Define a prompt, pick a model, attach a trigger. Opencode runs your routine on cron, on filesystem changes, or anywhere code lives.
42 total runs · 7 routines
Schedule routines on a cron, or watch a directory and react to file changes — pull requests, log files, screenshots, anything.
Anthropic, OpenAI, or anything OpenAI-compatible. Pick a default, or per-routine.
Streamed output, tool calls, cost, and a follow-up prompt — all in one timeline.
Local CLI, your laptop, or a tiny VPS. State stored as plain files. No vendor lock-in.
See the prompt, the tools the model used, and the result. Reply with a follow-up and the routine picks up right where it left off.
Found 3 open PRs:
#142 — Add rate limiting to API endpoints
Token bucket limiting on /api/routines. No edge-case tests — recommend adding timeout scenarios.
#139 — Fix SSE connection leak
Small focused fix. Looks good.
#136 — Migrate schema to support tags
Migration is not idempotent — add an IF NOT EXISTS guard.
Connect a provider once, then pin the models you actually use. Switch per routine when you need cheaper, faster, or smarter.