#Events

Every run yields a typed AgentEvent stream — the single channel for streaming text to a UI, showing tool activity, and servicing approvals and user questions. Events are observe only: handling them never changes agent behavior, so you can build any rendering or logging layer on top without touching the agent.

You already consume the stream whenever you iterate query():

import { query } from "@lite-agent/sdk";
import { anthropic } from "@lite-agent/provider";

for await (const ev of query({
  prompt: "Summarize this project.",
  model: anthropic(),
  modelName: "claude-sonnet-4-6",
  cwd: process.cwd(),
})) {
  if (ev.type === "text_delta") process.stdout.write(ev.text);
}

Events forwarded from a subagent carry an agentId; the main agent's events don't.

#Rendering text

Text arrives incrementally as text_delta chunks — write them through for a streaming UI. When a turn's full assistant message is ready, a message event carries it; when the whole run finishes, done carries the final RunResult:

for await (const ev of query({ /* ... */ })) {
  switch (ev.type) {
    case "text_delta":
      process.stdout.write(ev.text);       // stream chunks as they arrive
      break;
    case "tool_call_start":
      console.error(`\n[tool] ${ev.call.name}`);
      break;
    case "error":
      if (ev.fatal) console.error(ev.error);
      break;
  }
}

#Interactive events: approvals and user input

Two event pairs correspond to points where the run suspends and waits for your handler:

• approval_request → approval_resolved — the permission gate returned ask for a tool call; your onApproval handler answers "allow" or "deny". See Permissions.
• input_request → input_resolved — the model called ask_user; your onAskUser handler returns the answer string.

Use the events to render the prompt in your UI (e.g. show which tool call is awaiting approval); use the handlers to actually answer. The run blocks until the handler resolves.

#Full event reference

#See also

• Agent loop — the five steps that produce these events.
• Sessions — events are what gets persisted per session.
• Permissions — the approval flow behind approval_request / approval_resolved.
• Core strategies — ApprovalHandler and InputHandler interfaces.