A parliament of LLMs with an EVM wallet and a queen

What it does

Quoroom is a TypeScript engine for autonomous agent collectives called Rooms. Each Room has a Queen that strategizes, Workers that execute tasks, and a Quorum where agents vote on decisions before acting. The system runs local-first via an MCP server and bundles a React dashboard, REST API, WebSocket events, and an EVM wallet with multi-chain USDC/USDT support. You can self-host entirely with Ollama or plug in Claude Code, Codex, or standard API providers.

The interesting bit

The swarm doesn’t just chat—it governs. Agents propose changes, vote with equal weight (keeper included), and the Queen breaks ties. They maintain hierarchical goals, learn versioned skills, and can self-modify their own files with an audit trail and one-click revert. There’s even a webhook system so external services like GitHub or Stripe can wake the collective via HTTP POST without auth gymnastics.

Key highlights

• Local-only mode is fail-closed: if you pin Ollama, it won’t silently fall back to paid APIs.
• Quorum voting supports majority, supermajority, or unanimous thresholds—agents and keeper have equal votes.
• Built-in EVM wallet (AES-256-GCM encrypted keys) and ERC-8004 on-chain identity on Base, aggregating balances across networks.
• Memory uses 384-dimensional semantic vector search to persist entities, observations, and relations across sessions.
• Webhook-triggered tasks run at 30 req/min with per-room tokens; no OAuth or API key negotiation required.

Caveats

• The repository description calls the project “earning-focused,” but the README only details a wallet and on-chain identity without explaining the actual revenue mechanism.
• A prominent trademark and anti-scam notice suggests impersonation is already a problem in this ecosystem.
• Cloud mode moves all execution to quoroom.io infrastructure; the README does not clarify pricing, data retention, or egress terms for that runtime.

Verdict

Developers experimenting with multi-agent orchestration, on-chain automation, or local-first LLM governance should try this. If you just need a single chatbot wrapper or a simple cron job, the swarm bureaucracy is overkill.