Quickstart: the shared foundation

Almost every app in AI Workforce needs the same five ingredients. Set them up once and every department guide gets shorter.

1. A model to think with

You need a model to run inference. Everything here speaks the OpenAI-compatible chat API, so you can point at whatever you like — swap the endpoint, not the code.

  • Local (private / free): Ollama exposes an OpenAI-compatible endpoint at http://localhost:11434/v1. 
    curl -fsSL https://ollama.com/install.sh | sh
ollama run llama3.1
  • Self-hosted gateway: LiteLLM or LocalAI give you one OpenAI-compatible URL in front of many models.
  • Hosted: any provider that exposes an OpenAI-compatible endpoint — set LLM_BASE_URL and LLM_API_KEY.

Every app in this repo is provider-agnostic. The three config values are LLM_BASE_URL, LLM_MODEL, and (optionally) LLM_API_KEY.

2. A place to remember things (vector store)

For anything that does retrieval (support bots, doc Q&A, research):

docker run -p 6333:6333 qdrant/qdrant

Qdrant is the easy default. pgvector on your existing Postgres also works.

3. Docker + Docker Compose

Nearly every app ships a docker-compose.yml. Install Docker Desktop (Mac/Windows) or Docker Engine (Linux). Verify:

docker compose version

4. A domain + reverse proxy (for anything public)

Put a proxy in front so you get HTTPS and can host several apps on one box:

  • Caddy — automatic HTTPS, tiny config.
  • Traefik — great for many containers.

5. A place to keep secrets

Never commit keys. Every app here reads from a .env file:

cp .env.example .env
# edit .env, then:
docker compose up -d

For teams, graduate to Infisical (open-source secrets manager).

That's the foundation. Now pick a department and ship.