Use Tracing to see every request

Every chat request, from every channel (API, widget, Playground and workflow Agent blocks), is recorded as an OpenTelemetry trace. Tracing in the sidebar lists them newest first. Traces are kept for 90 days.

Before you start

• At least one request to an agent. Send one from the Playground if you have none yet.
• Optional: the trace id or your own X-Request-Id for the request you're after. The API returns the trace id in the meta and done events and the X-Trace-Id header.

Steps

1. 1

   Open Tracing

   Click Tracing under Observability in the sidebar.

2. 2

   Pick a time range

   Choose 24 hours, 7 days, 30 days or 90 days.

3. 3

   Filter

   Narrow the list by agent, provider, model, status, error code, whether tools were used, streaming and feedback. Search looks in the message preview.

4. 4

   Read the row

   Each row shows the status (OK or the error code), when it ran, the agent and model, duration, tokens and how many model rounds and tools it took.

5. 5

   Open the trace

   Click a row. Preview shows the conversation turn and the steps in order, Events shows the span events (model input and output, tool input and output), and JSON shows the raw spans and attributes.

What a trace captures

• The full prompt sent to the model and the reply, the knowledge passages it was given and each tool call's input and output, as span events.
• Emails, API keys and phone numbers written with a leading + are redacted before storage, and each value is cut to 10 KB.
• Tokens in and out, the model and provider, time to first token, duration, whether the answer came from the answer cache, the error code if any, and your credit balance.

Traces are yours, in your account. SentientOne's own product analytics never include message text, prompts, replies or keys.

If something goes wrong