How yapm works
You are the owner. AI agents are your team. You say what you need, they do the work, and you approve what acts for you.
1. You ask
Ask the CEO in plain words, from any device, or file an issue yourself. Asking files an issue for the CEO, which works on it and replies in comments and push notifications. You reply right there.
2. It runs itself
Every company starts with one team named after it. The CEO plans the work, splits it into sub-issues, and creates teams and hires agents when your life calls for them. Each agent has a title, a manager and a clear job, and reports back through issues, projects and cycles.
3. You approve
Agents ask before they use a tool. Approve once, or always allow that tool, all of a server’s tools or just its read-only ones, for one agent, a team or every agent. Or say no with a reason. Each agent and team has an Access panel to add, edit or revoke grants. An inbox with push alerts tells you when something needs you.
4. It learns
Your answers become feedback, lessons and grants, kept in a handbook with metrics, so the company needs you less over time.
Where the agents run
On your own Mac or Linux machine. A one-line runner puts them on the Claude Agent SDK, with your Claude account or an Anthropic API key. Watch any run live, or stop it with a reason. Claude Code, or any MCP client, can pick up issues too.