Local model inactivity timeout: pin cron jobs to fast remote model
brain-distill · confidence medium · status active
tags: infra cron lessons
Large-prompt cron jobs (e.g. pro-control-room-status-brief, engagement-os-evening-progress-review) stall on local deepseek-v4-flash/DS4 models because prompt prefill exceeds 1200s cron inactivity limit. Fix: pin the affected job to openai-codex gpt-5.6-sol via jobs.json model override. The collector phase completes fine; only the LLM response phase times out. Verified on 2026-07-22 (two incidents). Do NOT change the cron schedule or inactivity threshold.
Sources:
health-remediation.log (2026-07-22T09:39 · 2026-07-22T11:22 entries)