Get started

    Your first trace

    Open the first trace the SDK sent, read the model call behind it, and understand tokens, estimated cost and what the tabs show.

    After this page you can find the request you just made, read the exact prompt and completion of the model call inside it, and know what every number on the trace page means. It is the third step of the Home checklist.

    Find it

    The quickstart page shows a toast, First trace received, with View trace. Otherwise open Traces: the table lists time, operation, service, status, latency, tokens, estimated cost, score and quality chips for the selected Environment. Click the row.

    Read it

    The header carries the trace id, status, duration, tokens, estimated cost and — when your code set them — session, user and app version. Four tabs: Waterfall (the span tree in time, with a type legend for model call, tool, http, db, function, log and eval result), Logs (the log lines the SDK captured, in order), Evaluation (empty until something scores the trace) and Metadata (Environment, app version, SDK version, content capture, a span breakdown by type and the slowest work by self time).

    Click the llm_call span. The panel shows the model and provider, the request as a chat transcript (system · user · assistant · tool), the completion, a thinking block when the model produced one, tokens in and out with cached and reasoning tokens listed separately, time to first token when the call streamed, the estimated cost, tool calls with arguments and results, and the raw attributes. Cache-read and cache-write tokens are subsets of the input tokens, so they are shown beside the total rather than added to it.

    If the prompt says “not captured”

    A service started with capture_content=False (or a mask hook that dropped the body) still records tokens, latency, cost, tool names and the span tree, and the panel prints the reason in place of the text. The Metadata tab names the setting and where it lives. See Configuration and privacy.

    What the checklist does next

    Home’s Getting started card has five steps — Environment key, Install SDK, Receive first trace, Add an evaluator, Run first evaluation — each detected from the project’s data on every load (GET /onboarding/checklist). Steps two and three are done now. Add an evaluator continues in Your first evaluation. The card can be dismissed from the user profile; it reappears only if the project loses its last Environment.

    EvalKit is built by Syntropylabs. Published on PyPI and npm.