Uninitialized team repository reports "please check your network connection" | Uninitialized and network failure errors are now classified separately, accurately reporting "team repository not initialized" and offering an initialization entry in the working directory panel |
Code Cleanup
| Description |
|---|
| Removes the embedded "Conversation" tab from the shared space; space pages become read-only views and conversations open uniformly as standalone session tabs |
| Flushes pending unsaved edits in bulk before closing the window, preventing pending-review changes from being lost on a forced close |
| Enhances internal engine code quality governance and automated verification |
Previous Versions
v3.1.11
Release Date: September 11, 2026
Release Note
This version introduces the brand-new BiBrain team collaboration foundation: team collaboration is organized around a single flow of "Team → Team Goal → Task." AI steps in only to decompose goals, execute tasks, and build team dashboards; in every other step, you review first, make changes if needed, and then confirm the result. The team's task structure, statuses, documents, deliverables, and conversation archives are all stored in the team's own Git repository, which can be cloned and inspected at any time instead of being kept in a black-box database. BiBrain is provided as a standalone window, accessible from the top bar of the main window, and can be used after enabling "Gateway Connection" in the Lab. This release also changes login to direct gateway-to-OA verification and adds application ratings, 2D office meeting status, file history evolution review, overwrite-and-refresh external session import, and more, while fixing multiple stability issues.
New Features
| Feature | Description |
|---|---|
| BiBrain Standalone Collaboration Window | Open the standalone BiBrain window from the main window's top bar; log in with an OA account and enable "Gateway Connection" in the Lab to load teams. All team assets are stored in the team's dedicated Git repository and can be cloned for inspection at any time |
| Create a Team (Two Methods) | Supports forming a team through an AI conversation, selecting people from the organization roster and viewing each person's current workload, or filling out a form manually; after creation, a Git repository is automatically created for the team and AI work guidelines are distributed |
| Team Goals and Milestones | Supports goal trees, milestones, KPIs (four metrics plus monthly changes), status transitions, pinning, and archiving; goals can manage child tasks hierarchically |
| AI Team Goal Decomposition | The decomposition page provides an AI conversation on the left and a real-time task graph on the right; supports adjusting "dependency density" to control granularity, directly editing task fields in the form for the AI to read in the next round, and viewing a version evolution timeline. Decomposition results are saved as local drafts by default and applied as official tasks with one click after confirmation |
| Task Workflow Board | Once tasks are created, they enter the task workflow board, where they can be grouped by status, assignee, priority, or top-level task. Dragging a card changes its status; dragging it to "Blocked" requires a reason. List view can switch among "Assigned to Me," "Following," and "Created by Me" |
| AI Single-task Execution | Click "Execute" on a task card to chat with AI on the left while viewing generated files on the right. AI automatically searches existing materials in the team's shared space, so context does not need to be pasted again. When you exit, the conversation is automatically archived to the team repository for teammates to review |
| Task Execution Session Archiving | Every decomposition and execution conversation is automatically archived in the team repository and can be reviewed from the goal details, task details side panel, or shared-space session page, allowing teammates to see how each task was completed |
| Team Shared Space | Provides tabbed views for file trees, documents, code repositories, goals and tasks, sessions, and deliverables; supports change review, staging / committing / pushing, history, recycle bin and recovery, full-text search, and archived conversation browsing. AI never commits automatically; you always confirm what enters the team repository |
| Invoke AI from the Shared Space | Open a floating AI conversation directly beside a file or session record in the shared space to ask AI to work, such as resolving a Git conflict; also supports "Continue Session from Here" to continue with the previous context |
| Team Home Dashboard | Displays the number of goals, completed goals, risks, member workload, recent activity, and Top 5 blocked items in real time, with WebSocket-based real-time team activity updates |
| AI Team Retrospective Dashboard | Describe what you want to see to AI, which generates a self-contained HTML dashboard stored in the team repository for the entire team to open. The workspace supports draggable splitters and selected states for template cards |
| Invitation and Join Approval | Supports inviting people via invitation codes / links; join requests go through an approval process. Member roles and permissions, member profile summaries, and workload can all be maintained. Users already in a team can join a second team |
| External Code Repository Viewing | Supports attaching external GitLab repositories and viewing merge request (MR) lists in read-only mode |
| Login via Direct Gateway-to-OA Connection | Employee ID and password verification is completed by the gateway directly with OA instead of being relayed through the legacy server, making login more stable; departments can be synchronized from OA |
| Application Rating | Supports rating the application from 1 to 5 stars; ratings below 5 stars require at least a 20-character reason. Ratings can be submitted from "Help → Rate the Application" in the top bar or in response to server-side requests, reusing the feedback ticket workflow |
| 2D Office Meeting Status | Adds a "In a Meeting / Available" status pill beside the AIKO mark in the sidebar, with one-click switching. While executing tasks, the status is automatically reported according to whether the task is personal or team-owned and displayed in the 2D office workspace |
| File History Evolution Review | Deliverables and project files support viewing an evolution timeline of past modifications and comparing changed lines version by version, making it easy to understand how a file changed step by step |
| Overwrite-and-refresh External Session Import | External session import now re-parses and refreshes on every import instead of skipping existing sessions. Updated sessions are marked with an "Updated" badge, and selecting all after a search imports only the search results |
Known Limitations
This release is the trial version of the BiBrain foundation. The following capabilities are not yet available:
| Item | Description |
|---|---|
| Data Source Access | Cannot yet pull workorders from Jira / Redmine / DingTalk / Email to create goals; goal descriptions must be entered manually |
| Profile-based Assignee Recommendations | Member profiles can be filled in and viewed, but assignees must still be selected manually; automatic recommendations are not yet available |
| Task Relationship Graph | The relationship graph only displays parent-child hierarchy and does not yet show dependency / blocking arrows between tasks |
| Status Gates | There are no deliverable or review gates yet; any task can be moved directly to "Completed" |
| Some Team Settings Cards | "Shared Profiles," "Prompt Skills," and "Scenario Extensions" are not yet available |
| Old and New Team Interfaces Coexist | The original "Team" interface in the main window remains alongside the BiBrain window; during the trial, use the BiBrain window |
Bug Fixes
| No. | Problem Description | Solution |
|---|---|---|
| 1 | Creating a new push message returns 500 | Fixes abnormal decoding of a boolean message field that caused errors in the admin console's create and list paths and the client's polling path |
| 2 | Execution record timestamps are 8 hours early | Formats session times according to the local time zone instead of directly truncating UTC strings as local times |
| 3 | Decomposition / execution sessions cannot be found under goals or tasks | Correctly archives and associates drafts and exited sessions with the goal code; conversations can be reviewed from goal details, the task details side panel, and other supported locations |
| 4 | AI does not follow the work guidelines in the first decomposition message | Restores implicit injection of goal context and work guidelines into the first decomposition message so AI does not fall back to a generic assistant |
| 5 | Previous goal content remains after switching goals | Resets decomposition state when switching goals, preventing tasks from being created under the wrong goal or duplicate tasks from being dispatched |
| 6 | Existing teams cannot receive task execution guidelines | Detects existing teams individually and adds the work guidelines, so the execution page no longer continuously reports that guidelines are missing |
| 7 | The "My Work" badge remains at 0 after decomposition | Subscribes the badge to goal events so it refreshes in real time after decomposition generates tasks |
| 8 | Pushing after uploading a code repository fails | Fixes multiple issues in shared-space code repository upload, commit, and push flows |
| 9 | Shared-space file tree and synchronization are abnormal | Changes the file tree to a local view and fixes Git synchronization and conflict handling; Agent output that has not been committed can also be viewed in the interface |
| 10 | Data source menu does not refresh after cold start | Reloads the data source list and availability each time the "+" menu opens, removing the need to visit the settings page first |
| 11 | Expert recent task list contains many empty entries that cannot be opened | Shows only real execution snapshots, aggregates them by session, and filters out inspection / weekly-report and other non-execution records and other users' private sessions |
| 12 | Gateway migration verification always fails on Windows | Fixes silent failure caused by python3 resolving to the system placeholder executable |
Code Cleanup
| Description |
|---|
| Consolidates data source injection into a unified specification and removes injected body previews (chips are retained), preventing time-window drift between previews and actual injected content |
| Standardizes the shared-space template project; creating a team distributes the directory structure and work guidelines together with the skeleton |
| Enhances internal engine code quality governance and automated verification |
Previous Versions
v3.1.10
Release Date: September 4, 2026
Release Note
This version unifies new gateway login across all users, adds external session import, @ reference for historical sessions, proactive IM message push, a first-time user guidance page, knowledge base chips mounted above the input box, project-level memory isolation, and other features. It greatly optimizes team collaboration and workorder dispatch experiences, and fixes multiple stability issues including streaming interruption errors with no diagnostic detail, Chinese Markdown files failing to preview, and deliverable path display abnormalities.
New Features
| Feature | Description |
|---|---|
| Unified Gateway Login | All users now log in through the new unified gateway; the old-version toggle switch has been hidden |
| External Session Import | Supports importing external session files from Claude Code, Codex, etc. The sidebar shows a source badge; imported sessions support continued chat and deletion |
| @ Reference Session | The @ popup in the input box now has a dual tab of "File / Session", allowing one-click reference of historical session content |
| New User Guidance Page | A guided tour for scheduled tasks and TEAM features automatically pops up on first launch; can be rewatched later from Help → User Guidance, supporting Chinese, English, and Japanese |
| Proactive IM Message Push | Feishu / DingTalk support Aiko proactive push: task completion summaries, execution failure alerts, and progress updates for long-running tasks. Push targets and granularity are configurable |
| Knowledge Base Session Mounting | Knowledge base chips can be mounted directly above the chat input box and removed at any time; knowledge base selection is card-based, unified with MCP connectors under a "+" menu |
| Project Memory Isolation | Each project's memory is stored independently without interference, while global memory merge-and-recall is retained. Switching projects no longer causes memory cross-contamination |
| File Reference Button | After selecting file content, click "Reference" before entering a session; references automatically expire when switching sessions, preventing content from a previous file from being carried into another session |
| Sub-Agent Info Enhancement | Sub-Agent bubbles now display the task title; the read-only session activity group is expanded by default when opened, making it easy to see what each sub-Agent is doing at a glance |
| Workorder Dispatch Enhancement | Dispatch plans support modification by the builder / dispatcher; workorders gain "Complete" acceptance, rejection of assignments, and ticket reset; PM gains view of assigned workorders and custom filtering |
| Feature Collaboration Flow DAG | The collaboration flowchart is now laid out as a topological DAG, clearly showing fork / merge structures with more accurate fan-in wait prompts |
| Multilingual Task Creation | Workorder decomposition and status transition content are generated in sync with the interface language (Chinese / English / Japanese); the Japanese interface no longer produces Chinese workorders |
| Team Dynamics Enhancement | The comment area returns to pure human conversation; task lifecycle announcements move into the dynamics stream with phase badges |
Bug Fixes
| No. | Problem Description | Solution |
|---|---|---|
| 1 | Streaming answer interruption only shows "Unknown stream error" | Errors are classified by source (connection dropped / upstream error, etc.); the bubble can be expanded to view raw diagnostic information, no longer misleading troubleshooting direction |
| 2 | Pure Chinese Markdown cannot be previewed | Fixes the issue where Chinese documents were incorrectly identified as binary files and could not be previewed |
| 3 | Deliverable display abnormality | Fixes deliverable paths showing temporary sandbox prefixes and deleted deliverables still being displayed |
| 4 | Completed task card shows "Unknown Status" | Fixes the board card status mapping loss; status display is restored to correct |
| 5 | Supervised mode question automatically rejected by timeout | In supervised mode, questions now wait indefinitely for a human reply, no longer auto-answered by timeout |
| 6 | Login server address incorrectly judged as duplicate | Login server deduplication now only compares against currently visible addresses, no longer falsely reports duplicates due to hidden addresses |
| 7 | Workorders fail to load after modifying filter conditions | Filter changes now automatically reload workorders using the new conditions; collaboration sync no longer forces team selection |
| 8 | Tasks with incomplete predecessors show execute button | Tasks waiting for predecessor completion no longer show execute / view buttons |
| 9 | Newly created project session in sidebar not associated with project | New project sessions now automatically inherit the project directory, properly isolating project memory and working directory |
| 10 | Pagination bar of synced data pushed off screen | Lists now use internal scrolling with fixed headers; long lists no longer overflow the viewport |
| 11 | Normal files with "tmp" in filename hidden | Fixes the intermediate artifact filter incorrectly hiding normal files |
Code Cleanup
| Description |
|---|
| Removes the TEAM v2 temporary entry, unified to be hosted by the BiBrain client |
| Gateway database supports unified migration across SQLite / MySQL / PostgreSQL |
| Internal code quality governance and automated verification enhanced |
Previous Versions
v3.1.9
Release Date: August 28, 2026
Release Note
This version adds conversation message editing fork, R&D management module, workorder decomposition prompt templating, Team Center redesign (members/prompts/shared profiles), task list dispatch/execution dual view, expert memory browsing, and other features, greatly optimizing the Team collaboration experience and workorder dispatch flow, and fixes multiple stability issues such as cross-session draft carryover and background todo update loss.
