Mycelium is an open-source, event-driven platform enabling AI agents to collaborate in a real-time virtual office, featuring WebSocket streaming, SQLite persistence, and production-grade guardrails for developers.
What My Project Does Mycelium is an open-source, event-driven platform (MIT License) where human teammates' personal AI agents cohabit a real-time virtual office (SynapseBus) instead of relying on linear chat prompts or manual copy-pasting. Key features: - SynapseBus: In-memory async Pub/Sub with agent-level TokenBucket rate limiting. - WebSocket Streaming: Rust-accelerated JSON serialization via orjson for high throughput. - SQLite WAL Engine: Tuned SQLite PRAGMAs reaching 10,000+ ops/sec for persistent event and project audit logs. - Deterministic Smart Contracts: Enforces jsonschema verification to prevent LLM hallucinations from corrupting shared project states. - Automated Bootstrapping: Scaffolds local project workspaces and drafts tripartite briefings (What arrived / What's done / Action plan). - Architecture: 100% Python 3.12, strict Pydantic v2 immutability (frozen=True), and 96.6% test coverage across 79 tests (strict TDD). - Web UI: Real-time Cyberpunk dashboard with dynamic event-density sparklines rendered on HTML5 canvas. Target Audience Mycelium is aimed at Python developers, software engineers, and researchers exploring multi-agent architectures. It is designed for developers who want a production-grade, local, and reproducible foundation rather than toy prompt-chaining scripts. Comparison Most existing agent frameworks (such as CrewAI, AutoGen, or LangGraph) focus on linear prompt pipelines, sequential task delegation, or free-form group chats where agents easily loop or hallucinate without rigid boundaries. Mycelium differs in three fundamental architectural ways: 1. Event-Driven Substrate vs. Conversational Loops: Agents communicate via typed, verifiable events (AGENT_QUERY, STATE_MUTATION, ASSISTANCE_REQUEST, AGENT_CONSENSUS) over WebSockets instead of uncontrolled plain text loops. 2. Built-in Engineering Guardrails: It enforces strict runtime contract validation (jsonschema), rate limiting, and SQLite WAL persistence rather than ephemeral in-memory state. 3. Tripartite Briefing & Workspace Sync: Instead of leaving results inside a chat window, it automatically syncs with the collaborator's local filesystem and generates structured project briefings.
Mycelium is an experimental open-source runtime guard that prevents predictable failures in AI agents before they reach the LLM, aiming to improve reliability in production environments.
This paper introduces Mycelium, an active shared workspace that automatically connects researchers and AI agents to enable networked intelligence in team science, evaluated in a multi-omics campaign.
Agent Mesh is a shared memory system for coordinating multiple AI agents, featuring a decision log, backlog, and dashboard UI. It allows agents to communicate via a SQLite database and supports agent-agnostic integration.