you-agent-factory

you-agent-factory is an AI agent factory. It orchestrates AI agents for you so you can do more work without doing everything manually.

Why?

Leverage.

With you-agent-factory, you codify your process into a workflow with different AGENTs.md and run them as wrappers around OpenAI codex.

For example:

• dispatch 10 agents to run independently in separate work trees
• have one agent loop through a series of tasks, and then have a reviewer review the output and retrigger the loop if it failed
• tell the agents a series of plans, and run them in dependency order
• have a cron setup to autonomously look at git tasks or whatever and submit tasks that go through a write/review cycle loop

Install

1. install codex npm i -g @openai/codex
2. install on macOS/Linux: curl -fsSL https://github.com/portpowered/infinite-you/releases/latest/download/install.sh | sh
3. install on Windows PowerShell: irm https://github.com/portpowered/infinite-you/releases/latest/download/install.ps1 | iex
4. go cd your-project-directory
5. run you
6. submit a work task on the website interface, like "go write a report on my codebase at TEST.md",
7. wait till complete
8. finished

claude variant

you init --executor claude --dir my-factory
you docs workstation

Inspect Runtime State

With the service running, ask the live API which factory is currently active:

you factory query
you factory query --port 7437 --json

factory query reads the running service's current-factory API. It does not infer the answer from local factory.json files, so the output reflects the active default-root runtime or the currently activated named factory on that server.

Submit Work From The CLI

Use submit for single-work API submission when the factory service is already running:

you submit --name "driver-incident-review" --work-type-name task --payload request.md

--name, --work-type-name, and --payload are required for unary CLI submission.

Example

Here's an example of you-agent-factory dispatching roughly 5-10 agents.

How It Works

The default no-argument starter flow looks like below: you give it a task, it spawns a basic agent CLI run and does stuff.

flowchart LR
   classDef place fill:#000,stroke:#333,color:#fff,stroke-width:2px
   classDef transition fill:#333,stroke:#333,color:#fff,rx:0,ry:0

   P0((task:init)):::place
   P1((task:complete)):::place
   P2((task:failed)):::place

   T0[process]:::transition

   P0 --> T0
   T0 --> P1
   T0 -.->|on failure| P2

Loading

Customization

See authoring-factories for the full configuration guide. you-agent-factory lets you customize your flow however you want.

The overall system of how you-agent-factory works is relatively simple.

1. You have work.
2. Work goes to workstations where the work gets worked on by workers (agents, or just shell scripts)
3. When the workstations complete the, work is converted to other work.
4. you-agent-factory stops when no work remains.

For packaged terminal docs, use the fixed you docs <topic> topics: config, workstation, workers, resources, batch-work, and templates. Those packaged pages point back to the maintained canonical docs in docs/reference/work.md, docs/reference/config.md, docs/reference/workstations.md, docs/reference/workers.md, docs/reference/resources.md, docs/reference/batch-inputs.md, and docs/reference/templates.md.

Shipped example factories

Drag the images from the examples/factories directory into the web interface's flow graph, and it'll load the factory for you.

Doc reviewer Write and review workflow.	you-agent-factory Meta factory that runs the factory.	Ralph Iterative plan, code, and review loop.
Timer Cron-based factory trigger.	Worktree Spawns work in a git worktree.	Writer reviewer Iterative loop for writing docs.

References

• Analysis on current projects