New features:
- Removed the menai codebase and tools and put them in a sibling git repo.
- Menai is now installed from PyPI (
pip install menai) instead of requiring a local clone of the sibling repository. Pre-built wheels include the C VM for all supported platforms, sofetch-menai-vm.pyis no longer needed. - The filesystem AI tool's
search_filesoperation now supports wildcard matches. - Added a recent-mindspaces list to the sidebar header button. Clicking the mindspace name now shows a popup menu with recently used mindspaces for quick switching, plus "Open Mindspace..." and "New Mindspace..." actions.
- Added an "Open Recent Mindspace" submenu to the Mindspace menu in the main menu bar, mirroring the sidebar's recent-mindspaces quick-switch feature.
- Added Kimi K3 support to Ollama cloud and removed Kimi K2.6 and Kimi K2.7.
- Added Claude Opus 5, removed Claude Opus 4.8. and Claude Sonnet 4.6.
- Added a conversation tab "prompt mini-map" feature to make navigating to user prompts easier.
- Added paren matching in the editor to make it easier to understand and debug code (especially s-expr languages).
- Added GLM 5.3 and removed GLM 5.1 for the Zai backend.
- Added Grok 4.6 and removed Grok 4.3 for the X.ai backend.
- Added a
USR1signal handler for MacOS/Linux that will terminate the app and log the current backtrace status to a new logfile.
Bug fixes:
- Fixed a problem with main menu colours on Windows/Linux.
- Fixed the diff application tools (filesystem and editor) so that the "\ No newline at end of file" marker is honoured, allowing diffs to add or remove a trailing newline. Previously the marker was discarded, causing silent no-ops or garbled editor buffers.
- Fixed a problem with
find_filesand wildcard access outside the current mindspace. - Fixed a problem with parsing multi-line bold/italics in markdown processing.
- Resolved a minor label problem with Ollama model updates.
- Resolved problems with the mindspace settings dialog where text would not correctly update on a language change.
- Instantiate a new Menai VM per AI tool use to avoid threading race conditions.
- Improve the handling of detecting that Menai help has been read in a conversation. Ensure this persists across restarts, across forking, and correctly acts when messages are deleted.
- Updating existing conversation models did not honour the provider correctly.
- Fixed a usage tab problem when more than 4 models from the same provider were used. Updated scrolling of the tab to make it more consistent with other tab types.