One durable queue
Jobs, attempts, state transitions, and output live in one SQLite database that every worker can safely share.
spoolctl
A local job queue built for operators that die.
Work survives when nobody can watch.
spoolctl runs shell jobs through one SQLite file. Workers are equal peers: there is no daemon, broker, or coordinator whose failure can stop the queue.
Recovery after a crash
If a worker dies during a job, another worker reclaims the work after the safety threshold and retries it.
No daemon or broker
Start as many symmetric workers as you need. Nothing central must remain alive for the queue to work.
Safe for automation
Use stable JSON envelopes, explicit exit codes, retries, backoff, and a durable dead-letter state in scripts and agents.
Composes with the toolchain
evalctl can send long evaluation runs to spoolctl for durable execution. Use inferctl to inspect local model routing before a model-backed job starts. spoolctl keeps the command, output, and recovery state safe while work runs.
spoolctl add