Changelog
Agotalk Changelog
All notable changes to the Agotalk project will be documented in this file.
[0.1.55] - 2026-07-07
- Voice replies start faster with streaming TTS, no more unexpected cutoffs, and better language detection.
- Local model startup is faster with quicker first-token response.
- Browser tool results are now auto-cleaned up instead of accumulating.
[0.1.54] - 2026-07-06
- Built-in GNU coreutils toolbox for stronger script execution.
- Model runtime errors now show clear causes instead of vague failures.
- Fixed page scroll jump-back on Stop and message bottom being covered by status bar.
[0.1.53] - 2026-07-06
- Better browser privacy protection and lower website fingerprinting sensitivity.
- More accurate browser content extraction; internal tools are now partition-isolated from external browsing.
- Work steps are displayed more clearly with one-click copy-all support.
[0.1.52] - 2026-07-04
- Model switching now shows progress and a confirmation window; empty threads and failure states are clearer.
- Browser tools support DOM-reference actions and geometry awareness; the blank window opened from the menu is fixed.
- Long tasks now actually stop when stuck; pressing Stop now fully terminates the child process tree.
[0.1.51] - 2026-07-02
- Image pending user projection is now deduplicated; the same image no longer queues multiple sends.
[0.1.50] - 2026-07-02
- Added browser window host support and browser actions.
- Left sidebar workspace list can now be collapsed and expanded; first frame loading is more stable.
[0.1.49] - 2026-07-01
- Browser window zoom and drag interactions are more stable.
[0.1.48] - 2026-07-01
- Added in-app browser support; open web pages directly in the workspace.
- Role package import and edge-case handling are more stable.
[0.1.47] - 2026-06-30
- Voice playback speed now auto-adjusts by language; manual scrolling during streaming is no longer interrupted.
[0.1.46] - 2026-06-30
- Voice replies are more natural and stable; overall improvements to speech recognition and TTS playback quality.
[0.1.45] - 2026-06-29
- Browser window now supports adjustable size, hidden address bar, and custom titles.
- Live Local avatar animation is smoother, flicker is eliminated, and recording auto-stops when the call ends.
[0.1.44] - 2026-06-27
- Role skill configuration metadata is more complete; handoff UX is much smoother.
[0.1.43] - 2026-06-27
- Fixed first-batch UI collaboration issues; shell runner state and error output are more stable.
[0.1.42] - 2026-06-24
- Copying system output messages now shows a "Copied" feedback toast.
[0.1.41] - 2026-06-24
- Right sidebar redesigned; work steps and reasoning are displayed more clearly.
- Finer interface details, smoother interactions, and more stable text streaming.
[0.1.40] - 2026-06-23
- Download links on the website now point directly at the latest installer; no manual version selection needed.
[0.1.39] - 2026-06-21
- Settings page layout is cleaner and calendar rendering is more stable.
[0.1.38] - 2026-06-20
- AI tool call failures now surface explicit errors instead of failing silently.
[0.1.37] - 2026-06-20
- Fixed MiniMax provider description labels and history display.
- Clearer boundaries for image-related tools, and more stable internal state after stopping generation.
[0.1.36] - 2026-06-19
- Stopping generation now shows immediate feedback with the latest reply ready to replay.
[0.1.35] - 2026-06-19
- Improved desktop workflow polish and status feedback.
- Fixed original user messages disappearing after @role replies.
- Improved stability around model context, tool authorization, and config binding.
[0.1.34] - 2026-06-18
- Live Local voice permission restoration is more reliable, and calls no longer drop on transcription failure.
[0.1.33] - 2026-06-18
- Live Local calls are more stable, avatar animation no longer flickers, and pausing the call no longer disconnects.
[0.1.32] - 2026-06-18
- Voice reply first audio is faster, with sentence-boundary chunking for immediate playback.
[0.1.31] - 2026-06-17
- Update download and installation is smoother; click to download new versions.
[0.1.30] - 2026-06-17
- First-launch API recommendations are now simplified to DeepSeek.
[0.1.29] - 2026-06-17
- "Check for updates" can find and download new versions again.
[0.1.28] - 2026-06-17
- Auto-update package signature verification is more reliable.
[0.1.27] - 2026-06-17
- Downloads and auto-updates now resolve to the same version more reliably.
[0.1.26] - 2026-06-17
- Installation and auto-update verification are more reliable.
[0.1.25] - 2026-06-17
- Installation packages open and verify more reliably.
- Fixed issues affecting the installation experience.
[0.1.24] - 2026-06-15
- Conversation now shows process rows; run steps can be collapsed and expanded.
- Returning to a past conversation shows a session recap.
- Finer interface details; scroll-follow and main window refresh are more reliable.
[0.1.23] - 2026-06-13
- Videos and images in chat now show thumbnails with more stable loading.
- Long conversations auto-compact context after sending a message.
- Automation tasks have an entry point in the conversation, with a clearer execution process.
[0.1.22] - 2026-06-11
- Runner injection and interaction detection are more reliable with fewer false positives.
- Bug fixes improving overall smoothness.
[0.1.21] - 2026-06-10
- Streaming text during AI replies renders more cleanly, no more unclosed Markdown markers.
- Bug fixes improving overall smoothness.
[0.1.20] - 2026-06-10
- Message input now uses an async command queue; sending and voice no longer block each other.
- Manual interjection messages get a distinct visual style in the conversation.
- Update check and model caching are more reliable.
[0.1.19] - 2026-06-09
- Bug fixes improving overall smoothness.
[0.1.18] - 2026-06-09
- Agotalk browser supports standalone windows instead of tabs.
