Reasoning

Runlet exposes provider-native reasoning when the underlying provider returns it. Runlet does not synthesize reasoning text on its own.

Non-streaming

Use RunResult.reasoning or ModelResponse.reasoning:

result = await runtime.run_request(agent, request)
print(result.output)
print(result.reasoning)

Streaming

Reasoning deltas are emitted as:

• model.stream.reasoning_delta

async for event in runtime.stream_request(agent, request):
    if event.type == "model.stream.reasoning_delta":
        print(event.payload["delta"], end="")

Important note

Whether reasoning is actually returned depends on the provider or gateway. Some OpenAI-compatible gateways expose reasoning tokens in usage data without returning readable reasoning text.

That behavior is upstream. Runlet only surfaces what the provider returns.