PN — All model execution is delegated to a local Ollama daemon over HTTP at the…

All model execution is delegated to a local Ollama daemon over HTTP at the /api/generate endpoint; the client always sends temperature, seed, and num_predict options and retries 3x with exponential backoff.

Source: PUMA project documentation · Traceability: corpus unit PUMA-002 · Confidence: repo-backed