- Installer size reduced by 34MB.
[0.1.17] - 2026-06-09
- Voice crash recovery is faster; load failures no longer hang.
- Conversation list no longer flickers on refresh.
- Tool pills and action hints are more compact.
[0.1.16] - 2026-06-09
- Underlying path resolution is more reliable; voice and network connections are more stable.
[0.1.15] - 2026-06-09
- Tool execution panel supports live output with smoother animation.
- The "check for updates" feature on the About page works again.
- Chat experience is more stable; file opening and terminal output are more reliable.
[0.1.14] - 2026-06-08
- Model picker only shows available items; remote models grouped by provider.
- Runner supports interaction detection and auto-reply.
- Grid mode is more stable with finer layout details.
[0.1.13] - 2026-06-08
- Role library launched: pick roles from candidates and add them to the workspace.
- Package installation entries removed from settings.
- Live voice opens in-app; assistant tool results show in the conversation.
[0.1.12] - 2026-06-07
- Voice reply playback speed slowed slightly for a more natural feel.
[0.1.11] - 2026-06-07
- Keep-alive connections are no longer auto-disconnected.
- Maintenance actions on the About page are clearer.
- Live Local page labels are more concise.
[0.1.10] - 2026-06-06
- "Check for updates" can now download new versions properly.
[0.1.9] - 2026-06-06
- Voice reply playback speed slowed.
- Runtime status display is clearer.
[0.1.8] - 2026-06-06
- Desktop voice playback path is more stable.
- After stopping speech, recording is auto-trimmed before recognition.
- Reply voice playback is smoother.
[0.1.7] - 2026-06-05
- Overall stability improvements.
- Test build install and startup details resolved.
- Live Local connections are smoother on weak networks.
- Live Local voice reply transcoding now uses the bundled ffmpeg, avoiding missing system ffmpeg in packaged builds.
[0.1.6] - 2026-06-05
- Overall stability improvements.
- Local runtime install, update, and cache maintenance are smoother.
- Live Local voice page adds local avatar status experience, with reinforced connection, VAD, and error feedback.
- Voice VAD switched to a shared sherpa/Silero mainline; a clean install auto-prepares bundled VAD material.
- Test build startup and thread read smoke issues resolved; pre-release in-place build / CDP / VAD smoke passes.
[0.1.5] - 2026-06-05
- Voice start and stop recognition is more accurate.
- Local network encryption is more stable.
- Clicking "check for updates" no longer false-negatives.
[0.1.4] - 2026-06-04
- Voice / settings full-link rewrite.
- Local OOXML parsing without LibreOffice.
- Feishu final replies rendered as cards.
[0.1.3] - 2026-06-03
Added
- Voice input/playback overhaul: full link rewrite with Kokoro speaker mapping restored; auto-speak and manual playback coexist cleanly, with proper preemption cleanup on thread switch.
- Settings shell unification: standardized target components, models page, field patterns, and control baseline; new role library tab.
- Workspace autopilot: toggle and dream-records loop; external hook + webhook adapter for runtime automation.
- Local OOXML parsing: office documents (Word/Excel/PowerPoint) parsed without LibreOffice, with bundled soffice fallback.
- Feishu rendering: final replies rendered as Markdown cards; inbound trigger mechanics hidden in transcript.
Changed
- Chat attachment polish: unified tool-window open intent, direct file icon rendering, tightened thumbnail frame.
- DeepSeek prefix-cache diagnostics: hit ratio surfaced to runtime.
- Role library: builtin role packages shipped with drift adapter; archived role homes detached from bootstrap.
Fixed
- Thread ledger: raw input artifact references preserved end-to-end.
- Voice input status indicator added; scroll-follow button contrast adjusted and feedback polished.
- Viewport pinning: live updates no longer jump the thread view; reserved-space observer removed.
- Typecheck / runtime boundary debt cleared across multiple subsystems.
[0.1.2] - 2026-05-08
Added
- Smart Model Settings: Automatic parameter matching for both local and remote models, ensuring you always have the right tuning options.
- Persistent Custom Profiles: Your tweaks to temperature, tokens, and other settings are now saved automatically per-model.
Changed
- Compact Model Inspector: A new high-density, sleek design for model settings that fits more information into less space, perfectly matching the Agotalk aesthetic.
- Enhanced Local Model Control: Full tuning support for local models (like Gemma), bringing parity between local and cloud-based AI settings.
Fixed
- Settings Reliability: Improved the loading speed and stability of the settings shell.
- Language Support: Localized labels for all specialized model parameters to ensure a consistent experience.
[0.1.1] - 2026-05-05
Added
- Intelligent Context Management: Automatically compresses long conversations to keep them running smoothly without losing important project context.
- Project Memory System: Established a long-term architectural memory to ensure the system stays true to its design principles.
Changed
- Streamlined Memory Storage: Switched to a faster, more reliable file-based memory system that stays in sync with your workspace.
[0.1.0] - 2026-05-02
Added
- Multi-Tab Editor: Work on multiple files simultaneously within the same workspace window.
- Standalone File Access: Easily open and edit files outside your current workspace in dedicated windows.
Changed
- Clean Interface Overhaul: Refined the title bars and sidebars for a zero-distraction, "pro" editorial look.
[0.0.1] - 2026-04-14
Added
- Zero-Dependency Local AI: Run models locally on your Mac out of the box, with no complex setup or Python installation required.
- Smart Workspace Discovery: Automatically detects and initializes your projects for immediate work.
- Modern Pro UI: A production-ready interface inherited and optimized from professional productivity tools.