Summary

• Add AI_PROVIDER env var to switch between openrouter (default), ollama, and openai backends
• When AI_PROVIDER=ollama, the OpenAI SDK client points at Ollama's local API (http://localhost:11434/v1) with a dummy API key — no data leaves the machine
• Embedding service also respects the provider switch, using nomic-embed-text (768-dim) for Ollama
• Existing OpenRouter behavior is completely unchanged when AI_PROVIDER is unset or set to openrouter

What changed

New env vars

AI_PROVIDER=ollama              # openrouter | ollama | openai
OLLAMA_BASE_URL=http://localhost:11434/v1
OLLAMA_MODEL=llama3.2
OLLAMA_EMBEDDING_MODEL=nomic-embed-text

How to test

1. Install Ollama: https://ollama.com
2. Pull a model: ollama pull llama3.2 && ollama pull nomic-embed-text
3. Set AI_PROVIDER=ollama in backend/.env
4. Start the backend — all LLM calls now go to localhost

Honesty contract

• I have tested that npx tsc --noEmit passes with 0 errors
• No frontend changes were made
• Default behavior (OpenRouter) is preserved when AI_PROVIDER is unset

Closes #53

🤖 Generated with Claude Code