What is clodfarm?

clodfarm operates a long-running fleet of Claude Code agents inside a single Docker container. You assign it a goal or a list of tasks; workers pull from a queue, divide large jobs into sub-agents on isolated git branches, execute tests, and merge only after checks succeed. Every worker is a genuine Claude account measured against its real 5-hour and weekly quotas, so the system automatically throttles when you personally use Claude. Several machines and accounts can participate in one farm via a shared DynamoDB table and git repository, each account governed by its own budget. You converse with your agent from the Claude mobile app using Remote Control, observe all sub-agents in the web UI at localhost:8080, and can connect your local Claude Code through MCP. The farm also links once to Stripe, Google Ads, X, Blender, and Slack so every agent can call those services. It installs with one command on any Docker host, or deploys to AWS in about ten minutes with zero inbound ports. Licensed MIT.

Should you self-host clodfarm?

clodfarm scores 48/100 on selfhost.directory (how scoring works) and ranks #110 of 209 in AI Agents. It has 213 GitHub stars.

It shipped 5 stable releases in the past year. The latest, v1.13.0, came out on 11 October 2026. New versions land roughly every 2 days.

Setup is moderate: expect some configuration beyond a single command. An official Docker image is available. It is MIT-licensed and written mostly in Python.

Its closest alternative on selfhost.directory is Orca (91/100, 108 446 stars). It scores higher overall, so compare the two before you pick. See clodfarm vs Orca side by side.

Features

AWS serverless app deploymentOptional; requires AWS account and apps role deployment
Bots for other models
Inter-agent messaging
Live dashboards
Logged-in browser
MCP server for local Claude Code
Multi-seat / multi-box farm
Per-account budget governor
Remote Control from phone
Schedules (cron/interval/one-off)
Slack integration
Sub-agents with visible worktrees
Test-gated merges
Whiteboards
Zero-Downtime Upgrades

FAQ

Can it use other models, like free ones?

Yes, as bots: Claude Code on another model (OpenAI, xAI, Gemini, Groq, OpenRouter, Ollama, or any compatible gateway). They use no Claude usage, take only sub-agents sent to them, and are counted at their model's list price.

Does it work with an API key, Bedrock or Vertex?

API keys work with a daily dollar cap instead of subscription pacing. Bedrock and Vertex should work through Claude Code's own environment variables but aren't tested yet.

Is it safe to give agents a shell?

Agents run as an unprivileged user inside the container. Mount only what they may change, give git a deploy key for one repo, and consider FARM_PERMISSION_MODE=auto.

What does it cost?

On a subscription, nothing beyond your plan. Locally or on your own server it's free (single box needs no database). The AWS box is roughly $25/month for a t4g.medium plus cents of DynamoDB.

Will it lock me out of my own Claude?

The governor stops agents at 80% of your week and 85% of any 5-hour window, and those numbers include your own usage, so the farm backs off when you're working.

Is this allowed?

clodfarm drives the official Claude Code CLI within documented limits. On a subscription it's meant for your own projects on your own login; don't run it as a service for others, don't share logins, use Team/Enterprise seats for collaboration, and use API keys for commercial workloads.