Lunen documentation
Agents that do real work, inside the limits you set.
Describe the outcome, connect the tools, decide what needs a human. Start where you sit — each path runs from first setup through to the reference at the end of it.
Or jump toQuickstartAPI
Choose your path
Switchable any time from the headerSet the limits
Connect servers, decide what agents may read and what needs sign-off, then prove afterwards what happened.
- Quickstart4 min
- Policy & approvals6 min
- Audit log5 min
Ship the agent
Write the brief in plain language, read the plan it produces, then put it on a schedule.
- Describe an agent5 min
- Execution plans5 min
- Schedules & runs6 min
Wire it in
Drive Lunen from your own systems — REST for state, webhooks for everything that happens.
Three ideas worth reading first
Everything else in these docs is a detail of one of these.
A plan is a contract. Lunen never improvises against your systems. The brief becomes an ordered plan you can read and correct before anything runs, and the plan is what actually executes.
Policy is per tool, not per agent. Reads are usually safe to allow; writes should wait for a human until you have watched them run. That pair — allow or approve — is the whole model, and it governs scheduled and ad-hoc runs alike.
The log is the product. Every tool call, policy change, and approval lands as an append-only audit event with the actor attached, because “the agent did it” is not an answer anyone accepts.