Events
Read the normalized events Agentdock emits while a run is active.
Await agent.stream() to get two values: stream, which gives you events as they happen, and result, which resolves to the final run result. Each event has a type, run and session IDs, a timestamp, and sequence numbers.
Event types
| Event | What it means |
|---|---|
run.started | The run began. |
message.started | A user, assistant, or tool message began. |
message.part.delta | A message content part arrived. |
message.completed | A message finished. |
tool.called | The agent requested a tool. |
tool.progress | A tool reported progress. |
tool.completed | A tool returned a result. |
tool.failed | A tool failed. |
interrupt.required | The run is waiting for approval. |
interrupt.resolved | An approval decision was applied. |
usage.updated | Token usage or cost information arrived. |
run.completed | The run finished. |
run.failed | The run failed. |
run.cancelled | The run was cancelled. |
Read text as it arrives
const { stream, result } = await agent.stream(prompt, context, { sessionId });
for await (const event of stream) {
if (event.type === "message.part.delta" && event.part.type === "text") {
showText(event.part.text);
}
}
const finalResult = await result;Message parts can also contain reasoning, media, files, citations, tool calls, tool results, and custom JSON. Use the type field to handle each part.
Check event versions
Events include protocolVersion. Compare it with AGENT_EVENT_PROTOCOL_VERSION before processing if your app may receive events from a different Agentdock version. The @agentdock-ai/contracts package also exports reduceAgentEvent() to build a UI state from the stream.