New Features
| Feature | Description |
|---|---|
| Conversation message editing fork | Editing a historical message automatically forks into a new session; the child session inherits the previous conversation content, working directory, and permission settings, while the original session remains unchanged |
| R&D management module | Adds R&D institute management functionality, including four sub-modules: Plan, Team Tasks, My Tasks, and Team Management, supporting team role and permission configuration |
| Workorder decomposition prompt templating | Workorder decomposition prompts support custom templates and placeholders; sorting is moved to header click; the data source and filter bar switch to unified-style dropdown selectors |
| Team Settings upgraded to Team Center | The Team Settings page is upgraded to Team Center, with three tabs: Member Management, Prompt Configuration, and Shared Profiles; supports overriding personal profiles at the team level, displayed per-person as a grid to facilitate comparing members' expertise during dispatch |
| Task list dispatch/execution dual view | The left task list adds "Dispatch" and "Execution" view switching: Dispatch shows conversations initiated on this machine, Execution shows workorders assigned to you; both views are grouped by team and support collapsing |
| Workorder data source unification | Jira and Redmine workorders adopt a unified data format; subtasks serve as the dispatch entry, parent tasks provide context; adding new data sources requires no client update |
| Expert memory browsing | When building team tasks, you can browse the expert memory library, supporting associating local files (e.g., daily reports, inspection sheets) as task context |
| Server-side message push | Administrators can push notification announcements to clients via the gateway; clients display them as banners or popups |
| Workorder owner filtering | Synced workorders support searching by ticket number, name, and email; the "Show Unbuilt Only" filter is changed to "Show Only Mine" |
| Department dynamics optimization | Department dynamics no longer follow group switching; grouped by team with collapsible display; workorder links can jump to the corresponding task and locate comments |
| Model capability update | Adds image understanding and context window capabilities for glm-5-3-flash and qwen3-8-flash; removes the offline ox-alpha |
| Plugin project template | The plugin template is updated to the xcu project plugin; the sidebar shows only enabled plugins with valid navigation entries; the project creation popup supports multi-language |
| Custom tools | Supports declaring custom tools via configuration file and associating script execution; adds a tool management settings page |
Bug Fixes
| No. | Problem Description | Solution |
|---|---|---|
| 1 | Attachments carry over to other sessions when switching | Unsent attachments (pasted images, etc.) are now isolated per session; they are retained when switching back, no longer carrying over |
| 2 | Switching model in one window affects other sessions | Model switching now applies only to the current session, no longer affecting running sessions in other windows |
| 3 | Background session todo progress not updating | During switching to other sessions, todo list updates of background sessions are no longer lost; the latest status is displayed when switching back |
| 4 | Synced workorders pushed repeatedly | Fixes the duplicate push issue caused by identifier changes after workorder updates |
| 5 | "Show Unassigned" filter ineffective | Fixes the issue where built workorders could not be filtered out; filter results are consistent with the status column display |
| 6 | File link display format abnormal | Fixes the layout issue where file links in messages were squeezed into a single line; corrects non-responsive clicks on links in non-project mode |
| 7 | Plugin project config occasionally ineffective | The plugin project config path passing is integrated into the startup flow, eliminating occasional failures caused by async timing |
| 8 | Profile edit popup mask too dark, input box color mismatch | The popup mask and input box colors are unified to a globally consistent style |
| 9 | Task flow status judgment abnormal | Fixes the issue of incorrect status retrieval when the instance changes after flow; only the flow operator can decide the next status |
| 10 | Deliverable upload identification error | Fixes the incorrect identity and stage identification when uploading deliverables |
| 11 | Some internal addresses point to old server | Fully updates the repository address configuration to ensure pointing to the correct servers |
Code Cleanup
| Description |
|---|
| The tool execution bubble is changed to a compact two-line layout, details collapsed by default, improving long-session browsing efficiency |
| Fixes the AI assistant text duplication display issue |
| Removes the old R&D management functionality |
| Adds a device fingerprint drift diagnostic tool (for License activation issue troubleshooting, internal support only) |
| Engine internal code quality governance and automated verification enhancement |
v3.1.8
Release Date: August 21, 2026
Release Note
This version adds features such as extension-provided custom new-project forms and fields, one-click sync task data operation (merged pull and upload), task flow and My Tasks UI redesign, conversation trajectory visualization, and gateway group management. It optimizes error message guidance, cache hit rate, and sub-Agent interruption recovery, and fixes multiple stability issues including visual capability failure after quick model switch, duplicate artifact paths, License deactivation due to network card changes, and reactivation after upgrade.
New Features
| Feature | Description |
|---|---|
| Extension custom new-project | Extensions can provide custom forms and fields for new projects, which can be centrally viewed and managed within the project; when an extension is disabled, the fields are retained as read-only without affecting existing data |
| One-click sync task data | "Pull to Local" and "Upload to Cloud" are merged into a single "Sync Task Data" button, which auto-syncs once on page entry; sync failure aborts immediately to avoid overwriting the cloud with stale data |
| Task flow and My Tasks redesign | Adds task flow and manual execution interfaces, supporting task confirmation and status transitions; the My Tasks interface is fully redesigned |
| Conversation trajectory visualization | Adds a conversation trajectory timeline panel to review every step of the conversation and tool execution; the trajectory entry is presented as a "Trajectory" text label, unified in style with "Replay" |
| Pre-execution preparation for subtasks | Before subtask execution, you can select a working directory, associate files, and add notes; the working directory is auto-remembered and backfilled per team |
| Sub-Agent interruption recovery | A sub-Agent that fails midway can automatically recover and resume, with failure messages including trajectory-location guidance |
| Cache hit optimization | Optimizes cache hit-rate accounting and display, fixes the issue where streaming conversations for some models did not hit the cache, reducing repeated context billing |
| Gateway group management | Adds a group administrator role for gateways, supporting group member management and quota allocation |
| Group context isolation | The @ list when starting a task is narrowed to the current group; task history and conversations are displayed grouped by group |
| Project Git repository initialization guidance | When no Git repository is detected for a project, one-click initialization guidance is provided; after initialization, the file change list is automatically displayed |
| Right-click menu for starting tasks | Starting tasks from the sidebar supports a right-click menu: Rename, Replay, Open in Parallel, Export Logs, Share, Delete, etc. |
Bug Fixes
| No. | Problem Description | Solution |
|---|---|---|
| 1 | Model-layer instability error bubbles misread as backend failure | Infrastructure/unknown-cause errors are uniformly changed to network-environment guidance text (429 "Network access restricted"/others "Check network environment"); auth/badRequest retains the original tone; status code and raw error log are preserved |
| 2 | Visual capability does not take effect after quick model switch | When quickly switching to a visual model, the AIKO_API_IMAGE_INPUT env retained its old value, causing images not to be read; setModel now synchronously rewrites the process env when receiving an imageInput hint |
| 3 | Compression summary request writes cache in vain | Adds skipConversationCacheBreakpoint; summary requests no longer write a new conversation breakpoint at the end of one-shot instructions, saving hundreds of thousands of tokens of useless cache writes |
| 4 | Duplicate artifact path causes open failure | joinArtifactPath produced .../report.md/report.md; adds a guard to collapse adjacent duplicate file segments, with a dev-time warning when base is not a directory |
| 5 | DingTalk approval issue | Fixes DingTalk approval flow anomalies |
| 6 | Network card change causes License deactivation | v2 fingerprint removes unstable MAC addresses (retains only machine_guid + board_serial + disk_serial); either v1/v2 candidate passing is accepted; existing licenses do not deactivate after upgrade |
| 7 | Reactivation required after upgrade | Retrospectively probes all historical storage locations of license.dat; the first one that passes validation is used; when a non-official location is hit, it is migrated once to the official location |
| 8 | AIKO balance display error | Fixes the incorrect balance display issue; quota query interface upgraded |
| 9 | API Key application and renewal failure | Fixes the inability to apply for keys and renew |
| 10 | External data source connectors do not use proxy | External connector requests such as Redmine / Jira uniformly follow the HTTP proxy in general settings, supporting proxy access to intranet self-hosted instances |
Code Cleanup
| Description |
|---|
| Removes the custom schema cleanup logic on plugin disable/uninstall; schema snapshots are persisted locally to the project and no longer cleared on plugin start/stop |
| Refactors the custom field file/folder selection control into a unified path-selection component with icons and placeholders |
| Trajectory panel colors are taken directly from the underlying color-scale tokens, bypassing conversation-flow color interference and optimizing render performance |
| Custom tools support script deletion, and the enable/disable strategy for custom tools is optimized (can be included in scheme management, quantity counts toward custom tools) |
v3.1.7
Release Date: August 14, 2026
Release Note
This version adds full-link session sharing (send/receive/deep-link/real-time notification), tool and scheme management, end-to-end reasoning depth setting (6-level instant switching), sub-Agent session read-only viewer, file selection as session context, plugin system enhancements (scheduled tasks/form popups/lifecycle notifications), Team detail page and collaboration flowchart aligned with prototype, personnel collaboration relationship graph, and other features; it greatly optimizes cross-gateway usage normalization and cache hit rate display, and fixes multiple stability issues such as post-stop freezing, MCP config BOM, and compression failure routing.
