Inngest

What Inngest does

• Checkpointed step functions Each named step stores its result as a checkpoint, so retries and resumed runs reuse completed-step results instead of re-executing their callbacks.[1]

• Compute-free waits Steps can sleep for a delay or wait for a matching event with a timeout, and no worker runs during the wait, so long pauses cost no idle compute.[1]

• Automatic retries Functions declare retry counts and the executor re-runs failed steps against saved state, covering transient downstream errors without bespoke retry code.[1][2]

• Queue flow control Concurrency limits, throttling, prioritization, debouncing, rate limiting, and batching cap load against fragile APIs and databases during bursts and replays.[1]

• Event, schedule, and webhook triggers Runs start from ingested events, cron-style schedules, or inbound webhooks, covering background jobs, recurring tasks, and event-driven fan-out.[1]

• Per-run traces Inngest records steps, waits, retries, and errors for every run, so teams inspect what happened without adding separate instrumentation.[1][2]

• Agent execution primitives Agents and harnesses map onto series of steps, and paid cloud tiers add concurrent sandboxes for isolated runs of multi-tool loops.[1][2]

• Own-compute execution Functions deploy on the team's current infrastructure while Inngest coordinates the run, and local development starts against the single-binary server before promoting to self-hosted or cloud production.[1][2]

Who will love it

• Offload image processing, imports, notifications, and webhooks to durable functions so request paths stay short while retries and traces cover the background half.[1][2]

• Model order flows, onboarding sequences, and LLM tool loops as checkpointed steps with event waits and flow controls that plain queues express awkwardly.[1][2]

• Keep events and run state on controlled infrastructure via the single-binary or Docker self-host path, provided the team can operate it without guaranteed vendor support.[1]

Where it falls short

• Functions are defined in code with the TypeScript or Python SDKs and operating runs assumes backend skills; the dashboard monitors runs rather than replacing development.[1]

• The core ships under the Server Side Public License 1.0 with an Apache 2.0 future notice, which is source-available; license-sensitive adopters should clear that with legal first.[1]

• Direct support for self-hosted instances is not guaranteed, so teams that cannot carry that operations load should price the managed cloud tiers instead.[1][2]

Inngest versus Hatchet