Autonomous AI Agency logoAutonomous AI Agency

An AI company that runs itself, on your servers

Paste a website URL and get a scored SEO, GEO and security audit. Then a CEO agent and a team of specialist agents plan the work, write the code, test it and open pull requests. The only thing they ask you for is the merge.

Run it in one command View the source
SAM, the agency's mascot, answering 'Brief me' with the CEO loop's live status and queuing the top roadmap item
SAM runs the agency from any screen. Anything risky still waits for you.

What it does

Self-hosted, MIT licensed, and OpenAI-compatible. Your code, customer data and prompts stay on infrastructure you control.

Reads your business

Scans a site's stack and runs a deterministic audit across SEO, GEO, AIO and security, with a 0–100 score and every finding priced.

Staffs itself

Provisions the specialist agents that business needs, coordinated by a CEO agent with a C-suite advisory board.

Checks its own work

Every change goes plan → execute → verify. It compiles and runs the tests before accepting a step; an LLM grading its own diff doesn't count.

Fixes itself and learns

Dashboard errors become fix tasks. Failures become lessons the planner reads next time, and repeated failures are filed as improvements.

Stays on a leash

Approval gates, a kill switch, per-agent daily spend caps, a policy engine and an audit trail. Nothing merges or deploys without you.

Free models first

Defaults to free NVIDIA NIM models and fails over across Cerebras, Groq and local Ollama, so the agency keeps working when a provider is down.

How it works

The same loop for every piece of work, whether it starts from your goal, an audit finding or an error on the dashboard.

  1. You give it a goal or a URL. Type it, or tell SAM.
  2. The CEO agent plans. It breaks the goal into tasks and assigns specialists, prioritised by value and urgency.
  3. Specialists execute in an isolated workspace. Each change is a small step with the files it touches named up front.
  4. A verifier checks every step. Code must compile and tests must pass before a step is kept; the PR carries the evidence.
  5. You approve. A pull request waits for your review. Deploys, deletes, payments and messages always ask first.

Don't trust it, check the proof

Most agent projects show a demo video. This one publishes its own output, including where it falls short.

Built by its own agents

A large share of this repository's merged pull requests were opened by AI agent sessions, each verifiable from its branch name. See the count and how to check it.

43.1 / 100

The audit engine run against this project's own docs page, published unedited with every finding. Read the self-audit.

Open source

MIT licensed. Every gate, prompt and test is in the repository, with a public changelog of what changed and why. Changelog.

Run it in one command

No cloud account, no database server, no Docker. SQLite storage, the dashboard and the 24×7 loops run in one process.

git clone https://github.com/strikersam/autonomous-ai-agency.git
cd autonomous-ai-agency
./scripts/quickstart.sh          # or: make up

Open http://localhost:8001/ and sign in with the credentials the script prints. Requires Python 3.11+ and Node 20+. A free NVIDIA NIM key gives top-tier output with no GPU. Setup guide.

Who it's for

  • Teams paying per seat for AI dev and ops tools who want the same leverage on infrastructure they control.
  • Anyone who needs agents that can write to production without choosing between no gates and a human on every step.
  • People evaluating autonomous-agent platforms who want one that runs on its own codebase, in public.
  • Organisations that care where their code, customer data and prompts go.

If you only want help writing code in one repository, a coding assistant such as Claude Code, Cursor or Aider is a better fit.

Frequently asked questions

What is Autonomous AI Agency?

A self-hosted, MIT-licensed, OpenAI-compatible AI proxy and multi-agent platform. A CEO agent coordinates specialist agents that plan, build, test and open pull requests for your approval.

Does it merge or deploy code on its own?

No. Every change goes plan, execute, verify, then waits as a pull request for your review. Deploys, deletes, payments and messages always ask first.

What does the website audit check?

A deterministic audit across SEO, GEO, AIO and security, with a 0-100 score and every finding priced. You can paste any website URL to run it.

Which AI models does it use?

It defaults to free NVIDIA NIM models and fails over across Cerebras, Groq and local Ollama, so the agency keeps working when a provider is down.

How do I run it myself?

Clone the repository and run ./scripts/quickstart.sh. It needs Python 3.11+ and Node 20+, uses SQLite storage, and needs no cloud account or Docker.

Is it free?

Yes. It is open source under the MIT licence and free to self-host.