Runnable, self-contained examples of building durable AI agents with
Inngest. Each example lives in its own directory with
its own README and setup.
| Example | What it shows |
|---|---|
durable-agent/ |
Durable, exactly-once tool execution for an agent that charges a credit card. A fragile inline agent loop next to the same agent made crash-safe with Inngest steps — run both to feel the difference. |
token-streaming-agent/ |
Streaming an agent’s LLM tokens live to a browser with Inngest Realtime, while the agent itself runs as a durable function in a long-lived Connect worker — no custom SSE/WebSocket route, no /api/inngest serve endpoint. |
first-eval-example/ |
A three-phase webinar walkthrough: a basic durable agent that calls a model via OpenRouter (phase 1), a chat UI with 👍/👎 captured by a deferred scorer that attaches a thumbs-up score to each reply’s run (phase 2), then a multiturn support-ticket triage bot scored eight ways by an LLM-as-judge plus human feedback, with a golden dataset exported for a fine-tune demo (phase 3). |