New Features Added
| Feature | Description |
|---|---|
| Full-link Session Sharing | Supports the full link of session sharing: send/receive/deep-link/real-time notification; the shared space integrates the "Sessions Shared with Me" and "Sessions I Shared" panels |
| Tool and Scheme Management | Adds tool enable/disable and scheme management capabilities; the engine filters tool registration by whitelist/blacklist, with visual management on the frontend settings page |
| End-to-end Reasoning Depth Setting | Connects the reasoning effort end-to-end link, expanded to full 6 levels (off/low/medium/high/xhigh/max), supports per-session instant switching, OpenAI protocol transparently passes reasoning_effort |
| Sub-Agent Session Viewer | Sub-Agent execution bubble adds a glasses icon; clicking opens a read-only session tab in the right temporary space, displaying the full message stream and refreshing dynamically with execution |
| File Selection as Context | Text selected in the file preview area is automatically written into the session context; code is referenced by line number, documents are located by section; supports multi-file/multi-segment display and auto-clears after sending |
| Plugin System Enhancement | Adds scheduled task page and API, custom form popup (8 field types + validation), theme adaptation (hostTheme), lifecycle notifications, append-message immediate execution, engine streaming incremental event display, host entry for creating tasks/projects/sessions, plugin Skill creator |
| Team Detail Page and Collaboration Flowchart | Feature detail page aligned with BiBrain prototype, Issue card expands status instance chain/rating/comment area; collaboration flowchart nodes supplement icons and assignees, clicking a node links and expands the Issue |
| Personnel Collaboration Relationship Graph | Adds a personnel collaboration relationship graph view and task reassignment |
| Team Member Management Enhancement | Adds add-member/settings/pagination components, uses authoritative data to refresh member lists, preserves existing data when updating roles |
| Cache Hit Rate Display | Adds a cache hit indicator to the right of the interactive badge; normalizes usage fields across gateways (Anthropic/OpenAI/DeepSeek/Gemini), fixes cache read/write accounting and cost double-counting bugs |
| Floating Conversation History Sidebar | Floating window left side adds a history mini-panel, supports collapse persistence, complete history list, click to switch session in place; "Start Task" supports a back button |
| Model Image Capability Alert | Pops up a warning when switching to a non-visual model while history contains images; the kernel automatically downgrades images to text placeholders |
| Login Window Gateway Shortcut | Login window adds a new gateway shortcut entry |
| Session History Pagination and Sorting | Project/chat/launched task lists default to 12 items + button pagination to expand; sessions sorted by recent activity; title row is fully collapsible |
| Engine Default Tool Whitelist | workflow/browser/worktree three heavy capabilities require explicit host listing to load; new builtin tools are not in the default set by default |
Bug Fixes
| No. | Problem Description | Solution |
|---|---|---|
| 1 | Deadlock stuck on "Thinking" after stop/switch model | remove_engine first removes the slot then wraps stop packet with 8s timeout, interruptEngine moves UI reset down to finally, no longer depending on whether approval succeeds |
| 2 | MCP config file BOM causes parsing failure | Strips UTF-8 BOM, empty file tolerantly returns empty list, parsing failure supplements path and content preview for localization |
| 3 | Compression failure 403 treated as PTL retry burning quota | Routes by error class: quota/auth throws immediately without retry, PTL directly mechanically degrades, timeout goes through threshold, the real cause is no longer masked by the threshold |
| 4 | cmd.exe branch quote corruption | cmd.exe branch switched to windowsVerbatimArguments to pass arguments as-is, adds regression tests |
| 5 | Pasted text cannot be undone | Fixes the issue where PromptInput cannot perform undo after pasting text |
| 6 | Received session cannot be continued | import_conversation_bundle writes back rollout to convs/ directory so --resume can load context |
| 7 | Session sharing notification jump error | conversation_shared notification changed to jump to TEAM shared space - Shared tab; fixes history/reconnect notifications landing on detail popup due to missing target |
| 8 | Tool management not actually taking effect | Fixes tool policy filtering logic and frontend state synchronization |
| 9 | Each popup mask darkens and dark mode is messed up | Login/connector/gateway/settings popup masks uniformly remapped, dark mode panel/input corrected |
| 10 | Floating box dragged to top is covered by toolbar and cannot be dragged back | Raises the y lower bound of drag/resize/collapsedTop to toolbar height, title bar events are not intercepted |
| 11 | Floating dialog cannot be dragged out of main area | Changed to portal to team-view-root, can stretch left/right boundaries when creating task |
| 12 | History ticks hidden prematurely | Changed to use content-box right edge by subtracting paddingRight, ticks retract only when close to bubble/text |
| 13 | Overly long tool description bursts Agent bubble | getToolDescription truncates command/text/pattern/query to 120 characters |
| 14 | Unsent content display anomaly when switching sessions | Fixes display issue of unsent sessions in session edit box |
| 15 | File panel width reset after minimizing window | Fixes ChatView/ProjectSessionView panel width loss after window restore |
| 16 | File render area not updated in real time after editing file | Fixes useProjectFiles file change monitoring |
| 17 | Quick issue location icon covers main session | Fixes MessageNavigator icon layer and position |
| 18 | TUI cannot exit | Fixes aiko-app TUI exit logic |
| 19 | Context ring turns red with usage, causing discomfort | Changed to a fixed single-color green, removing the three-tier color logic |
| 20 | API Key project integration token treated as personal key | Excludes project integration bindings when querying ownership table, NULL apply_type historical bindings still issue keys normally |
| 21 | Traditional login quota query interface v1 outdated | Upgraded to v2 interface |
| 22 | 400/401/402 errors not retried | Adds 400/401/402 retry strategy |
| 23 | Pending message queue not sent immediately | Engine skips queue and sends immediately when idle |
| 24 | Task not removed when status is deleted | Adds deleted status handling logic |
| 25 | Target loop target→goal rename residue | Completes goal.noGoal/noGoalHint trilingual keys, fixes traj_features else branch NameError |
| 26 | Connector popup mask too dark | Raises selector specificity to align with access channel popup, test connection reuses running engine environment |
| 27 | Dashboard TS build error | Fixes WeeklyReport/FinancePage type inference issues |
| 28 | Startup error UI lifecycle notification reference error | Fixes view.id field referenced by ExtensionViewContainer |
Code Cleanup
| Description |
|---|
| Target loop feature target root globally renamed to goal (types/events/tools/rollout/slash commands/CSS/components/i18n) |
| Removes deprecated tools, improves authority of tool management configuration |
| Gateway compatible with OneAPI series |
v3.1.6
Release Date: August 7, 2026
Release Note
This version adds Browser-Use browser automation toolchain, full-screen floating dialog (draggable/resizable/pin), Team view refactoring (My Tasks/Shared Space/BiBrain Cloud Drive), external connector architecture (Redmine/Jira), engine hook system, Skill content encryption and other features, greatly optimizes streaming rendering performance and tool bubble experience, and fixes multiple stability and interaction issues.
New Features Added
| Feature | Description |
|---|---|
| Browser-Use Browser Automation | Complete browser toolchain based on chromiumoxide (launch/navigate/click/type/screenshot/extract_content/close), full frontend-backend pipeline connected; added Browser-Use settings module (capability toggle/dedicated proxy/connectivity pre-check), short-timeout reqwest probe for target site before navigation; system prompt adjusted to prioritize browser access when browser_* tools are available, fallback to WebSearch on failure |
| Full-screen Floating Dialog | Added FloatingComposer three-segment floating input box (title bar/message area/input area), supports 8-direction resize and drag, pin lock, collapse capsule, geometry persistence to localStorage; file browser and preview panel mounted in fullscreen, current file auto-injected as attachment to session |
| Team View Refactoring | Navigation changed to "Start Task/My Tasks/Team Activity/Shared Space/Department Settings"; My Tasks adds Feature board (stats bar + cards + Issue details); Shared Space rewritten as complete cloud drive view (breadcrumb/folder tree/upload/move/link/share/preview/download), integrated with BiBrain cloud drive capability |
| AIKO Pool Changed to Right Drawer | AIKO pool changed from bottom popup panel to right full-height drawer (slide in/out + semi-transparent overlay + Esc close), list page has persistent floating dialog supporting "Start Task/Quick Chat" dual tabs |
| Data Source Adds Redmine and Jira Support | Connector architecture refactored to support external connectors, Redmine and Jira adapters changed to JavaScript external implementation, decoupled from Rust built-in |
| Skill Content Encryption | Self-developed Skill content AES-256-GCM encryption on disk, only model can read at runtime, UI/Read display placeholder; added encryption button and encrypted state persistence |
| VSCode Extension Built-in Installation | VSCode plugin built into client installer, settings page adds one-click install/manage entry; plugin itself reuses client kernel, maintains cross-platform |
| History Session and Project Multi-select Delete | History sessions and projects support Ctrl multi-select + right-click batch delete |
| File Panel Browse Any Directory | Project file panel supports browsing files in any directory, no longer limited to project directory |
| Model Selector Reasoning Depth Switch | Model selector footer adds low/medium/high three-tier reasoning depth segment selector, changes saved instantly and linked with settings page |
| Default Language Follows System | UI default language follows OS language setting, activation UI and backend prompts fully multi-language |
| Error Bubble Connectivity Test | Error bubble "View Log" changed to point to agent.log, added "Test Connection" button to test current model connectivity in-place |
Bug Fixes
| No. | Problem Description | Solution |
|---|---|---|
| 1 | Malformed file path causes chat area white screen | Path containing unescaped % or truncated multi-byte escape causes decodeURIComponent to throw URIError reaching ErrorBoundary; added safeDecodePath fallback, uses literal path when decode fails |
| 2 | Floating box drag/resize lag | High-refresh-rate devices pointermove per-event setState accumulation, streaming content crushes main thread; setRect via requestAnimationFrame frame merging, component memoization |
| 3 | Background session cold cache persistence overwrite loses session header | Non-active session cold cache builds headless tail from empty array and overwrites disk; changed to merge by id with disk, never overwrites complete history with tail |
| 4 | Session restore loses context and assistant text duplication | Mid-way cancel and reopen engine with empty session-id causes model to not see history; marked resumeOnStart after first successful start, subsequent reopen uses --resume; also fixed text deduplication |
| 5 | Chrome startup exit code 21 (PROFILE_IN_USE) | Orphan process locks shared profile directory; changed to use independent unique user data directory each time, added first-start robustness flag |
| 6 | click/type index misalignment | click/type scripts use different selectors and dom_extractor numbering inconsistent; unified INTERACTIVE_SELECTORS constant shared across three places |
| 7 | Skill directory repeated injection causes context bloat | Skill directory originally published as independent user message each round, repeated injection accumulation after resume/compression/hot-reload; migrated to system skills tier dynamic segment, not in transcript |
| 8 | Foreground command orphan grandchild process holding pipe causes turn hang | Orphan grandchild process inherits stdio pipe handle, close event never triggers; finalize extracted, with exit+1.5s grace period fallback, 8s timeout forced cleanup |
| 9 | Restoring session with dangling tool_use causes 400 | restoreFromDisk previously only fixed tool pairs when compression summary exists; changed to unconditionally execute sanitizeToolPairs |
| 10 | Task detail return always goes to homepage | Clicking detail from "My Tasks" and then ← returns to wrong homepage; adopted "source snapshot" model, restores source tab when closing detail |
| 11 | File preview placeholder icon not displayed | Global border-box squeezes SVG drawing area to 0 by padding; changed to content-box, empty state split into placeholder container + inner button |
| 12 | Session/project right-click menu truncated by window bottom | Right-click menu flips at viewport boundary based on real dimensions |
| 13 | Split session jumps back to homepage | After clicking "Split Session", new session jumps back to homepage new session; changed to stay in file panel tab for interaction |
| 14 | Data directory modification not fully effective | Fixed data storage path modification effectiveness logic, optimized change flow |
| 15 | Deleting history session in project causes project order reordering | Fixed project display order being accidentally reordered after deleting session |
| 16 | Model dropdown list options right-aligned | Top toolbar select's text-align:right inherited by Chromium to option list; explicitly reset left-align for options |
| 17 | Connector probe idling failure log when not logged in | Probe lacks auth prerequisite, blindly runs detection when not logged in; only probes when auth_state is Ok |
Code Cleanup
| Description |
|---|
| "Plugin" terminology unified to "Extension" across project |
| aiko_eval directory renamed to aiko_eval_repo |
| Removed redundant TeamMembersView / TeamArtifactsView / TeamSimpleTaskList components |
v3.1.5
Release Date: July 30, 2026
Release Note
This version adds Git Worktree isolated workspaces, Supervised Mode (three-tier permission control), plugin extension system, message navigation ticks, session log export and other features, comprehensively upgrades Team group architecture and brand visuals (Ultra-Soft Blue default theme, SDW registered trademark), optimizes Target Mode interaction and session list performance, and fixes multiple experience issues.
