Inngest

Event keys are used to send events — each key is a credential that authorizes event ingestion. Signing keys verify that Inngest is calling your functions (request signing). The SDK handles request verification automatically. Event keys can be rotated in the dashboard. No fine-grained scoping on event keys — a key can send any event type. The cloud platform handles function execution security. For self-hosted development, the Inngest dev server runs locally with no auth required. Production deployment on Inngest Cloud requires account authentication. The security model is appropriate for the event-driven pattern: event keys for ingestion, signing keys for verification, platform auth for management.

Inngest provides event-driven function orchestration — agents send events, and Inngest routes them to registered functions with automatic retries, concurrency control, and step-based workflows. The model is particularly agent-friendly: agents fire events (e.g., 'user.signed_up', 'order.completed') and Inngest handles reliable execution of downstream processing. Step functions enable multi-step workflows where each step is independently retried on failure. Fan-out enables one event to trigger multiple functions. Scheduled functions (cron) run on a recurring basis. The SDK-first approach means functions are defined in code alongside the application — no separate job infrastructure to manage. Concurrency controls prevent resource exhaustion. For agents building event-driven architectures on serverless platforms (Vercel, Netlify, etc.), Inngest eliminates the need for separate queue infrastructure.

The primary integration pattern is SDK-based: define functions in code using the Inngest SDK, register them with the platform, and trigger via events. Events are sent via SDK or REST API (POST to /e/{event-key} with event name and data). Functions are declared with trigger conditions, step definitions, and configuration (retries, concurrency, timeout). Step functions use step.run(), step.sleep(), step.waitForEvent(), and step.sendEvent() primitives for workflow composition. The REST API provides event sending, function management, and run inspection. Event replay enables re-processing historical events. Batch events send multiple events in a single request. The data model: events → functions → steps → runs. The Dashboard API provides observability into function runs, errors, and performance.

Automatic retries are a core feature: failed steps retry with configurable backoff (default: exponential with jitter). Retry count is configurable per function (default: 3). Non-retryable errors can be thrown to skip retries. Step-level retries mean only the failed step re-executes, not the entire function — this is efficient for multi-step workflows. Function timeouts prevent runaway executions. Concurrency limits prevent resource exhaustion with configurable per-function limits. Dead letter queues capture permanently failed runs. The dashboard provides real-time visibility into function runs, errors, and retry attempts. Rate limiting on event ingestion is generous. The main consideration for agents: step idempotency is the developer's responsibility — Inngest guarantees at-least-once execution, so steps must be safe to replay.

Documentation at inngest.com/docs is excellent — well-structured with quickstart guides, concept explanations, SDK references, and integration guides for frameworks (Next.js, Express, Flask, Django, etc.). The TypeScript SDK documentation is particularly strong. Python SDK documentation is newer but adequate. Concept guides explain events, functions, steps, retries, concurrency, and scheduling with clear examples. The local dev server provides a visual dashboard for development-time function testing and event inspection. Community Discord is active with responsive maintainer support. The documentation addresses serverless deployment patterns explicitly, which is relevant for agents deploying to Vercel, Netlify, or similar platforms. For agents new to event-driven architectures, the conceptual documentation provides a good learning path.