Git Tools

Plot twist: Gitea now hosts AI models alongside your repos. Self-hosted, OpenAI-compatible API, Ollama backend. If you've been wanting a GitHub Copilot equivalent that lives entirely on your own metal — behold.

The Setup

Gitea AI is a new opt-in module that bundles an Ollama runtime, a model registry, and an OpenAI-compatible inference endpoint into the existing Gitea container. Pull a model, point your IDE at the endpoint, profit.

# docker-compose.yml
services:
  gitea:
    image: gitea/gitea:1.24-ai
    environment:
      - GITEA__ai__ENABLED=true
      - GITEA__ai__BACKEND=ollama
      - GITEA__ai__DEFAULT_MODEL=qwen2.5-coder:7b
    volumes:
      - ./gitea:/data
      - ./models:/var/lib/ollama
    ports:
      - "3000:3000"
      - "11434:11434" # ollama
    deploy:
      resources:
        reservations:
          devices:
            - capabilities: [gpu]

The Money Pattern

Once it's up, the inference endpoint speaks the OpenAI dialect. Point Cursor, Continue, or any SDK at it:

curl http://gitea.local:3000/api/ai/v1/chat/completions \
  -H "Authorization: Bearer $GITEA_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen2.5-coder:7b",
    "messages": [{"role":"user","content":"review this diff"}]
  }'

The Catch

It's separate from Forgejo. The Gitea/Forgejo fork is already a sore subject and this widens the gap — Forgejo isn't shipping the AI module and probably never will. If you care about the open-governance fork, you're stuck running a second service. Ecosystem fragmentation is real.

The Verdict

For a small team running everything on-prem — think a Rebuild Relief-style internal tooling setup — Gitea AI is a stupidly good zero-config win. One box, your repos and your code-completion model. If you're already on GitHub, ignore this. If you're already on Gitea, upgrade tonight.