Choose your AI provider
Each role uses an AI agent with its own working directory — Claude Code by default, or Codex, opencode, a local model, and more. Agents run commands, write and execute code, and edit files to complete tasks.
Autonomous AI organizationsAn AI organization
An AI organization
that runs itself.
Give Quorum a mission and a budget in one file. It plans the work, hires the specialists it needs, and does it — then keeps operating: reacting to events, scheduling its own work, and reaching real services like email, hosting, and payments. You set the spend caps and approval gates, and steer it live from the browser.
$ npx @meffecta/quorum initWrites the org's starter QUORUM.md — edit it, then boot with npx @meffecta/quorum.
- ✓ Runs on Claude Code, Codex, a local model, and more
- ✓ Ships as a standalone binary — zero runtime deps
- ✓ One directory per org
org chart · live
starts with one CEO — hires specialists as needed
Runs continuously, reacting to events·Schedules its own work·Hires specialists as needed·Independent review of all work·Spending tracked and capped·Reaches real services via MCP
What is Quorum?
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's free to use — you pay only your own AI provider for the model calls. Everything runs on your machine, so your mission, your files, and your agents' work never pass through anyone else's servers. Where other multi-agent frameworks give you a library to assemble agents in code, Quorum is a runtime you point at a directory.
Not a chatbot. An organization with a job to do.
Quorum keeps intelligence and coordination separate. The brain — a headless agent CLI — does the work of one task. Everything around it — how those agents work together — is Quorum itself.
Goals become a task plan
A coordinator breaks goals into subtasks with dependencies, assigns each to the right role, and runs them in the correct order — independent tasks run simultaneously.
The organization hires specialists
When a task needs a skill the team lacks, the organization adds a specialist agent. Managers delegate work down the reporting structure, then review and integrate the results.
Independent review of all work
No agent evaluates its own output. An independent reviewer reads the actual files the task produced — not just the agent’s summary — and scores it 0–100; anything below the threshold is returned for revision. Tasks anchored to a deliverable only complete when the artifact verifiably exists.
Budget limits stop spending
Every API call is tracked in a ledger. Daily, monthly, and lifetime caps halt work — and you can stop everything immediately, even mid-task.
Watch it work, live
A live dashboard shows the organization structure, task board, spending, revenue, and activity — inspect any agent, task, or run, message the organization, and answer the requests it raises for you.
The organization works in cycles.
The system runs one cycle on a fixed interval (default 10 seconds). Each cycle processes the current task board — if the system crashes, it simply resumes where it left off. Five steps, then it repeats.
- 01
Absorb
Your messages become goals on the shared board, processed in order — multiple messages are never reordered.
- 02
Plan
One goal is selected in rotation. The team discusses and breaks it down into a task plan with dependencies.
- 03
Execute
Ready tasks run in dependency order across all goals, with a limit on concurrent tasks. Hiring and delegating complete within the cycle.
- 04
Review
An independent reviewer reads the artifacts each task produced and scores them. Below threshold returns for revision with feedback; invalid results fail.
- 05
Account
Progress and spending are recalculated; a goal completes only when all its tasks finish. Then the loop repeats — the organization keeps operating, reacting to new events and schedules, and stands itself down only when the mission is verifiably complete.
It doesn't wait for you.
Most agent tools run a task and stop. Quorum keeps operating — reacting to the outside world, scheduling its own work, and reaching real services — and asks for a human only when it genuinely needs one.
Reacts to the world
A ## Triggers section seeds goals on a schedule, from an inbound webhook, when a watched process finishes, or when its polled inbox has new mail — so the org advances off real events, not only your messages.
Schedules its own work
Agents book future goals and write their own cron jobs into ajobs/ folder — headless scripts that poll an inbox or drain a queue at no cost, waking the org only when there's something to act on.
Reaches real services
Through MCP servers and command-line tools you grant it, the org sends email, manages a calendar, queries a dashboard, or calls an API — and with its own channels, inbox, CRM, and merchant account it messages people, hears their replies, records leads, and bills customers with hosted checkout links. Each service is gated as its own allow / ask / deny capability.
Define the org
The whole company, in one file.
Drop a QUORUM.md in the directory to define the organization: a title, a budget, a mission, and one ### section per agent. The heading is the role name; the content describes what the agent does.
- No YAML, no schema. A recognized
key: valueline is metadata; everything else is plain text. - Reporting structure.
reports tonames a manager; omit it to place the agent at the top level as a coordinator. - Per-agent AI models. Assign cheaper models to junior roles; use advanced models for leads.
- Responds to events. A
## Triggerssection creates goals on a schedule, from webhooks, or when processes complete.
# Cobalt Ferry
A two-person studio shipping weekend projects.
- budget: 25
- provider: { "type": "claude", "model": "claude-opus-5" }
## Mission
Launch the landing page and collect 100 signups.
## Agents
### CEO
You are the CEO. Set direction, break the mission into tasks, delegate.
### Engineer
- reports to: CEO
- capabilities: { "crm": "allow" }
You are the engineer. Build what's asked and verify it works.Tools & permissions
Capabilities, allow / ask / deny.
Each agent has access to specific tools based on capability settings.allow grants the tool; deny blocks it;ask blocks it and sends you an approval request.
capabilitygrantsdefault
codeBash · Read · Write · Edit · Glob · Grepallow
consultTask — spawn sub-agent consultantsallow
webWebSearch · WebFetchallow
browsera real browser — portals, forms, signups (when wired)allow
delegateDELEGATE[id]: hand a subtask to a reportallow
hireHIRE[role]: employ a new specialistallow
crmCONTACT[email]: record a lead on the shared CRMask
chargeCHARGE[email | amount=…]: bill a customer — hosted checkoutask
Cost & budget
Caps that actually halt the loop.
Every AI model call reports its cost, tracked in a ledger with running totals. Set spending caps in the file, with environment variables, or using --budget.
Today$3.41
This month$62.90
Lifetime cap$100.00
quorum --budget 25 # halt for good at $25 spent
BUDGET_DAILY_USD=5 quorum # skip cycles past $5/daySteer it live
Watch it work. Steer by talking.
Open localhost:3000. The dashboard is a deliberately read-only window into the organization — every agent, task, and dollar inspectable — and chat is the steering wheel.
- Message the entire organization. Your directive becomes a goal in the next cycle, guiding the work.
- Reply to one agent. Select an agent from the dropdown; they respond directly.
- Answer requests. Agents ask for things they cannot do themselves — an account, credentials, or a decision. Approve, deny, or provide details, which are saved to their memory.
Dashboard · chat
you → whole orgShip the pricing page before outreach.
CEOUnderstood. Reprioritizing: pricing page first, then Growth starts outreach.
REQUEST · accessGrowth needs a Gmail account to run the campaign.
YesNoAnswer…
Ready when you areDefine a mission and a budget.
Define a mission and a budget.
The organization takes it from there.
$ npx @meffecta/quorum --prompt "Build and launch a tiny SaaS" --budget 20Drafts the org's QUORUM.md from your brief (one claude call), then boots it with a $20 cap.
Got a feature request? We'd love to hear it.