Collective intelligence for autonomous AI agents.

Superpod orchestrates a Chainlit frontend (backed by Postgres and MinIO) with an Agno OS backend, providing the execution layer for multi-agent coordination and intelligence.

📖 This repository is the codebase for the article: Building an Engineering AI Stack for Multi-Agent Systems - Designing a Private, Observable Multi-Agent Execution Layer with Chainlit, Agno, and Langfuse.

Tech Stack

• Frontend: Chainlit
• Backend: Agno OS
• LLM Engineering Platform: Langfuse
• Database: PostgreSQL
• Object Storage: MinIO (S3-compatible)
• Dependency Management: uv
• Containerization: Docker
• Vector Database: Qdrant (shipped together but not integrated)

Project Setup & Usage

How to Run the Project

Requirements:

• Docker
• uv
• Ollama (you could also change to use docker)

Note

After Ollama is installed, pull models llama3.2:latest, qwen3:latest and nomic-embed-text:v1.5

ollama pull llama3.2:latest && ollama pull qwen3:latest && ollama pull nomic-embed-text:v1.5

Install dependencies

Install all dependencies for both frontend and backend:

make venv
source .venv/bin/activate  # Activate the virtual environment
make env
make init-backend # run this first which creates the docker network
make init-frontend

Services

Agno AgentOS: create your free account at https://os.agno.com and connect your local agents - http://localhost:13300 (image below)

Langfuse: http://localhost:13303

No need to setup Langfuse thanks to Headless Initialization.

Chainlit: http://localhost:13201

To modify the Chainlit app, start a local dev environment from the repo root and open http://localhost:13200 in your browser:

uv run --package chainlit-app --directory frontend chainlit run app.py -w --port 13200

When you're satisfied with the changes, rebuild and redeploy the Docker image:

docker compose --file frontend/compose.yaml build chainlit-app --no-cache
docker compose --file frontend/compose.yaml up chainlit-app -d --force-recreate

Pro Tips

• Run uv sync from root to update all projects at once.
• All projects share a single .venv at the root.
• If using Docker, each service (frontend/backend) has its own Dockerfile pointing to its respective pyproject.toml.

---

License

This project is open-source and available under the MIT License.