Mycelium is an autonomous agent system that prevents engineering knowledge loss as teams and codebases evolve. It infers who understands what in a codebase, where knowledge is fragile, and acts directly inside GitLab to stabilize it.
Built for the Google Cloud Rapid Agent Hackathon 2026 using Vertex AI, ADK, Gemini, GitLab MCP, and MongoDB MCP. Runs an 8-stage autonomous pipeline: Observe → Model → Analyze → Decide → Act → Reflect → Persist → Summary.
| Document | Description |
|---|---|
| docs/SETUP.md | Prerequisites, environment config, local run, webhooks, and deployment |
| docs/ABOUT.md | Project writeup: inspiration, design, challenges, and roadmap |
| docs/TECH.md | Full technology stack: all frameworks, services, and libraries used |
| docs/PROJECT_INFO.md | Full project overview: design decisions, pipeline stages, agent architecture, and roadmap |
| docs/HACKATHON.md | Hackathon compliance requirements and partner integration notes |
mycelium/
├── docs/ # Project documentation
│ ├── SETUP.md # Setup guide: local run, environment, deployment
│ ├── ABOUT.md # Project writeup: inspiration, design, challenges, roadmap
│ ├── TECH.md # Full technology stack reference
│ ├── PROJECT_INFO.md # Full design and architecture reference
│ ├── HACKATHON.md # Hackathon compliance requirements
│ └── draw.io/ # Architecture diagrams
├── mycelium/ # Backend: FastAPI + ADK pipeline agents
├── mycelium-ui/ # Frontend: Next.js monitoring UI
├── deploy.py # Cloud Run deployment helper
└── LICENSE # Apache 2.0
Backend: mycelium/
FastAPI service hosting the 9-stage pipeline, ADK agents, MCP connectors, and MongoDB knowledge graph driver.
Frontend: mycelium-ui/
Next.js dashboard for live pipeline monitoring, agent activity, knowledge graph visualization, and configuration.
Deployment: deploy.py
Cloud Run deployment script for both the backend API and the UI.