Baize Makes Long-Running Agent Runs Restart-Safe and Duplicate-Free

Baize, an agent runtime, treats each Run as a first-class database row rather than an in-memory goroutine, so long multi-step tasks survive process restarts and scaling. Runs can span many tool calls, human approval waits lasting hours, scheduled triggers, and bounded child runs, all of which can be interrupted by crashes, timeouts, or disconnects. All triggers are unified into a single queued Job with three types: normal runs, approval resumes, and child receipts. Runs use one suspended state, waiting for an external signal, with the specific reason stored in the payload. Its queue, datastore, and blob storage use swappable drivers, and the database rather than the queue serves as the source of truth for recovery.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
Discussion (0)
Log in to join the discussion and vote.
Log in