Build AI agents that survive crashes and retries.
Guides the construction of durable AI agents whose model calls, tool actions, approvals, and progress survive retries and interruptions.
When to use it
Use when building, migrating, or debugging AI agents, multi-step workflows, tool-calling loops, human review flows, or realtime agent interfaces.
It changes how the agent builds AI workflows by placing agent runs, model inference, tool actions, and approvals inside durable Inngest functions.
What you provide
The AgentKit implementation uses the @inngest/agent-kit library to define agents and model providers.
This skill
AgentKit documentation
Checks current AgentKit documentation