New Features Added
| Feature | Description |
|---|---|
| Git Worktree Isolated Workspace | Supports creating Git Worktree isolated branch workspaces, AI automatically orchestrates worktree creation → sub-session → split-screen; provides full management UI for create/merge/rebase/remove, file tree and commit panel automatically sense worktree context |
| Supervised Mode (Three-tier Permission) | Permission selector upgraded from two-state to three-tier menu (Confirm/Supervised/Full-auto); in Supervised Mode, files and commands auto-pass, MCP tools require manual confirmation, AI questions remain unobstructed; set as new default, buttons colored by mode |
| Plugin Extension System | Rust backend host manages plugin installation, start/stop and event forwarding; frontend plugin view and management panel; supports both zip package and directory import methods; unified popup management and permission verification |
| Message Navigation Ticks | Compact tick cluster added to the right side of chat area (latest 8 user messages), hover pops up summary panel, click jumps to corresponding message, current position highlighted |
| Export Session Log | Session right-click menu adds "Export Log", packages engine diagnostic files and session records as zip for troubleshooting |
| Target Mode UI Enhancement | Target status changed to horizontal compact overlay, hover instantly shows progress hint, click expands detail card; existing input preserved when creating target |
| Ultra-Soft Blue Default Theme | Added "Ultra-Soft Blue" theme (SWD Blue #3B8FFF) and set as default, includes full set of primary/gradient/grid/glow variables |
| Brand Visual Update | Top bar SDW text replaced with registered trademark SVG (color follows theme), sidebar aiko changed to uppercase AIKO and unified theme color |
| Team Group Architecture Upgrade | Team layout aligned with BiBrain, supports group task initial prompt template, collaboration flowchart disables button and prompts reason when task cannot be accepted |
| Session List Scroll Loading | Session history changed to scroll dynamic loading (initial 12 entries, auto-append on scroll to bottom), resolves issue where history only showed latest 12 sessions |
| HUB Search by Username | HUB resources support searching by username and nickname, fixes no results when filtering by author |
Bug Fixes
| No. | Problem Description | Solution |
|---|---|---|
| 1 | Background sub-agent card only shows "Working..." | Completed background path tool_use/result/tool completion event passthrough, sub-agent card normally displays steps and tool details |
| 2 | Project mode layout swap/drag buttons not working | Fixed left-right swap button invalid and drag range limited issues, aligned with non-project mode |
| 3 | Permission mode hover tooltip needs to wait 1 second | Permission selector and target ring hover switched to self-drawn instant tooltip, no longer relies on browser native delay |
| 4 | Full-auto selector text coloring abnormal | Full-auto mode text changed to black, removed blue override eliminates hover flicker |
| 5 | HUB skill search by author filter no results | Falls back to frontend fuzzy filtering when server-side does not match |
| 6 | Diff patch display issue | Fixed code change difference display and style issues |
| 7 | Host prompt language rules conflict with kernel | Removed "first message" language rules from three host prompts, unified by kernel fallback; fixed result strategy self-contradiction |
| 8 | Right-click menu background transparent | File browser right-click menu background changed to opaque color |
| 9 | Check update build error | Fixed build error in version update check |
| 10 | BiBrain service address overwritten | Preserved user-customized bibrainServerBaseUrl value |
| 11 | Memory module description inaccurate | Updated cloud profile and local memory description text |
Code Cleanup
| Description |
|---|
| Removed old SWE-bench evaluation framework, switched to Harbor standardized adapter (ATIF trajectory format) |
| Removed sub-agent ConfirmCheckpoint tool |
Historical Versions
v3.1.4
Release Date: July 24, 2026
Release Note
This version is SDW AIKO V3.1.4, adding Target Mode (/Target), AGENTS.md project specification support, AI-generated commit messages, Exa semantic search, API/tool invocation timeout settings, HUB resource publishing enhancement and other features, greatly optimizing context compression stability (summary degradation/estimation distortion/mid-loop deadlock/pollution detection), and fixing multiple compatibility and experience issues.
New Features Added
| Feature | Description |
|---|---|
| New Target Mode (/Target) | Added TargetController state machine, /target slash command, target_stepped/target_closed events and rollout snapshot, supports multi-step goal auto-continuation; Team task advancement loop unified migration to /target bridge |
| AGENTS.md Support (AI Programming Universal Specification) | Auto-loads AGENTS.md from project root (CLAUDE.md fallback compatible) and injects into system prompt, Agent automatically follows project build, specs and conventions, no need to manually paste context |
| AI-generated Commit Messages | Generates Conventional Commits style commit messages based on real git diff, language follows UI |
| Exa AI Semantic Search | Added Exa AI search provider support, complete frontend and backend integration |
| Chat Area and File Panel Swap | Top bar added swap button, one-click left-right swap of chat area and file panel |
| API/Tool Invocation Timeout Settings | Supports custom API timeout and tool invocation timeout duration |
| HUB Resource Publishing Enhancement | Supports version-selectable publishing, resource takedown and publisher homepage |
| Code Review Local Credential Authentication | Repository authentication switched to local git credentials, added "pending authorization" status and authentication pre-check |
| Model Selector In-group Sorting | Within provider groups sorted alphabetically by name, same-series models displayed adjacently |
| Lightweight Model Query Auto-downgrade | Auto-downgrades to fallback model when quick model incompatible, with downgrade prompt |
| Task View Sub-agent Card | Sub-agent invocation independently rendered as Agent card with step list |
| Knowledge Base Loading | Project settings scan and load knowledge base directory, content injected into system prompt |
Bug Fixes
| No. | Problem Description | Solution |
|---|---|---|
| 1 | Compression summary degradation (output "Hi.") | Separator + first-person instruction prevents trailing user message semantics being overridden by preceding text after merging |
| 2 | Context estimation distortion and mid-loop compression deadlock | Fixed three root causes: fallback path underestimation, same-source anchor case misjudgment, mid-loop overflow window exemption |
| 3 | Auto compression main path moved to turn boundary | Fixed session permanent freeze caused by mid-loop compression result silently discarded |
| 4 | Claude 4.7+ thinking/temperature compatibility | Parameters branched by model generation, 400 error no longer misreported as service exception |
| 5 | LLMStreamIdleTimeout false positive | Idle timeout switched to emergency compression resend, no longer blindly retries three times wasting 12 minutes |
| 6 | Artifact deduplication by file identity | Fixed duplicate file display caused by inconsistent write/edit and Bash scan paths |
| 7 | Scheduled task weekday selection offset and Sunday error | Fixed UI and cron engine weekday encoding inconsistency (0-6 vs 1-7) |
| 8 | DeepSeek long thinking tool invocation silently truncated | Raised model output limit and added truncation recovery mechanism |
| 9 | Compression summary pollution detection | Detects tool call marker pollution and auto-retries, polluted text not written to history |
| 10 | Engine freeze after interruption | Pre-interrupt answer pending approval, startup phase adds timeout guardrail |
| 11 | Goal-task user interrupt auto-resumed | Interrupt goes through cancel_task to terminate directly, no longer resumed by auto_recover |
| 12 | Channel CLI GBK encoding garbled | Windows GBK fallback decoding, fixes Chinese error message all garbled |
| 13 | Fresh install default creates empty DeepSeek provider | Only migrates when legacy non-empty API Key exists, cleans up empty key providers on startup |
| 14 | Mechanical degradation compression missing summary boundary marker | Added summary/boundaryOriginal persistence requirements and anchor reset |
| 15 | BiBrain sync username mismatch | Sync registration explicitly sends username, fixes email prefix inconsistent with username |
| 16 | Email validation error message not exposed | Exposes exit code and original error on failure, no longer uniformly displays expired |
| 17 | TEAM/task session engine MCP warmup missing | New engine dispatches MCP servers config, fixes MCP tools unavailable |
| 18 | Replay card rendering exception | Fixed Team replay card rendering issue |
| 19 | Gitignore recursive exemption insufficient | Fixed nested target directory source silently ignored causing file miss in commit |
| 20 | Dropdown menu popup position error | Fixed model selector and attachment menu popup direction calculation |
| 21 | Today's highlights history item time overwritten | Fixed history item message time being overwritten |
| 22 | Cannot review early details after compression | Summary end restores session record path hint, four compression paths unified coverage |
| 23 | Skill card column width uneven and badge overlap | Fixed three-column grid uneven width, long titles truncated to ellipsis no longer under badge |
Code Cleanup
| Description |
|---|
| Error classification phrase criteria unified to error-taxonomy single source |
| Compression threshold unified to 90%, summary changed to fork self-narrative style |
v3.1.3
Release Date: July 17, 2026
Release Note
This version is SDW AIKO V3.1.3, adding CLI unattended mode, user self-service activation code and License issuance, Team task re-execution and session replay, homepage data source manual summary triggering and other features, greatly optimizing desktop performance (eliminating persistent GPU usage from frosted glass, frontend debouncing and repaint layer reduction) and Linux stability, and fixing multiple issues including context compression deadlock and notification lag.
New Features Added
| Feature | Description |
|---|---|
| CLI --ci Unattended Mode | Supports CI/CD one-click execution, oneshot + JSON output + auto-disable interactive tools |
| Full-auto Mode Optimization | Only retains hard-baseline confirmation, sensitive list no longer interrupts with popups |
| Custom Sub-Agent Recovery | Restores AGENT.md recognition and invocation capability |
| Authorization Check Log Persistence | Authorization check results written to aiko_license.log, startup failure can be self-checked |
| Team Task Re-execution | Team tasks support two "re-execute" methods, returning to the initial form |
| Team Session Replay | Team sessions add replay functionality |
| Authentication Management Enhancement | Authentication button always visible, added "Delete Authentication" operation |
| Homepage Data Source Manual Summary | Homepage supports manual triggering of data source summary |
| Artifact Path Optimization | Artifact hover tooltip changed to show relative path |
| Data Source Refactoring | Data source depends only on CLI, decoupled from external dependencies |
| User Self-service Activation and License Issuance | User self-service activation code and License issuance feature launched |
| Windows Installation Multi-language | Windows installer supports multiple languages |
Bug Fixes
| No. | Problem Description | Solution |
|---|---|---|
| 1 | Context compression deadlock causing long task freeze | Mechanical degradation fallback + recoverable circuit breaker, solves long task freeze issue |
| 2 | set_model model name collision exception | Fixed set_model model name collision issue |
| 3 | Agent.md content distortion when delegating sub-agent | Agent.md original text injected, eliminates LLM summary distortion |
| 4 | LLM request payload redundancy | Streamlined request payload, reduces token waste |
| 5 | Login lag | Fixed login process lag issue |
| 6 | chat cwd lost across restart | Preserved chat cwd state across restart |
| 7 | File browser panel semi-transparent | Fixed file browser panel semi-transparent display issue |
| 8 | Dynamic full-auto permission switching anomaly in session | Fixed dynamic full-auto permission switching issue in session |
| 9 | Homepage color scheme not following theme | Homepage color scheme follows theme skin |
| 10 | Code Review local repository storage path error | Fixed Code Review local repository storage path |
| 11 | Notification popup ghosting/lag | Fixed notification popup ghosting and lag issue (3 commits) |
| 12 | Bell notification time inaccurate | Bell notification time changed to use real event time |
| 13 | High GPU usage on all platforms | Removed persistent frosted glass backdrop-filter, reduced homepage persistent repaint layers, removed decorative animations |
| 14 | Resource change events causing frontend lag | Frontend debounce + single-flight optimization |
| 15 | Ubuntu/WebKitGTK lag and high GPU usage | Fixed rendering performance issues on Ubuntu/WebKitGTK environment |
| 16 | Multi-webview concurrent xcb crash (Linux) | XInitThreads fixes multi-webview concurrent crash |
| 17 | Read-only refresh storm (Linux) | File watcher filters by event type, eliminates read-only refresh storm |
| 18 | DMG expansion script syntax error | Fixed DMG expansion script syntax error |
| 19 | DingTalk authentication misjudged as connected | No longer misjudged as "connected" when authentication cannot be confirmed |
Code Cleanup
| Description |
|---|
| Removed old HUB and standalone marketplace code |
| Cleaned up expired License scheme documentation |
v3.1.2
Release Date: July 7, 2026
Release Note
This version is SDW AIKO V3.1.2, adding a brand-new homepage and Team collaboration module, DingTalk/Feishu/Email channel data source access, scheduled summary and notification center, multi-session parallel and split-screen, file online preview, memory system enhancement and other features, greatly optimizing multi-session stability and compression reliability, and fixing multiple user feedback issues.
