Server and Client Boundaries in AITracer

Agent Lab pages (run, chat, coach, training) are mostly client-driven: they call /api/local/* and /api/coach/* and show results immediately.

Trace persistence, auth, and ingest APIs stay on the server.

Server responsibilities

• session auth and API keys
• writing traces to Postgres (/api/local/run, /api/traces, coach ingest)
• optional policy evaluation and verification on ingest (advanced)

Client responsibilities

• Ollama model picker, prompts, coach comparison UI
• training export downloads and model build actions

Keeping inference on your Ollama instance (or your provider keys for coach) means the server stores records — it does not run your models for you unless you configure cloud provider execution separately.