Skip to content

MegooColab 0.20.0

Choose a tag to compare

@github-actions github-actions released this 13 Nov 16:37

MegooColab 0.20.0 lets each workflow stage run on its own model and effort, so an agent can design and plan on one model and build on another.

New

  • A model and effort for each workflow stage. An agent's settings (Agents → the agent) have a new Model by stage table under its Model and Effort, with a row each for Design, Plan, Build (steps and their fixes), Review (step, final and pull request reviews) and Finish. For example, set Design and Plan to Fable 5.2 and leave the rest "As above" to build on Opus.
    • Each row offers the models the agent's own program lists, with the effort levels each model has, or a typed name. A row that sets a model but no effort uses that model's own default effort. Rows with their own choice are outlined in violet.
    • Every run of a stage starts on its choice, also when it continues an earlier session: a session designed on Fable goes back to the agent's own model for building.
    • A task's Activity already says which model and effort each run used, so you can see the stages switch.
    • Chats, scheduled jobs, answers to other projects and Knowledge Base drafts keep using the agent's own Model and Effort.
    • Changing the agent's own model resets a stage that set only an effort the new model doesn't have, and says so. The same goes for a stage's own model and its effort.

Upgrading

Pull and restart as usual. The update adds a database column for the stage settings and one for what each run was started on; it is applied automatically when the server starts, and there's no new setting.

Update the runner on each machine to 0.9.0 to use stages with their own model. Until a machine's runner is updated, runs of a stage with its own model wait, and the agent shows "runner too old to choose a model per stage — update megoocolab-runner". Agents without stage settings, and stages left "As above", work with the older runner as before.