New Features Added
| Feature | Description |
|---|---|
| Brand-new Homepage | Homepage fully redesigned, added cards for today's highlights, running tasks, recent executions, resource and model status connected to real data, sidebar layout and theme style synchronized optimization |
| Team Collaboration Module | Added Team module, supports creating teams (selected members join immediately), @members, team settings and workspace switching, can directly create AIKO sessions within team, team dynamics sync to homepage, collaboration flowchart accept/cancel task adds confirmation dialog |
| Channel Data Source Access | Built-in DingTalk, Feishu, Email three channel skills and CLI programs, message pulling and consumption decoupled and landed in local inbox, added data source settings page and chat injection, email local direct connection IMAP/SMTP |
| Scheduled Summary and Notification Center | Supports DingTalk/Feishu/Email scheduled summary tasks, added general file message bus and self-drawn notification popup (replacing system native toast), notification bell real-time refresh, added send-notification skill to push in-app notifications to notification center |
| Today's Highlights "Process Now" | Today's highlight items support one-click entry to new session with data source context and original message reference, supports inline input to create new highlights |
| Marketplace Upgrade | Marketplace added Agent category and likes/favorites/downloads query, supports .skill format import and skill category auto-inference, marketplace auto-login, UI fully upgraded |
| Web Search Toolchain | Added multi-search provider support and parallel racing search architecture |
| Memory System Enhancement | Added expert experience memory and user profile collection, supports positive experience accumulation and negative dead-end memory, added memory_write main model direct-write tool, accumulated experience auto-mirrored to BiBrain |
| Multi-session Parallel and Split-screen | Supports multiple simultaneous conversations and conversation split-screen, engine bucket limit raised to 20, running state supports follow-up queue and steer insertion, sidebar restores running state icon |
| File Preview Enhancement | Project files support PDF/Word/Excel/PPT online preview and auto-typesetting, file links in task panel can be previewed in popup in-place |
| BiBrain Task Enhancement | Tasks support scoring, custom initial prompts, working directory selection and task restart after interruption, conversation tasks support inputting related info when started, tool invocation reported to BiBrain |
| Experience Optimization | Supports custom avatar, setting shortcuts, drag-and-drop file upload in conversation, session input box overflow expand/collapse, session-level working directory and permission mode persistence, non-vision models prohibited from sending images and model image capability labeled |
Bug Fixes
| No. | Problem Description | Solution |
|---|---|---|
| 1 | Frontend lag and memory usage surging to 2GB during session sync | Session upload pipeline moved to Rust, fundamentally resolves sync lag and memory issues, logout anti-freeze |
| 2 | Windows app unresponsive after receiving first notification | Fixed self-drawn notification window triggering main thread deadlock |
| 3 | File content search full-disk scan causing freeze | Search tool switched to ripgrep and built into install package, maintains fast search when system has no rg |
| 4 | Bash output path scan regex catastrophic backtracking causing entire window freeze | Fixed scan regex |
| 5 | Repeated compression after reopening session, manual /compact race condition and countdown not stopping | Fixed compression state race condition |
| 6 | Manual stop during compression causing session permanent freeze | Fixed stop logic, compression persistence supports reopen continuation |
| 7 | Application abnormal termination during /compact compression | Fixed un-awaited promise rejection escalated to process exit by global handler |
| 8 | Context usage surging after compression | Post-compression re-injection slimming and honest validity judgment |
| 9 | Empty signature thinking block causing session polluted by 400 | Discard empty signature thinking blocks, also fix 400 error when model returns empty signature |
| 10 | Task silent interruption or misjudged as complete | Fixed three types of completion judgment issues: model writing tool call as text, thinking-only idling, incomplete rounds |
| 11 | Tool image reading invisible to model when going through translation gateway, answering based on hallucination | Fixed image passthrough pipeline |
| 12 | Large request body (accumulated large images/complex project files) hanging gateway, software freeze | Added request body protection, relaxed connection timeout |
| 13 | TUI long thinking rendering freeze | thinking stream only re-renders tail each frame, eliminates O(L²) overhead |
| 14 | macOS input method enter accidentally sending message | Fixed input method combination key judgment |
| 15 | Attachment type small icon mismatch (xlsx displays as PDF) | Attachment tag changed to infer by file extension |
| 16 | After exit, bun/aiko process leak, command line black window pops up while running | Cross-platform binding engine subprocess lifecycle, hide console window |
| 17 | Scheduled task reusing engine loses full-auto permission mode, repeatedly triggers cold start engine | Fixed permission mode inheritance and engine hot reuse |
| 18 | Quick switching session cannot switch/switches wrong | Fixed session switching race condition |
| 19 | Gateway model and built-in model name collision but inconsistent protocol causes 404 request | Fixed model routing match logic |
| 20 | Multi-provider API Key mutual influence | Removed broadcast logic on credential save |
| 21 | False permission confirmation popup appears | Engine broadcasts approval result ensures frontend-backend state consistency, also fixes forced popup when temporary workspace write misjudged in full-auto mode |
| 22 | In-round process killed causing message loss, UI inconsistency | Session messages changed to per-message async persistence |
| 23 | Long startup white screen time | Startup entry chunk slimming, shortens first screen white screen time |
v3.1.1
Release Date: June 16, 2026
Release Note
This version is SDW AIKO V3.1.1, adding Bibrain WebSocket real-time message notification, session running status icons, Windows Git Bash execution environment, Webhook execution history, skill marketplace security audit and other features, greatly optimizing compression stability and gateway fault tolerance, and fixing multiple tool invocation and task management issues.
New Features Added
| Feature | Description |
|---|---|
| Bibrain Real-time Message Notification | Uses WebSocket to synchronize Bibrain message notifications, supports notification panel, Toast prompts and multiple languages |
| Session Running Status Icon | Sidebar session title shows three-state status icons: running (blue spinner), idle engine (green dot), no engine (hidden); disappears immediately after the engine is evicted by LRU |
| Windows Git Bash Execution Environment | On Windows, the bash tool prioritizes Git Bash to execute commands, eliminating incompatibility between cmd.exe and POSIX syntax, supports AIKO_SHELL_PATH override |
| Webhook Execution History | Each Webhook/scheduled task trigger leaves a record in the "Execution History" tab, showing trigger time, completion time, duration and status |
| Skill Marketplace Multi-iteration | Added archive extraction (7z/rar/tar), skill security audit and scoring, user profile page, knowledge base page replacing plugin page, refactored marketplace paths and navigation |
| Skill Creation Logic Refactoring | Skill default approval and visibility toggles changed to off, new skills automatically set to public visibility, optimized document tag scrolling and infinite loading |
| AIKO CLI Auto-install | Automatically installs the aiko cli command when installing AIKO, supports loading native Agent definition files as sub-agent system prompts |
Bug Fixes
| No. | Problem Description | Solution |
|---|---|---|
| 1 | Compression no-thinking failure causing excessively long compression time | DeepSeek-compatible endpoint explicitly sends thinking disabled, compression time reduced from 40-232s to about 9s |
| 2 | Context progress bar not refreshing in real time | Changed to refresh in real time on each agent step, reusing real usage from assistant events, displays >100% honestly when overflowing |
| 3 | Path hallucination when searching .claude files using Claude model | Eliminated model path hallucination, fixed prompt construction logic |
| 4 | MCP tool returning screenshots triggers gateway 400 | Reused downsampling protection, eliminated gateway errors triggered by large images |
| 5 | Approval backfill parameters cause tool crash | Three-layer protection fixes the regression where updatedInput overwrote validated parameters, tools like Read/Bash/Grep/Edit no longer crash |
| 6 | Inflated context usage after compression | Uses summary message timestamp as compression boundary, skips expired anchors, optimized summary template into 8-segment structure |
| 7 | Iterative summary mechanism redundant | Removed previousSummary state, each compression changed to an independent one-time summary |
| 8 | Unfriendly MCP connection error prompt | Optimized error prompt message when MCP connection fails |
| 9 | Gateway 429 rate limiting causes session freeze | Added rate-limit matching rules and dedicated backoff strategy, added effectiveContextWindow to prevent compression threshold from being inflated by falsely reported contextWindow |
| 10 | AskUserQuestion stops or enters dead loop after submitting answer | Fixed updatedInput passing, no-answer branch changed to isError to interrupt repeated questioning, improved bash Windows prompts |
| 11 | Inline Skill invocation with parameters loses user task text | Automatically appends original input to the end of rendered result when no placeholder is declared |
| 12 | SSE stream JSON parsing error causes dialog failure | Classified JSON parsing errors as retryable server-side errors, performs silent retry with exponential backoff |
| 13 | Scheduled task/Webhook disappears from schedule after termination | Termination changed to stopping the current execution and returning to Scheduled status, no longer hard-deletes the task |
| 14 | Background bash child process becomes orphan and keeps running | Started as an independent process group, killTask changed to use negative PID to kill the entire process group |
| 15 | Stream idle watchdog misjudges long thinking as freeze | Default timeout adjusted from 60s to 180s, adapting to block-buffered reasoning phase scenarios |
| 16 | CJK token estimation too low causing context overflow | Chinese/Japanese/Korean estimated at about 0.9 token/character, fixing the original chars/4 underestimation of 3-4 times |
| 17 | Garbled tool invocation result display | Fixed bash tool execution result encoding issue |
| 18 | aiko-baseline.exe error reported during installation | Fixed installation script and engine entry configuration |
| 19 | tsc build failure | Removed empty @ts-expect-error directive |
v3.1.0
Release Date: June 15, 2026
Release Note
This version is SDW AIKO V3.1.0, adding general memory system (Genesis Memory), automatic compaction during mid-loop context, sub-agent architecture refactoring and prompt management, TUI interactive approval rewriting, Hook mechanism, file-level cross-restart rollback, automatic Code Review triggering and other features, greatly enhancing compression stability and sub-agent output integrity, and fixing multiple user feedback issues.
