Forge agents New in 0.8.5
You know what the job is, but not which agents would do it. Say it in your own words and attach your files. Captain plans a small team of standing agents, you approve it, and it recommends what should run on a schedule and what software is worth building.
In the kiosk: New agent → Forge agents. Close the dialog at any point: it carries on, and the bell brings it back.
Your request in your own words, the way you'd tell a new colleague, and up to 10 files of 25 MB each.
An agent opens your files, and the planner picks the fewest standing agents that cover the request, writes each one's job and may ask you questions. Rename, rewrite or leave out anything.
Each agent gets its way of working, its job and its teammates. The team shares a folder and one set of data tables, and they can ask each other.
Captain recommends what should run by itself and what software is worth building, with why and how you'd use each. Only what you keep gets made.
The real interface, with made-up people and data. A plan never exceeds the room your agent limit leaves.
Finish Schedules and software
A standing agent mostly runs on a schedule or for a linked system, with nobody watching. So Captain tells it its job before each scheduled turn, and before a linked system's first turn after it starts. A link also follows its agent when a restart moves it to another port.
What it doesn't do yet: it has run end to end with a stand-in model, CLI and app, but a real model's plan and a real worker building to the contract have yet to be tried together. A team's tables aren't listed in the Data library. A non-admin can't host a runnable app, so Connect then needs an admin or another host. Agents come from the archetype catalog.
Forge agents docs · Agent Forge, in the full layout, designs a team around a goal and spawns it in one click
Join the waitlist for Cohort 001, our first early-access group, and you'll get an email when you're in. Or run Captain on your own machine today.