> ## Documentation Index
> Fetch the complete documentation index at: https://evalgate.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Capture your first trace

> Plain-English setup for sending one safe interaction to EvalGate.

A trace is one recorded interaction in your app. It can be an AI answer, an API
request, a background job, or a tool call. EvalGate uses traces as evidence so a
team can turn real behavior into checks and catch regressions before shipping.

## Start local

You do not need an EvalGate API key, model provider key, or telemetry collector
to inspect a repository, create local health checks, or run a local gate.

```bash theme={null}
npx @evalgate/sdk understand
npx @evalgate/sdk health --config evalgate.application.json --format github
npx @evalgate/sdk gate
```

## Send traces when ready

When you want EvalGate to store trace history in a workspace, choose the path
that matches your app:

* Python AI app: use OpenInference or OpenTelemetry instrumentation and export
  OTLP/HTTP traces to `/api/otel/v1/traces`.
* TypeScript or Node app: use an OTLP HTTP trace exporter with the EvalGate
  endpoint and an EvalGate API key.
* Existing collector: add EvalGate as an `otlphttp` exporter.
* No tracing library yet: send one manual collector payload to `/api/collector`.

Send one deterministic test payload first. That proves the workspace can accept
and store trace evidence without sending your app's real data.

## What to expect

After a trace is stored, open **Evidence** from **Search** (traces lens) and
search for its trace ID. You should see
the trace title, spans, inputs and outputs that survived your redaction policy,
token and model metadata when present, and a source of `otel` or `api`.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.