New Features Added
| Feature | Description |
|---|---|
| Genesis Memory | Brand-new general memory architecture, includes memory-core pure library (12 modules), aiko-agent six-point wiring, GUI card panel (profile/facts/experience management, filtering, archive/restore/pinning), automatic migration on startup, explicit memory signals ("remember/remember") and prompt v2, enabled by default |
| Mid-loop Compaction | Supports in-loop detection of token over-threshold and automatic compaction triggering during long sub-agent tasks, solves context infinite expansion problem; migrated to beforeStep hook unified compaction entry |
| Automatic Compaction Circuit Breaker | Stops trying after 3 consecutive compression failures, avoids infinite retry dead loops; emits UI event to show "compression failed" on failure |
| Compression In-progress Indicator | Added "compressing..." status indicator, supports Chinese/English/Japanese three languages |
| Compression Event Protocol Alignment | Compression events aligned to compact_boundary format, added CompactionIndicator component to show compression reason, released token count and summary content |
| Sub-agent Prompt Management Refactoring | Built-in agent definitions migrated to independent files (explore/plan/verify/general/fork), added Fork boilerplate injection and "When to fork" guidance, fixed default routing |
| Agent Factory Pattern | Unified main agent and sub-agent construction, added main logic comments |
| Hook Mechanism | Implemented hook mechanism, supports both command + prompt hooks, covers 11 events (PreToolUse/PostToolUse/Stop etc.), includes matcher regex and if condition filtering |
| File-level Rollback (Cross Restart) | rewind_files refactored from memory snapshot to disk backup scheme, supports file-level rollback across engine restarts |
| acceptEdits Permission Mode | Added acceptEdits permission mode, automatically approves editing tools, bash/mcp still require confirmation |
| MCP Elicitation Bridge | MCP server can initiate queries to users, three-layer transparent transmission, supports kotei Type-B checkpoint |
| TUI Interactive Approval Rewriting | TUI rewritten from Ink/React to pi-tui, completed interactive approval, AskUserQuestion, diff preview, image input, session-replay, tasks browser |
| Code Review Auto-trigger | AIKO automatically fetches repository updates and calls Code Review sub-agent |
| Feedback Editing & Deletion | Feedback大厅 submitted tickets support editing and deletion, added my feedback panel |
| User Message Auto-collapse | User messages exceeding 200px collapsed by default, hover shows expand button, supports multiple languages |
| Context Compression Threshold Config | Added configurable context compression threshold and model context length |
| Project Intermediate Artifact Auto-cleanup | In project mode, .tmp/ intermediate artifacts automatically cleaned up, distinguishes final artifacts from intermediate files |
| MCP List Refresh | Added MCP list refresh UI |
| Image Copy Function | TUI mode supports image copy and boundary fault tolerance |
| Auto-extract Project Name | Auto-extract default project name from local/Git repository path when creating new project |
| Global ESC Key Interrupt | Supports global ESC key to interrupt engine execution |
| Bibrain Task Lifecycle Optimization | Switch workspace refresh config, remote cancel sync local, orphan task cleanup, aggregate interface replacement |
Bug Fixes
| No. | Problem Description | Solution |
|---|---|---|
| 1 | Dialog line break has no prompt | Added shift+enter line break prompt to input box |
| 2 | Generated document format cannot be selected, format conversion failed | Optimized artifact generation process, improved file output format management |
| 3 | SDW thinking timeout no response when pressing Esc | Added global ESC key to interrupt engine execution, improved interruptEngine state cleanup |
| 4 | Feedback ticket cannot be edited | Added feedback editing and deletion functionality |
| 5 | Dialog still spinning after terminating task | Extended clearLoadingStates function, cleans up streaming markers and all output item loading states |
| 6 | Generating PPT file delivers py file as artifact | Improved temporary file filtering logic, covers .tmp directory and _tmp_ intermediate artifacts, optimized artifact identification |
| 7 | New conversation still shows previous context usage | Fixed ModelPicker reverting to default model after restart, follows persisted model after sync effect and other settings loaded |
| 8 | Artifact identification inaccurate, generates image but delivers py file | Optimized artifact display and output rules, separates intermediate from final artifacts, supports bash command artifact extraction |
| 9 | Compression mechanism token count always reads uncompressed state causing repeated compression | Rebuild session tree after compact/reactiveCompact succeeds, fixes token counting |
| 10 | Iterative compression always loses previous template summary | Writes previousSummary after reactiveCompact succeeds, fixes iterative compression context loss |
| 11 | Manual compression and automatic compression may execute concurrently | Added compacting mutex flag prevents concurrent compression |
| 12 | Manual compression still has AI reply after | Slash command bypasses FILE WRITE RULES injection, fixes /compact being treated as normal message |
| 13 | Sub-agent result loss | extractResultText rewritten, completely拼合 multiple segments generated by continuation; agent added to exemption list prevents result truncation |
| 14 | Sub-agent output truncated | max_completion_tokens increased from 16384 to contextWindow, removed disk spill replaced with in-place truncation |
| 15 | Sub-agent tool result invisible | turn_end transparently transmits sub-agent tool result to UI, removed redundant idle event fixes engine busy state race condition |
| 16 | Ghost tool call (empty parameter tool_call) | Filters empty parameter ghost tool_call, added dedupeToolCalls merges duplicate tool_use |
| 17 | After error handling user message bubble not displayed | Added clearing selectedBranch=-1 sentinel in three error/fallback result handlers |
| 18 | First round conversation cannot load and recognize MCP Server | Fixed first round conversation MCP Server loading recognition issue |
| 19 | Online marketplace installed Agent status display failed | Fixed installed Agent status display logic in online marketplace |
| 20 | MCP reconnect button invalid | Fixed MCP reconnect button invalidation issue |
| 21 | Artifact path cross-platform matching issue | Added normalizeArtifactPath utility function unifies different file path formats |
| 22 | Local task unavailable when Bibrain not configured | Fixed local task availability when Bibrain not configured |
| 23 | Click validation and close doesn't work | Fixed interaction issue with click validation and close button in collaboration flowchart |
| 24 | Preview sound doesn't work after packaging | Fixed task prompt preview sound not working after packaging |
| 25 | todo tool accidentally truncated | EXEMPT_TOOLS naming fixed todo_read/todo_write → todoread/todowrite matches actual registration name |
| 26 | Sub-agent duration always 0 | writeSubagentMetadata startTime changed to captured variable |
| 27 | Fork mode security boundary vulnerability with disabled tools | Fork mode also filters SUBAGENT_FORBIDDEN_TOOLS |
v3.0.10
Release Date: June 5, 2026
Release Note
This version is SDW AIKO V3.0.10, adding @mention file picker, Auto Compact automatic context compression, Bash dangerous command interception, sub-agent parallel execution, Code Review module, memory document system and other features, refactored streaming rendering pipeline greatly improving conversation fluency, optimized project file refresh performance and scheduled task experience, and fixed multiple user feedback issues.
New Features Added
| Feature | Description |
|---|---|
| @mention File Picker | Chat input box supports @ mentioning files/folders, supports search filtering, keyboard navigation and selected insertion |
| Auto Compact | Added automatic context compression mechanism, adapts to prompt too long scenario, automatically triggers context compression to avoid exceeding limit |
| Bash Dangerous Command Interception | Added Bash command safety interception mechanism, automatically intercepts high-risk commands like rm -rf /, sudo rm -rf /, normal commands are not affected |
| Sub-agent Parallel Execution | Refactored tool execution pipeline, supports true parallel execution of multiple sub-agents, fixes budget pollution and timeout interruption issues |
| Sub-agent Background Invocation | Background sub-agent output switched to JSONL single data source, added tool disable enhancement and recursive protection |
| Plan Mode | Added Plan Mode tool, supports entering/exiting plan mode for task planning |
| Code Review Module | Added Code Review module, supports enabling/disabling Code Review listening for specified Git repositories in conversation, review results displayed in dedicated panel |
| Memory Document System | Added memory document view and JSON read/write, supports Markdown editing and save-back for profile/facts memory, automatically injects memory context into new conversation |
| Remote Knowledge Base Fusion Search | Added remote knowledge base synchronization and fusion search function, supports MCP remote knowledge base metadata synchronization and vector search |
| Streaming Rendering Pipeline Refactoring | Added deterministic stream-reducer replaces fragile streaming marker scanning, folding tool group changed to stable count summary, tool bubble delayed rendering and fade-in animation |
| Log Writing Independent Task | Engine log writing decoupled from read loop into independent tokio task, bulk disk writing via channel, avoids blocking hot path due to disk flushing |
| Scheduled Task Session Mode | Scheduled tasks support reuse/fresh two session modes, old session archived retains history in fresh mode |
| Webhook Task Editing | Scheduled plan list supports distinguishing trigger types (Webhook/file monitor/scheduled), Webhook tasks use dedicated edit dialog |
| Scheduled Task Workspace Editing | Scheduled task editing supports modifying workspace path, edit page pre-fills existing configuration |
| Collaboration Flowchart Optimization | Optimized collaboration flowchart line display, supports viewing data flow graph by Feature, adds shortcut jump buttons when task expands |
| Agent Task Multi-language | Agent task related text fully integrated multi-language support (Chinese/English/Japanese) |
| Marketplace Page Refactoring | Refactored marketplace page UI, card installation status changed to badge display, details popup changed to modal, added top classification navigation bar |
| File MIME Type Extension | Added MIME type support for SVG/ICO/BMP files, optimized binary and Office file reading logic |
| Login View Apply Entry | Login popup added apply jump button, server address changed to clickable link |
| Bash Command Result Persistence | Bash tool execution results support streaming persistence, archived by session path for historical traceability |
Bug Fixes
| No. | Problem Description | Solution |
|---|---|---|
| 1 | Output artifacts cannot refresh display in project folder | Optimized project file refresh logic, added useMemo caching, concurrent read限制, debounce throttling control and ignore directory filtering |
| 2 | SDW memory leak, freezes during automatic analysis | Optimized file change listening and directory refresh logic, fixed parent directory not loading issue when refreshing automatically |
| 3 | Memory Prefetch caused hardcoding Sonnet model | Fixed hardcoding getDefaultSonnetModel() in related memory recall, defaults consistent with main session model |
| 4 | Hardcoded claude-haiku-4-5 when quick model not configured | Defaults consistent with main session model when quick model not configured, no longer hardcoded |
| 5 | Busy engine does not passively restart causing session switch deadlock Working | Added busy_flag mechanism, skips stop restart reuse existing process when engine busy, adds fallback补发 engine:stopped event avoids frontend silent deadlock |
| 6 | Dialog folder selector default path inconsistent with current session working directory | Fixed folder selector default path synchronized with current session working directory |
| 7 | HTML preview cross-origin resource loading failed | Uses Tauri convertFileSrc handles local paths, adds allow-same-origin permission to iframe sandbox |
| 8 | Old tool bubbles still display Working status | Fixed issue where tool group still shows "running" after switching session branch, isRunning added engineBusy fallback |
| 9 | AIKO Agent tool events incomplete causing frontend state error | Fixed toolcall_start push missing content writing caused tool_id empty, tool bubble permanently stuck pending issue |
| 10 | AIKO Agent thinking streaming text duplication | Removed redundant message_start event emission,补上 missing text_start processing |
| 11 | Parallel sub-agents cannot return normally | Closed Agent tool 120-second timeout limit, fixed budget key pollution parent budget issue |
| 12 | Webhook task editing shows scheduled task template | Webhook task editing uses dedicated dialog, distinguishes trigger types |
| 13 | Scheduled task editing does not show existing Prompt | Fixed startEdit pulls complete task回填 prompt from taskGet |
| 14 | Scheduled task repeatedly triggers same session has context overflow risk | Added fresh session mode, supports creating new independent session each trigger |
| 15 | After scheduled task manually deleted task still running but UI shows nothing | Detects invalid session reference automatically creates new session avoids dangling reference |
| 16 | Open containing folder path logic error | Fixed file explorer open containing folder path calculation |
| 17 | AIKO restart re-runs completed tasks | Fixed state judgment logic for restarting completed tasks |
| 18 | Blank directory display when switching projects due to lazy loading requires manual collapse expand | Fixed incorrect reset of lazy loading state when switching projects |
| 19 | Answer text duplication (ghost message appears from second round in multi-round conversation) | Removed redundant message_start event emission, fixed incorrect opening of non-first round text blocks |
v3.0.9
Release Date: June 1, 2026
Release Note
This version is SDW AIKO V3.0.9, adding true streaming output, session grouping management, conversation forking, sub-agent checkpoint confirmation, PlantUML preview, project file lazy loading and memory restoration and other features, greatly optimizing conversation fluency and response speed, and fixing multiple user feedback issues.
