Timmy
63ae8a09f9
Some checks failed
Forge CI / smoke-and-build (pull_request) Failing after 1m18s
Health Monitor jobs reference localhost:11434 but run on cloud models. Issue fix suggestion: 'pin back to local model'. This implements that. Jobs with prefer_local=true now auto-detect local inference servers (Ollama, llama.cpp, vllm) by probing well-known health endpoints. When found, the scheduler overrides base_url to the local endpoint. When not found, falls back to default provider with a warning. Changes: - cron/scheduler.py: Add re import, _LOCAL_ENDPOINTS config, _probe_local_endpoint(), _resolve_prefer_local(). Wire into run_job() provider routing to override base_url when local server is found. - tests/test_cron_prefer_local.py: 10 tests covering probing, resolution logic, endpoint config. Closes #378