---
description: Trace, debug, and monitor AI agents running on the BEAM.
---

# WeaveScope

WeaveScope shows you what happened inside every agent run. Follow model calls,
tool calls, errors, latency, token usage, and cost from one trace, then monitor
those signals across production.

[Start free](https://app.weavescope.com/auth/login) · [Send your first trace](quickstart.md)

## What you can do

### Trace an agent run

See the complete run tree, including model calls, tool calls, inputs, outputs,
errors, tokens, and cost.

### Debug a failure

Find the step that failed, compare timings in the waterfall, and inspect the
data recorded for that observation.

### Monitor production

Track traffic, latency, errors, token usage, model cost, and tool performance
over time.

## How WeaveScope organizes your data

**Projects** keep traces for different applications or environments separate.
Each project has its own API keys and retention settings.

**Traces** represent complete agent runs, such as one user request, background
job, or graph execution.

**Observations** are the individual steps inside a trace. An observation can be
an agent, model call, tool call, retriever, guardrail, chain, or generic span.

**API keys** connect telemetry to a project. Keys begin with `ws_`, are shown in
full only once, and should be stored as secrets.

## Start tracing

If your application uses Elixir, the fastest route is
[BeamWeaver](beamweaver-integration.md). Configure a project API key and add
trace metadata at the boundary of the operation you want to observe. BeamWeaver
then exports the agent, model, and tool observations in the background.

If you use another client or an OpenTelemetry collector, send data through the
[Ingestion API](ingestion.md).

Once traces arrive:

1. Use [Tracing](debug-traces.md) to inspect individual runs.
2. Use [Monitoring](monitoring.md) to follow production trends.
3. Add [model pricing](model-pricing.md) if a custom model shows zero cost.