New Features Added
| Feature | Description |
|---|---|
| True Streaming Output | Changed pseudo-streaming to true streaming output, greatly reduces first token response time, eliminates conversation stuttering |
| Session Group Management | Supports managing historical sessions by both time and custom grouping, can create/rename/delete groups, move sessions to specified groups |
| Conversation Forking | Supports forking new conversation from any historical message, preserves complete context |
| Sub-agent Checkpoint Confirmation | Added ConfirmCheckpoint tool, sub-agent can confirm intermediate results to user or main agent during execution |
| PlantUML Preview | Project files support PlantUML sequence diagram online rendering preview, automatically falls back to server rendering when local rendering fails |
| Project File Lazy Loading | File tree loads first two levels by default, remaining subdirectories load on demand, improves large project loading performance |
| Project Memory Restoration | Automatically restores last file browsing state and tabs when reopening project |
| Normal Session Working Directory Switching | Supports switching working directory during normal conversation |
| Loop Detection Mechanism | Added loop detection service prevents model from falling into infinite optimization dead loop on same file |
| Streaming Timeout Protection | Streaming response added watchdog timeout mechanism (90s), shows prompt and supports cancellation on timeout |
| Waiting Input Notification | When AI issues confirmation request, triggers sound notification, system notification and task bar blink if window not focused |
| Collaboration Flowchart | Added task collaboration flowchart, visually shows task flow and dependencies |
| Agent Task Unattended Execution | Added TaskExecutor, supports automatic Judge loop and automatic approval for scheduled/Webhook tasks |
| Image Upload Optimization | Optimized image compression and upload validation logic, supports automatic scaling for extra-large images |
Bug Fixes
| No. | Problem Description | Solution |
|---|---|---|
| 1 | AI answer running over 30 minutes no response | Changed to true streaming output, displays generated content in real-time |
| 2 | Generated md file empty when opened | Fixed artifact generation path parsing issue, sets working directory for chat session |
| 3 | Document content off-screen cannot be displayed and operated | Adjusted UI, md document out-of-bounds can be displayed and operated normally, supports embedded HTML preview and fullscreen |
| 4 | Cannot input HTML format content | Supports content input in JSON, HTML and other formats |
| 5 | Cannot see changes after modified file | Supports artifact difference viewing |
| 6 | Too many historical sessions cannot quickly find | Added custom session grouping function |
| 7 | Project file cannot remember last opened path | Added project-level file browsing state persistence |
| 8 | Window does not automatically scroll to bottom during generation | Optimized automatic scrolling logic prevents false triggering |
| 9 | Stuck thinking for ten minutes with no response | Added timeout logic and error prompt mechanism |
| 10 | AI occasionally falls into infinite loop thinking | Added loop detection interception mechanism |
| 11 | Mermaid parsing error causes UI crash | Added rendering error capture and friendly prompt |
| 12 | No prompt for task interruption/completion/query | Added sound, notification and task bar blink reminders |
| 13 | Loading shows blank after importing large project | Optimized to lazy loading, only shows two levels by default |
| 14 | Session filtering escape causes conversation streaming to wrong session | Fixed session filter escape condition causing streaming cross-talk issue |
| 15 | UML format preview not supported | Added PlantUML online rendering preview support |
| 16 | Retry carries previous answer as context | Fixed context accumulation problem on retry, uses rewind/clear to cut off old context |
| 17 | YAML extraction loses multi-line YAML content | Extracted shared frontmatter module unifies block scalar processing |
| 18 | MCP add save error | Optimized default id processing logic when no id |
| 19 | Image upload reports dimensions exceed 2000x2000px | Optimized upload compression logic, removed size limit check |
| 20 | Directory does not automatically refresh | Triggers file tree refresh when task completes or project updates |
| 21 | Output occasionally appears folded | Fixed streaming bubble first frame starting from zero length issue |
| 22 | DeepSeek model thinking takes too long | True streaming output displays thinking process in real-time |
v3.0.8
Release Date: May 22, 2026
Release Note
This version is SDW AIKO V3.0.8, adding Plan mode, personal memory system, PM Agent Bibrain integration, artifact difference viewing, TUI interactive REPL and other features, greatly optimizing file browser experience and sidebar layout, and fixing multiple user feedback issues.
New Features Added
| Feature | Description |
|---|---|
| Plan Mode | Added Plan mode, supports planning and step-by-step execution for complex tasks, improves task controllability |
| Personal Memory System | Added Personal Memory system, supports manual/automatic memory extraction, /remember command one-click saves session memory |
| PM Agent Bibrain Integration | PM Agent integrated with Bibrain, allows creating tasks and having AIKO execute |
| Artifact Difference Viewing | Artifact card added "View Diff" button, supports Diff comparison for file changes |
| Input History | Conversation input box supports up/down arrow keys to browse input history (limit 50 entries) |
| Rich Text Pasting | Supports pasting tables, links and other rich text content, automatically converts to Markdown format |
| File Browser Enhancement | File preview supports image zoom drag, code highlighting, Office file opening, embedded HTML preview, fullscreen viewing |
| Task Completion Notification | Shows completion status prompt and summary information after task completion, automatically refreshes file tree when completed |
| Task System Optimization | Task system UI redesigned, added goal-loop convergence optimization, task recovery state fixes |
| TUI Interactive REPL | aiko-agent refactored interactive REPL based on Ink/React, solves streaming output and input coverage issues |
| VPN Login Support | Supports login authentication through VPN network |
| Simplified Knowledge Base Creation | Merged knowledge base creation process, creation done by just filling name + description, optional files |
| Sidebar Simplification | Simplified sidebar layout, low-frequency operations folded into menu, frees up space |
| Task Completion Notification | Plays notification sound and sends system notification after Agent task and Chat reply completed |
Bug Fixes
| No. | Problem Description | Solution | Related Case |
|---|---|---|---|
| 1 | Using AIKO not in maximized window, new project popup covered by sidebar | Automatically fold sidebar when window smaller than 980px | |
| 2 | Dialog + button skills menu difficult to select skill, menu disappears during mouse movement | Fixed attachment submenu closing incorrectly when mouse leaves | |
| 3 | Session content cleared and error reported not logged in after switching models | Fixed session clearing and login state abnormal when switching models | |
| 4 | Rewrite or regenerate reports error cannot continue | Fixed retry message error issue, fixed Windows send message start engine failure | |
| 5 | Input line by line text, AIKO removes line breaks changing format | Added pre-wrap style preserves line break format for plain text paragraphs | |
| 6 | Cannot complete knowledge base upload using md | Fixed knowledge base creation process and Windows path separator issue | |
| 7 | Cannot execute write command in project conversation AIKO | Fixed command unavailable issue caused by non-Claude model injected prompt | |
| 8 | Project Setting added skill does not appear in actual conversation | Fixed issue where added skill not displayed in project | |
| 9 | Document content off-screen cannot be displayed and operated | Adjusted UI, md document out-of-bounds can be displayed and operated normally, supports embedded HTML preview and fullscreen | |
| 10 | Switching project occasionally replaces file content | File preview changed to async loading adds loading state | |
| 11 | Mermaid parsing error UI crashes | Added mermaid rendering error capture and friendly prompt, cleans up dirty DOM | |
| 12 | Project folder and files display narrow cannot resize freely | Default open half width, supports drag to resize | |
| 13 | Copy table to input box shows as table, after sending shows text | Added HTML to Markdown tool, automatically converts when pasting rich text | |
| 14 | Input box does not recognize link text | Automatically parses to link format after pasting send | |
| 15 | No user notification when task completed, no summary information | Added task completion prompt status and summary information display | |
| 16 | Cannot see changes after modified file | Added artifact change difference viewing function | |
| 17 | Fixed "Client disconnected" caused by bucket mismatch in memory | ||
| 18 | Fixed undefined is not an object (usage.speed/iterations) issue | ||
| 19 | Fixed GPT OSS truncation bug during usage | ||
| 20 | Fixed missing concluding text after one round (summary silently discarded) | ||
| 21 | Fixed aiko-agent editing tool CRLF line ending incompatibility issue |
v3.0.7
Release Date: May 15, 2026
Release Note
This version is SDW AIKO V3.0.7, adding Completions API protocol support, AI Lab, online marketplace, skill replacement and other features, optimizes IM Gateway stability and model management experience, and fixes multiple known issues.
New Features Added
| Feature | Description |
|---|---|
| API Protocol Configuration | Added provider API protocol field, supports OpenAI Completions and Anthropic Messages protocols |
| Model Management Refactoring | Refactored add model process, supports batch adding via comma separation, Base URL filling prompt and official example URL one-click copy |
| AI Lab | Added AI Lab function page, built-in PM Agent switch |
| Online Marketplace | Added online marketplace function (pending verification) |
| Skill Replacement | Supports replacing existing skills via SKILL.md, compressed package or folder |
| Normal Conversation Temporary Workspace | Creates temporary workspace for normal conversations, supports file browsing |
| Normal MCP Picker | Normal conversation window added MCP picker, optimized selection panel interaction |
| Sub-agent Optimization | Optimized sub-agent access method, supports model selection and intermediate execution status display |
| IM Gateway Enhancement | Group chat session isolation, @mention configuration, Markdown rich reply, streaming reply, dangerous command filtering |
| Frontend Error Logging | Frontend errors and warnings unified written to backend log, facilitates troubleshooting |
| Feedback Type Completion | User feedback added type options, feedback displays user name |
| Session Statistics Enhancement | Session statistics displays user name and department information; unlogged users identified using MAC address |
Bug Fixes
| No. | Problem Description | Solution | Related Case |
|---|---|---|---|
| 1 | Enter Settings-Skills interface, after adding Skill click replace button no feedback | Fixed skill replace button click no response issue | TC-SET026 |
| 2 | Enter Settings-IM Gateway interface, Feishu backend preparation list incomplete | Fixed IM Gateway Feishu backend preparation list incomplete | TC-SET056 |
| 3 | Enter Settings-IM Gateway interface, timer jumps irregularly after clicking scan code | Fixed irregular timer jumping after scan code | TC-SET064 |
| 4 | QR code disappears before timing complete when scanning code | Fixed QR code disappearing early before timing complete | TC-SET064 |
| 5 | After connecting robot there is no unbind entry | Added IM Gateway robot unbind entry | TC-SET063 |
| 6 | Switching proxy and SEP_base toolchain causes content jump | Fixed file tree not correctly reset causing content jump after switching toolchain | TC-SET042 |
| 7 | Clearing conversation content when switching models | Fixed conversation content clearing issue when switching models | |
| 8 | PlanList popup position not updated correctly when stretching window | Fixed progress list rendering position incorrect when stretching window | |
| 9 | Project mode right directory tree cannot load complete project directory (includes git ignored files) | Fixed incomplete directory loading issue in project mode | |
| 10 | Fixed MCP Server unrecognized state display error issue | ||
| 11 | Fixed undefined is not an object ($input_tokens) caused by different model return values, fixes token calculation exception | ||
| 12 | Fixed DingTalk multi-message reply out-of-order, serializes per-conv message processing | ||
| 13 | Optimized response latency, eliminates engine cold start waiting; fixed repeated AIKO_PROMPT sending | ||
| 14 | Improved AIKO_PROMPT prompt, shields tool description summary speeds up response | ||
| 15 | Fixed OpenAI API request path extra /v1 prefix |
v3.0.6
Release Date: May 9, 2026
Release Note
This version is SDW AIKO V3.0.6, adding DingTalk robot integration, file modification statistics, duplicate project detection, resource package online upgrade and other functions, and fixes multiple known issues.
New Features Added
| Feature | Description |
|---|---|
| DingTalk Robot Integration | Added DingTalk robot QR code integration function, supports talking with AIKO via DingTalk |
| File Modification Line Count Statistics | Shows added/deleted line count statistics in message bubbles and file summaries |
| Duplicate Project Detection | Checks for duplicate name and path when creating new project, disables submission if duplicate exists |
| Resource Package Online Upgrade | Resource packages support online upgrade installation, optimizes update description and installation experience |
| IM System Prompts | Added IM system prompts, adds prompt information for Feishu messages |
| PM Agent Optimization | Optimized proxy configuration when PM Agent calls Redmine; forces UTF-8 encoding on Windows avoids Chinese garbled |
Bug Fixes
| No. | Problem Description | Solution | Related Case |
|---|---|---|---|
| 1 | In Projects interface, clicking template under Template repeatedly can keep adding template projects with exactly same name cannot modify | Check duplicate name and path when creating new project, disable submission if duplicate exists | TC-PR003 |
| 2 | Starting historical conversation replay, click dialog input box can input, after replay end input can enter conversation | Disables input box input and click during historical conversation replay | TC-AIKO015 |
| 3 | After adding Skill in dialog, send button cannot click but enter can send successfully | Fixed dialog send button cannot click issue | TC-CHAT009 |
| 4 | After selecting auto-approve edit mode, still asks for write permission | Fixed still asking permission issue in auto-approve edit mode | TC-CHAT016 |
| 5 | After downloading latest version install, reports CPU does not support AVX2 instruction set | Fixed grep, glob tools not included in version package causing CPU instruction set error report | |
| 6 | After error occurred when talking, click view log button but log is empty | Wrote stdout errors to err log, fixes empty log issue | |
| 7 | After adding MCP in settings, tests successfully saved but no related MCP displayed in dialog | Fixed MCP status bug, after adding MCP current kernel automatically perceives and activates | |
| 8 | When two same-name artifacts generated after file modification, one is empty content | Simplified file modification count logic, removed modified field avoids duplicate artifacts |
v3.0.5
Release Date: May 6, 2026
Release Note
This version is SDW AIKO V3.0.5, adding Feishu IM Gateway access, custom Agent, PM Agent skills, session analysis and other features.
