// self-hosted · any ai · your machines, your keys

Your backlog, built overnight.

The Farm runs the AI of your choice on machines you own. The AI writes the code; your machine tests it and pushes it. Your git credentials never reach the AI.

See pricing →/who-holds-the-keys

runs onyour hardware

git tokenyour machine only

aiclaude · codex · gemini · +

The AI writes. Your machine holds the keys.

Every item runs on a machine you own, in a working copy of its own. The AI session can edit and commit there — nothing more. Testing, merging and pushing to your repository are done afterwards by the farm itself, with a token only the machine can read.

1 · your repository

An item is picked

The lane on your machine pulls the backlog and cuts a fresh working copy for the item.

git pull · worktree add

2 · the ai sessionsandboxed

The AI writes and commits

  • can
  • edit its working copy
  • run your tests
  • commit
  • cannot read
  • git / gitlab token
  • ~/.ssh · .netrc
  • .env · signing keys

no push rights · no deploy rights

3 · your machine

The machine lands it

After the session ends, the farm re-runs your gate, rebases onto main and pushes — with the one credential that never left the machine.

gate · rebase · git push

Code stays home

Builds, tests and working copies live on your Linux, Mac and Windows machines. We never host your code.

Keys stay put

Git tokens and AI keys sit in owner-only files on the host. Never in git, a log or the hub — the hub can hand a key over, never keep it.

Only green lands

A change the gate fails never reaches main. It waits on a branch for you, with the session that made it.

Bring the AI you already pay for.

Sign in with the subscription you have, or paste an API key. Choose the AI per machine, per project, even per item — and switch from the panel without touching a lane.

REFERENCE

Claude Code

Anthropic. The farm’s full feature set: remote control from your phone, usage metering, per-file access rules.

subscription or api key

openai

Codex CLI

Runs in its own workspace sandbox. Sign in with ChatGPT or an OpenAI key.

chatgpt sign-in or api key

google

Gemini CLI

Sign in with your Google account or a Gemini API key.

google sign-in or api key

open · local

OpenCode

Grok, Mistral, DeepSeek, OpenRouter — or a model on your own GPU through Ollama, LM Studio, llama.cpp or vLLM.

api key, or nothing at all

one farm · claude for the app · codex for the api · a local model for the docs

aider · goose · qwen code — coming

Frugal by design[ every item on a budget ]

Your tokens are your money, so the farm spends them carefully. Routine items run on a lighter model and step up only when a retry needs it. Every session has a context and turn budget. A small scout maps the code before the main session reads anything, and big files and images are handled in short sessions of their own. Nothing sits polling while it waits. And a usage governor pauses the lanes before your weekly allowance runs out.

lighter model firsta budget per sessiona scout reads firstnothing polls