Inngest
used in
0
stacks
Durable functions and event-driven queueing platform for agent backends, with the AgentKit framework for building AI agents.
Provides durable, step-based functions (step.run, step.waitForEvent, step.invoke) where each step retries independently and state survives failures and deploys. Includes event-driven fan-out, flow control such as rate limiting, throttling, and per-tenant concurrency, and built-in execution tracing with bulk replay. Functions deploy to existing infrastructure including serverless platforms and traditional servers; agents built with its AgentKit can pause mid-execution and resume on input.
Strengths
- Step-level retries and durable state without managing workers or queues
- Flow control (throttling, per-tenant concurrency, rate limits) as code, useful for LLM rate limits
- Runs on existing serverless or server infrastructure rather than dedicated workers
- AgentKit framework targets pausable, resumable AI agent workflows
Tradeoffs
- Execution orchestration depends on the Inngest service or self-run server, adding a platform dependency
- Less low-level control than operating your own queueing or workflow cluster