New Features Added
| Feature | Description |
|---|---|
| IM Gateway Feishu Access | Supports scanning QR code to access Feishu, enables conversation interaction with AIKO via Feishu |
| Custom Sub-Agent | Supports custom Sub-Agent in C:\Users\{username}\.sdw\agents |
| PM Agent | Integrated project management Agent skill, supports Redmine project management operations |
| Quick Model Configuration | Introduced "quick model" configuration, used for auxiliary tasks like topic naming and summarization |
| Session History Upload | Supports uploading session history file to server |
| Session Analysis Page | Added session statistics analysis view and feedback view |
| HMI End-to-end Deployment | HMI end-to-end deployment service connected |
Changed Features
| Feature | Description |
|---|---|
| Cloud Desktop Compatibility | Fixed runtime compatibility issue in cloud desktop environment |
| Packaging Optimization | Optimized packaging time and build process |
Bug Fixes
| No. | Problem Description | Solution | Related Case |
|---|---|---|---|
| 1 | Operation: Open settings during dialog output. Expectation: Hierarchical content displays normally. Actual: Window content rendering penetration occurred. | Fixed window content rendering penetration issue when opening settings during dialog output | TC-CHAT022 |
| 2 | Operation: Set approval permission in one dialog, open another dialog with different settings. Expectation: Different dialog approval permissions do not affect each other, run independently. Actual: Changing permissions in one dialog affects other dialogs. | Added default permission configuration, permission switching only affects current session | TC-CHAT019 |
v3.0.4
Release Date: April 30, 2026
Release Note
This version is SDW AIKO V3.0.4, adding user feedback function, and fixing multiple interaction issues.
New Features Added
| Feature | Description |
|---|---|
| User Feedback | Added feedback function, supports adding images and user information reporting |
| Error Bubble Optimization | Added log entry on right side of error bubble; same error no longer pops up continuously, changed to update count |
| Default Permission Configuration | Added default permission configuration, permission switching only affects current session |
Bug Fixes
| No. | Description |
|---|---|
| 1 | Fixed issue where multiple input boxes cannot input on page |
| 2 | Removed invalid statistics Token count calls |
| 3 | Fixed grep、glob tool call failure issue |
| 4 | Fixed issue where settings window not closed after Skill "Try in Chat" |
| 5 | Supplemented missing vendor dependencies |
v3.0.3
Release Date: April 29, 2026
Release Note
This version is SDW AIKO V3.0.3, focuses on optimizing login and account system, improving collaboration stability between client and server.
Changed Features
| Feature | Description |
|---|---|
| Login Service Upstream Replacement | Login service docked new upstream interface, optimized authentication process |
| Quota Unit Changed to USD | Quota field changed from Token to USD display, uses Intl currency formatting |
| OA Employee ID Login | Login form changed to "OA Employee ID / OA Password", three-language translation synchronized |
| Kotei Cloud Provider Uniqueness | Automatically cleans up duplicate Kotei Cloud providers and orphaned models after login |
| Model Display Name Optimization | Model display name changed to model_id, avoids repeated names in dropdown |
| JWT Expiration Time | Login expires_at changed to parse JWT exp, default timeout adjusted to 10 minutes |
| Artifact Download | Added "Download" button to artifact preview |
| Skip Certificate Verification | Added skip certificate verification option, facilitates internal network debugging |
Bug Fixes
| No. | Description |
|---|---|
| 1 | Fixed issue where issue_api_key incorrectly judged as NO_API_KEY for approved accounts |
| 2 | Fixed R&D management platform address error |
| 3 | Fixed AI does not remember previous issues after reopening session |
| 4 | Fixed Token count getting logic exception |
| 5 | Performance optimization: ContextProgress does not poll when idle, interval adjusted to 60 seconds |
| 6 | Optimized status bar UI, fixed issue cannot drag on Mac |
v3.0.2
Release Date: April 28, 2026
Release Note
This version is SDW AIKO V3.0.2, adding client login function and custom title bar, improves cloud desktop compatibility.
New Features Added
| Feature | Description |
|---|---|
| Client Login | Supports OA account login, automatically configures Kotei Cloud provider and model after login |
| Custom Title Bar | Added AppHeader component, supports window minimize, maximize/restore and close operations |
| AVX Instruction Set Detection | Added AVX instruction set detection, pops up reminder when not supported (cloud desktop scenario) |
Bug Fixes
| No. | Description |
|---|---|
| 1 | Fixed issue where jumping to browser from link in markdown file |
| 2 | Removed logic that automatically hides after minimizing window |
| 3 | Disabled "template project" functionality |
| 4 | Can set voice and check updates normally without login |
v3.0.1
Release Date: April 27, 2026
Release Note
This version is SDW AIKO V3.0.1, adding automatic version update, network proxy configuration and context progress bar and other functions.
New Features Added
| Feature | Description |
|---|---|
| Automatic Version Update | Supports automatic version update detection, added update dialog |
| Network Proxy Configuration | Supports configuring network proxy in settings, unified manages HTTP requests |
| Context Progress Bar | Added context usage progress bar, displays context usage in real-time |
Bug Fixes
| No. | Description |
|---|---|
| 1 | Fixed packaging failure on Mac |
v3.0.0
Release Date: April 22, 2026
Release Note
This is SDW AIKO V3.0.0.
New Features Added
| Feature | Description |
|---|---|
| Project Grouping | Projects grouping supported |
| Mermaid Diagram Rendering | Artifact popup added Mermaid diagram rendering support |
| @mention Feature | Implemented mention function and style optimization in chat message bubbles |
| Custom Query | Allows adding and refreshing custom queries |
Changed Features
| Feature | Description |
|---|---|
| Packaging Optimization | Optimized packaging process, no file changes generated |
| Multi-language Task Progress Bar | Task progress bar supports multi-language |
| Progress Bar Style Optimization | Optimized progress bar style, fixed ghost deletion issue causing delete error on Windows |
| Git Detection Guidance | Added Git detection on startup, guides users to download |
Bug Fixes
| No. | Problem Description | Solution | Related Case |
|---|---|---|---|
| 1 | Operation: 1. New dialog, input task, AI generates artifact, click artifact. Expectation: Click artifact can download. Actual: Click artifact prompts "associated project" and no entry to operate. | Added "Download" button to artifact preview | TC-CHAT011 |
| 2 | Operation: In settings, import large compressed package (600M+). Expectation: Can import normally, should have feedback during long import. Actual: No feedback after click import, cannot confirm if import completed successfully. | Optimized resource package import process and progress feedback | TC-SET019 |
| 3 | Operation: After upgrading software version, click Marketplace, enter skills and knowledge base interface, expect interface loads normally, icons display correctly. Actual: Icon content missing after upgrade. | Fixed missing marketplace icon content after version upgrade | TC-Inst001 |
| 4 | Operation: After upgrading software, uninstall tool chain in Marketplace, check Marketplace status shows still installed. | Fixed Marketplace status not updating in real-time after uninstall | TC-SET046 |
v2.9.8
Release Date: April 17, 2026
Release Note
SDW AIKO completes first official release. This release covers the following product features, see Product Feature List for details.
Release Notes
This version is SDW AIKO V2.9.8, paired with latest "Product Feature List" and "User Manual", completes function iteration and architecture optimization, software functions, checklist and manual content remain consistent.
Feature Highlights
| Module | Features |
|---|---|
| Left Sidebar | Supports quick switching between New chat / Search / Marketplace / Chats / Projects pages |
| Marketplace | Six-column resource management and one-click access: Featured / Skills / Knowledge Base / Detectors / MCP / Toolchains |
| Historical Conversation Management | Display grouped by time, supports playback control (pause / speed / jump / exit) |
| Chat Interaction Management | Multiple parallel dialogs, Skill enhanced response, file download |
| Projects Management | Project creation, viewing and deletion |
| Settings Configuration | Comprehensive configuration of language / theme / provider / model / data directory / resource package |
| Resource Management | Add / edit / delete / enable for Skills, Knowledge Base, Detectors, Toolchains, MCP |
| MCP Management | Supports STDIO / SSE / HTTP / WS multi-protocol access and connectivity testing |
| Installation & Engine | Wizard-based installation deployment, engine initialization startup |
Bug Fixes
| No. | Problem Description | Solution | Related Case |
|---|---|---|---|
| 1 | Operation: In new dialog asking for Tetris game code, dialog asks what language user wants. Expectation: User can choose in dialog or query info box. Actual: Only allow/deny two choices. | Fixed issue where only allow/deny choices available during conversation dialog, added interaction options | |
| 2 | Operation: In new dialog asking for Tetris game code, after entire dialog completed and output files generated. Expectation: Can see timestamp for each step completed. Actual: No timestamp display. | Added timestamp display for each step completed | |
| 3 | Operation: In new dialog create another task, switch window between two tasks. Expectation: Each task runs independently, does not affect each other. Actual: One dialog has another task's reply content overlapping. | Fixed message cross-talk issue when multiple dialogs running in parallel | |
| 4 | Operation: After configuring API KEY in Settings, you need to restart software before application can take effect. | Fixed issue requires restart after Settings configuration API KEY | |
| 5 | Operation: Create new Project, click new Project card. Expectation: Displays project files information normally. Actual: Automatically jumps to NEW CHAT dialog. | Fixed automatic jump to NEW CHAT dialog after clicking new Project card | |
| 6 | Operation: Switch Project, expects previous history conversation not overwritten. Actual: Switching Project overwrites previous history conversation. | Fixed issue where switching Project overwrites history conversation | |
| 7 | Operation: Install new version package. Expect: Can be used normally after installation, no error message. | Fixed Engine fail to start error after installation | |
| 8 | Operation: Waiting for AI reply when switching dialogs, other dialogs cannot input. | Fixed issue where other dialogs cannot input while waiting for AI reply | |
| 9 | Operation: After adding Skill, input area cursor has abnormal display. | Fixed cursor display abnormality after adding Skill to dialog | |
| 10 | Operation: Search not working in Customize-Skills interface. | Fixed search function invalid issue in Customize-Skills interface | |
| 11 | Operation: After adding Skill to dialog, cannot recognize configured Skill. | Fixed issue where dialog cannot recognize installed Skill | |
| 12 | Operation: After adding Skill to dialog, Skill disappears from dialog while waiting for AI reply. | Fixed issue where added Skill disappears from dialog while waiting for AI reply | |
| 13 | Operation: View Detectors in Customize-Detectors interface, expects information fully displayed. Actual: Information stage incomplete. | Fixed incomplete information display issue in Detectors interface | |
| 14 | Operation: New Chat dialog defaults to software installation path as working directory. | Removed working directory entry from dialog above new chat | |
| 15 | Operation: Settings page full selection input box causes crash. | Fixed crash when selecting all text in settings page input box | |
| 16 | Operation: After adding Skill to Customize-Skills, cannot find it after successful addition. | Fixed issue where added Skill cannot found after successful addition to Customize-Skills | |
| 17 | Operation: Invalid API-Key, no prompt automatically detected. | Added automatic API-Key validity detection and error prompt | |
| 18 | Operation: Enter Project dialog, create custom Project, click project card to enter conversation dialog. Actual: Two terminal interfaces appear then disappear. | Fixed issue where two terminal interfaces appear when entering Project conversation | |
| 19 | Operation: After entering Project dialog, File information bar does not have generated file after dialogue generation. | Fixed issue where File information bar has no generated file after dialogue generation | |
| 20 | Operation: 1. New dialog, send 3 rounds conversation waiting for AI complete reply; 2. Click replay the task; 3. Check replay dialogue content. Expectation: 1. Replay should completely show all 3 rounds conversation; 2. No missing third round. Actual: Third and subsequent content missing. | Fixed third and subsequent content missing during conversation replay | |
| 21 | Operation: After setting data directory base path in personal settings, restart software expects correct file structure under this path. Actual: No related files. | Fixed issue where no related files after restart when setting data directory base path | |
| 22 | Operation: Uninstall Skill does not work on market Skill detail page. | Fixed uninstall button not working on market Skill detail page | TC-SET046 |
| 23 | Operation: After downloading multiple Skills, some Skills cannot display completely. | Fixed issue where some Skills cannot display completely after downloading | |
| 24 | Operation: In dialog, after adding multiple Skills, some Skills cannot display completely. | Fixed issue where some Skills cannot display completely after adding to dialog |