Pacenote v0.5.29

Pacenote's own work

• Choose the AI. Settings > Todo > Pacenote's own work: Claude Code, Codex or opencode. It runs headless on this Mac, with your own login. Setup's first step sets it to the AI you pick for Pace's answers; Pacenote set up before keeps Claude Code.
• Switch when it can't. On by default. When the AI chosen is at its usage limit, unavailable, signed out or without a Slack connector, Claude Code or Codex does the work in the same run. One at its limit rests until the limit resets, so the next piece of work goes to the other first, and Looks done waits while both rest.
• Codex reads Slack through ChatGPT's Slack app. Read only: only that app is on, its tools that change or delete are off, and its other tools that write are refused. Both connectors should read the same Slack workspace; a task is keyed by its message, so collecting with either never adds it twice.
• opencode, with a model of your own. Pick opencode and, if you like, one of the models in your opencode config, such as a local model or one you host yourself, through an OpenAI-compatible API. Looks done and Write it up run there, with no tools; it reads no Slack yet, so Collect now and drafts go to Claude Code or Codex.
• Who did it. Todo's line under the title, a drafted reply, a recap's write-up and a collection's run say which AI did the work, and why it switched. Settings shows each one: here and signed in, or resting until its limit resets, with a sign-in for one signed out.

Also

• Usage limits, as the CLIs say them now. "Usage limit reached", a model's own limit and Codex's usage_limit_reached count as a limit, a context limit does not, so Pace's replies about a limit are right too.
• Collect now uses CLAUDE_BIN, as drafts, Looks done and Write it up already did.
• Short choices sit side by side in Settings: where code tasks open, and which terminal.