What Quorum is, what it costs, what it needs, and where it fits next to the other multi-agent frameworks.
Quorum is a multi-agent AI framework that runs an autonomous organization of AI agents on your own machine. You define the mission, budget, and roles in one Markdown file; Quorum plans the work, hires the specialists it needs, executes tasks in dependency order, reviews the output, and keeps going until the mission is done.
It is a runtime, not a library: you point it at a directory instead of assembling agents in code.
Quorum is free. There is no licence fee, no subscription, and no paid tier. The only thing you pay for is the AI provider you point it at — your Claude, OpenAI, or other model usage, billed directly by that provider. Quorum tracks every call in a local cost ledger so you can see the total.
A machine running macOS, Linux, or Windows, an empty directory, and a brain — an agent CLI such as Claude Code or Codex, an OpenAI-compatible server, or a local model. Quorum ships as a standalone binary with no runtime dependencies, so nothing else needs installing.
Run npx @meffecta/quorum init in an empty directory to scaffold the starter QUORUM.md.
Those are libraries you write agent workflows with. Quorum is a runtime you give a mission to. It decomposes goals itself, hires roles the team is missing, reviews work independently of whoever produced it, halts on budget caps, and keeps running on triggers instead of finishing after one graph executes.
If you want to program the control flow, a library is the better fit. If you want an organization that plans and staffs itself, Quorum does that part for you.
Any of several brains, chosen per agent: Claude Code, Codex, other agent CLIs, an OpenAI-compatible HTTP endpoint, or a local model. Junior roles can run a cheaper model while leads run a stronger one, and a fallback chain fails over automatically when a provider is unavailable.
See The brain — providers for the full list and per-agent configuration.
The organization runs entirely on your machine. Your mission, your QUORUM.md, your files, and your agents’ output stay local. The only outbound traffic is to the AI provider you configure and to services you explicitly grant, plus anonymous usage analytics you can turn off with one command.
See Analytics & privacy and the privacy policy.
Set daily, monthly, and lifetime caps in the QUORUM.md file, with environment variables, or with the --budget flag. Every model call is priced and recorded in a ledger, and reaching a cap halts the cycle rather than warning you. You can also stop everything immediately, mid-task.
See Cost & budget.
Yes, within limits you set. Through MCP servers and command-line tools you grant it, an org can send email, deploy a service, query an API, record leads, or bill a customer. Each capability is allow, ask, or deny — "ask" blocks the action and sends you an approval request.
See Acting in the world and Tools & permissions.
No. Quorum is proprietary software, free to install and use under its terms, but the source is not published and there is no public repository. You install it as a binary with npx or a direct download, and it runs locally on your own machine.
The package is published publicly as @meffecta/quorum on npm, and the licence terms are in the terms of service.
Quorum is young software and it is honest about that: it is provided as-is, without warranty. Autonomous agents make mistakes, so run it with budget caps, keep risky capabilities on "ask", and give it a directory whose contents you are willing to have changed.
Quorum is published by Meffecta AB, a software company registered in Sweden. It began as a side project exploring how far an AI organization can run without a human in the loop, and it is developed and released free of charge rather than sold.
More on the about page.