Describe the work. Read the plan. Put it on a schedule.

You build agents in Lunen by describing them — in writing, to an agent whose whole job is turning your brief into something reviewable. There are two kinds, and knowing which one you’re building is most of the design work.

Two kinds of agent

Task Agents do work without you. A Task Agent is a saved execution plan — ordered steps, named tools, explicit dependencies — that runs on a schedule, on demand, or when a webhook fires. Every run is recorded with its full execution log, and any tool the policy marks for approval pauses the run for a human.

Convo Agents work with you. A Convo Agent is a purpose-built chat partner — a synthesized system prompt plus the services you give it — that you and your teammates open when you need it. It runs as whoever is chatting with it, using their connections and their permissions.

A useful rule: if the trigger is a time or an event, build a Task Agent; if the trigger is a person with a question, build a Convo Agent. The longer version is in Designing reliable agents.

From brief to first run

1. Click New Task Agent or New Convo Agent in the sidebar.
2. Describe the job in plain language — goal, boundaries, what done looks like. The planning agent interviews you and drafts.
3. Read what it drafted: the plan (or the agent’s design) renders in the blueprint panel beside the chat as you go.
4. Refine by replying. The draft changes nothing until you save.
5. Save Agent — or Run Now for a Task Agent when you want to watch one run first.

Nothing your agent will need is assumed: the services it can call are the MCP servers your organization has enabled and you have connected, and every call it makes runs under the same policy as everyone else’s.