Temporal
used in
0
stacks
Durable-execution runtime that persists workflow state and automatically retries failures, used to run long-running agent workflows reliably.
Executes application logic as Workflows whose state is persisted, so processes survive crashes, deploys, and intermittent failures with automatic retries of failed Activities. Originated as a fork of Uber's Cadence; the MIT-licensed server is written in Go, with SDKs for multiple languages. Available self-hosted or as the managed Temporal Cloud, and increasingly used to make multi-step, long-running agent loops and tool calls durable.
Strengths
- Workflow state persists across crashes and deploys, suiting long-running agent loops
- Automatic retries and timeouts for failed activities such as LLM and tool calls
- MIT-licensed open-source server with multiple language SDKs
- Managed Temporal Cloud option avoids self-hosting the cluster
Tradeoffs
- Self-hosting the server cluster and its datastore is operationally heavy
- Deterministic workflow constraints impose a real learning curve on developers
- Heavier abstraction than simple job queues for short-lived tasks