Skip to content

Modus v0.1.12

Choose a tag to compare

View all tags
@github-actions github-actions released this 18 Oct 06:22
· 40 commits to main since this release
v0.1.12
ea240a6
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

What's changed

Local tool filter and REA on by default

  • The Local tool filter is now on by default, with no setup and no confirmation dialog. On supported machines, the local model (about 415 MB) downloads in the background the first time you open Modus after updating. The download is verified and resumes if interrupted. Until the model is ready, every tool is sent as before.
  • Background failures (offline, low disk space) don't show an error: the filter sends every tool and retries on its own. If you had turned the filter off before this release, it comes back on once.
  • To turn it off, use the switch in Settings › General. Turning it off cancels the download and stops the model process. Turning it back on doesn't ask for confirmation.
  • The filter is still unavailable on Intel Macs and on Windows without the Visual C++ runtime. On those machines it stays off with no warning.
  • REA was already on by default and no longer shows an approval card on each call. It still installs from a pinned lockfile, with a sha512 per package and install scripts disabled, and runs with your permissions, not in a sandbox. You can turn it off in Settings › MCP.

New define-goal skill

  • New built-in define-goal skill, adapted from openai/skills (Apache-2.0). It sets one concrete, verifiable goal for the chat, with the get_goal, create_goal and complete_goal tools. One goal per chat.
  • The tools only appear after the skill is used in the chat, and stay available until the chat ends. Before saying it's done, the agent checks the completion criteria and records the evidence.

Tool filter log

  • With the filter on, Modus logs one line per first-turn decision (filtered or sent everything, and why) and one line for each tools_expand call. The lines contain only numbers and fixed words: never message text, tool names, ids or paths.
  • The lines go to a local-models.log file in the app's data folder, at most about 2 MB:
    • Windows: %APPDATA%\@modus\desktop\logs\local-models.log
    • macOS: ~/Library/Application Support/@modus/desktop/logs/local-models.log
    • Linux: ~/.config/@modus/desktop/logs/local-models.log

Fixes

  • Fixed the "Run-scoped event history is unavailable because run boundaries overlap" error, which, once it appeared, came back on every message in the same chat. Each chat now runs one turn at a time: messages sent while a turn is finishing are queued and sent in order. Turns left open after the app closed mid-run are closed on the next message, so stuck chats start working again on their own.
  • The Settings sidebar footer shows just the app name and version again. The RERANK-03 credit paragraph was removed from there. The credit is still in the system About panel, and LICENSE and NOTICE still ship with the model files.

Licenses

  • Every third-party built-in skill now includes its original license in its own folder: the 15 Superpowers skills (obra/superpowers, MIT), reverse-engineer-anything (morluto/rea, MIT) and define-goal (openai/skills, Apache-2.0). All are listed in THIRD_PARTY_NOTICES.md.

Downloads

  • Windows: x64 installer.
  • macOS: Apple Silicon (arm64) and Intel (x64), DMG or ZIP.
  • Linux: x64 AppImage or Debian package.

Full changelog: v0.1.11...v0.1.12

Assets 17
Loading