Agent product

Chief

A personal agent that notices, proposes, and acts only after you approve.

How it works
  1. Notice
  2. Propose
  3. Approve
  4. Act

The problem

Assistants either do nothing until asked or act without asking. Neither is how a good chief of staff works.

How it works

Chief runs a quiet heartbeat over your week. When it finds something worth your attention, like three short nights in a row after packed meeting days with another packed day coming, it proposes a specific fix and waits.

You can also just talk to it from your phone: “What’s in my inbox?”, “Reply to Sam and say Friday works”. A large Stage screen shows its work as it happens: the inbox filling in, the calendar jumping to the day it will change, the sources it read.

Built for the Cline × Composio × Parallel hackathon during Open Source AI Week in San Francisco, October 2026. Released under the MIT license.

03 /

Key decisions

  • Every write waits for a tap

    Calendar edits and sent emails are proposals until you press Approve. Read-only work, like an inbox brief, runs freely.

  • Skills are single files

    A new skill is one file of about twenty lines, so the agent grows without a framework.

  • Suspicious mail is ignored

    Email that reads like instructions to an assistant is flagged and never acted on.

04 /

What’s next

Everything runs end to end on seeded mock data today. Next is running the real-service paths (calendar, Gmail, search, speech-to-text) against live accounts.