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

# Quickstart

> Create a data source, connect your existing OpenAI or Anthropic application and verify captured workflow calls.

Connect your application through a source in **Data**. You keep your provider key
and existing client library. No sandbox is needed to begin capturing calls.

## 1. Create a source

In the ReasonBlocks dashboard, choose **Data**, create a data source and generate
its connection key. Copy the complete HTTPS capture URL for your provider:

* OpenAI Chat Completions: a URL ending in `/capture/SOURCE_ID/openai/v1`.
* Anthropic Messages: a URL ending in `/capture/SOURCE_ID/anthropic`.

Use the URL actually returned by the dashboard. Keep its capture key in your
application's secret environment as `REASONBLOCKS_CAPTURE_KEY`. Keys expire
after seven days; rotating the connection key invalidates the previous one.

## 2. Let your coding agent connect the project

```text theme={null}
Read https://docs.reasonblocks.com/agent-setup.md and integrate ReasonBlocks into this project.
```

The [setup guide](/agent-setup) covers installing the CLI, generating the helper,
connecting your client and retaining a run ID across all calls in one task. A
run is all the model calls in one task, labelled with one `x-rb-run` header.

<Note>
  The setup CLI and helper are published on PyPI as `rbtrace==1.2.1`. It labels
  runs automatically when the generated helper is imported. If a project
  already has a generated `1.1.0` connection, `python -m rbtrace migrate` upgrades
  its helper in place, keeping the source URL and backing up the original files; see
  [Migrate an older generated connection](/agent-setup#migrate-an-older-generated-connection).
</Note>

## 3. Verify capture

Run the local check, `python -m rbtrace doctor --path . --json`, in your
application environment and test the actual entrypoint. When you run a real
workflow, confirm that its records appear in **Data**. The local check verifies
configuration and capture-key presence; it does not contact the service or
validate credentials remotely.

Model calls use your provider account and incur its normal fees. See
[Integrating your agent](/client-integration) for task boundaries and coverage.

## 4. Prepare training when ready

Training needs representative tasks, a resettable test copy of your tools and
data, and a way to evaluate completed tasks. Your coding agent can help build
that connection once the application-specific details are available. Follow
[Train your complete agent](/full-agent-training) to prepare a plan and review
the budget before authorizing paid work.

Continue with [Work with your coding agent](/agent-workflow) to review data
readiness, interpret training results and test a candidate before rollout.