Skip to main content

​
Overview

Run cancellation lets you abort an agent mid-execution using the standard AbortSignal API. The signal is checked at the start of each LLM roundtrip and before tool execution.

​
Quick Start

import { Agent, openai } from "@radaros/core";

const agent = new Agent({
  name: "researcher",
  model: openai("gpt-4o"),
  tools: [searchTool, analyzeTool],
});

const controller = new AbortController();

// Cancel after 10 seconds
setTimeout(() => controller.abort(), 10_000);

const output = await agent.run("Research quantum computing advances", {
  signal: controller.signal,
});

if (output.status === "cancelled") {
  console.log("Run was cancelled");
}

​
Cancellation Points

The signal is checked at three points in the execution loop:
  1. Before each LLM call — prevents starting a new API request
  2. Before tool execution — prevents running tools after cancellation
  3. At run start — immediately returns if already aborted

​
Output on Cancellation

When a run is cancelled, it returns a RunOutput with: 
{
  text: "",
  toolCalls: [],
  usage: { promptTokens: 0, completionTokens: 0, totalTokens: 0 },
  status: "cancelled",
  runId: "...",
  agentName: "researcher",
  durationMs: 5023,
}

​
Events

The run.cancelled event is emitted on cancellation: 
agent.eventBus.on("run.cancelled", ({ runId, agentName }) => {
  console.log(`Run ${runId} of ${agentName} was cancelled`);
});

​
Error Class

You can import RunCancelledError for custom handling: 
import { RunCancelledError } from "@radaros/core";

​
HTTP API Cancellation

When using the transport layer, you can implement cancellation via request timeouts or client disconnects by wiring AbortSignal in your middleware.