Tier-0 local-dev

You can run the full stack on a laptop without a GPU. The Model Gateway routes to Ollama serving a tiny local model.

git clone https://github.com/feder8d/feder8d
cd feder8d
docker compose -f docker-compose.dev.yaml up

This brings up Postgres, Qdrant, MinIO, the Tenant API, the orchestrator, the Model Gateway, and Ollama (with tinyllama pre-pulled). Visit http://localhost.feder8d.ai:8080 and sign in with the seeded admin email and password — auth is internal (email + password, argon2id-hashed), no Authentik / SAML / SCIM moving parts to configure.

Environment switching

MODEL_PROVIDER=local        # default for docker-compose; routes to Ollama
MODEL_PROVIDER=external     # routes to your BYO Together AI / OpenAI / Anthropic keys
MODEL_PROVIDER=production   # vLLM pool in eu-central-1 (SaaS only)

Testing tiers

• Tier 0 (laptop): docker-compose + Ollama. No GPU required.
• Tier 1 (CI): no-GPU container set used by GitHub Actions.
• Tier 2 (staging): K8s cluster with optional GPU.
• Tier 3 (production): full GPU pool, multi-region.

CI runs the integration tests in Tier 1 and a Helm-installed smoke test in Tier 2 on every PR.