For one-lane LLM / agent runners
When LLM, agent, or opencode jobs share a single execution lane —
one concurrency slot, one rate-limited plan — naïve cron times quietly collide and
wake the model with a cold prompt cache. This planner lays your day on a 24-hour
timeline, flags every overlap, and auto-staggers the schedule so each heavy job runs alone.
Heavy jobs that finish, then restart within the cache TTL keep a warm cache. Gaps longer than the TTL are marked cache-cold — you pay a full re-read.
No jobs yet. Add one on the left, or load the sample day.
Lite LLM plans and self-hosted agents often run one heavy task at a time. Two cron jobs at 09:00 and 09:02 don't run in parallel — the second queues, stalls, or trips a rate limit. The planner treats heavy jobs as exclusive and shows where they overlap.
A job that just ends as the next begins still risks teardown lag and token-bucket recovery. A small buffer between heavy jobs absorbs that. Set yours and the planner enforces it.
Prompt caches expire after a few minutes idle. Restarting a related job inside the TTL reuses the warm cache; waiting longer means a full cold re-read. The planner marks each gap warm or cold so you can cluster related work.
Everything stays in your browser via localStorage. No account, no upload, no backend, no tracking.