Pi agent support
July 7, 2026 ยท View on GitHub
Minutes supports Mario Zechner's pi coding agent as an opt-in local agent CLI.
What is wired
- Desktop settings now recognize
pias a well-knownagent_command. - Desktop Recall can launch
assistant.agent = "pi"as an interactive Pi session. engine = "agent"can runpifor summarization.- Agent routing evals can be run with
--agent pi. - Pi can use the existing
.agents/skills/minutes/skill mirror; Pi's package manager auto-discovers ancestor.agents/skillsdirectories, so a separate.pi/skillstree would duplicate the same skill names.
Recall assistant config
[assistant]
agent = "pi"
agent_args = []
For the interactive Recall panel, Minutes launches pi in the assistant
workspace and passes agent_args through unchanged, after dropping only
approval-bypass flags that belong to other agents.
Pi owns provider auth and model selection. Use Pi's interactive /login and
/model flows, or explicit Pi flags such as --model <provider/model>, to
choose the model. If a GitHub Copilot model such as github-copilot/gpt-5.5
fails with "Personal Access Tokens are not supported for this endpoint", refresh
Pi's GitHub Copilot login with /login; do not paste a GitHub PAT into Minutes.
Summarization config
[summarization]
engine = "agent"
agent_command = "pi"
Minutes runs Pi with:
pi --no-session --no-tools --no-extensions --no-skills --no-prompt-templates --no-context-files -p @<private-prompt-file>
Configure provider/model defaults in Pi itself; Minutes does not currently pass
extra [summarization] CLI flags through to the pipeline invocation.
That invocation is intentionally narrow: no saved session, no tool access, no automatic context files, and transcript prompt content passed through a private temp file rather than the command line.
Inflection Pi boundary
Inflection's Pi is a different thing from the pi coding-agent CLI. Inflection-3 Pi is tuned for emotional intelligence and customer-support style chat, so it may be useful later for opt-in tone coaching or reflection features. It should not be a default transcript processor because meeting transcripts often include personal data, and Inflection's developer terms currently tell API users not to send personal information or other regulated data.