Documentation

Everything you need to trace, test and ship.

Guides for the SDK, the eval runner and deploys. Press / or Cmd K anywhere to search.

START HERE

Introduction

What Tracewell records, how a run becomes a trace and where evals and deploys fit.

Quickstart

Install the SDK, wrap your agent and open your first trace in five minutes.

Core concepts

Runs, spans, projects, versions and datasets, defined in one page.

Getting started

Introduction

What Tracewell records, how a run becomes a trace and where evals and deploys fit.

Quickstart

Install the SDK, wrap your agent and open your first trace in five minutes.

Core concepts

Runs, spans, projects, versions and datasets, defined in one page.

Tracing

Instrument your agent

Wrap functions, add custom spans and attach attributes you can filter by.

Spans and attributes

What each span type stores and which attributes are indexed for search.

Sampling and retention

Keep the runs you need and expire the rest on a schedule.

Redacting data

Remove sensitive fields in your process, before a span leaves it.

Evals

Write your first eval

A case, a check and a score, in under twenty lines.

Rubric evals

Grade open-ended output with a model, a rubric and two examples.

Datasets

Build a dataset from production runs and pin versions of it.

Run evals in CI

Fail a pull request when the pass rate drops.

Deploy

Gates

Set the bar a version must clear before it can be promoted.

Canary releases

Send a small slice of traffic to a new version and watch the gates.

Rollbacks

Return to the previous version in one click or one command.

Reference

SDK reference

Every function the SDK exports, with arguments and defaults.

REST API

List runs, fetch traces and trigger evals over HTTP.

Stuck on something?

Describe the run, paste the run id and a person replies, usually within a day.

Ask the team