The SDK includes a built-in tracing subsystem to give you full visibility into agent runs. You can track model calls, token usage, tool executions, memory operations, handoffs, and errors.
Using the Tracer
You can instantiate a Tracer and pass it to your Agent configuration. The agent will then automatically log telemetry to that tracer without changing its run() response signature.
typescript
import { Agent, OpenAIProvider, Tracer } from "@rabbit-agent-sdk/rabbit-agent-sdk";
const tracer = new Tracer();
const agent = new Agent({
provider: new OpenAIProvider(),
tracer,
});
await agent.run("Calculate 25 * 4");
// Retrieve the trace for the last run
const trace = agent.getLastTrace();
console.log(trace?.events);Trace Events
The Tracer captures a variety of fine-grained events. These are particularly helpful for debugging, monitoring costs (via token tracking), and auditing agent behavior.
Some of the events included in the trace are:
RunStartandRunEndModelCall(includingpromptTokensandcompletionTokens)ToolCallandToolResultMemoryReadandMemoryWriteHandoffError