agents npm package. Braintrust traces every runAgentTool call, capturing the agent class name, tool inputs, tool outputs, and errors.
For Cloudflare Workers deployments, use manual instrumentation with
wrapCloudflareAgent(). The --import auto-instrumentation hook only runs under Node, not in the Cloudflare Workers runtime (workerd). See the Cloudflare setup guide for enabling nodejs_compat and flushing traces with ctx.waitUntil().Setup
Install Braintrust alongside theagents package, then set your API keys. Requires agents v0.17.0 or later.1
Install packages
2
Set environment variables
.env
Manual instrumentation
Manual instrumentation is the recommended approach for Cloudflare Workers. CallwrapCloudflareAgent() on the base Agent class at module scope so all subclasses inherit Braintrust tracing without any changes to your class definitions. Initialize the logger inside fetch with your env bindings, then pass logger.flush() to ctx.waitUntil() so buffered traces ship after the response returns.cloudflare-agent-manual.ts
wrapCloudflareAgent() returns the same class it receives after patching its prototype. Calling it on the base Agent class instruments all subclasses without modifying your class definitions.Deploying to Cloudflare Workers requires the nodejs_compat compatibility flag and storing BRAINTRUST_API_KEY as a Wrangler secret. See the Cloudflare setup guide for the full deployment configuration.Auto-instrumentation
Auto-instrumentation patches the SDK at runtime without modifying your application code, but the--import hook only runs under Node (local development or tests), not in the Cloudflare Workers runtime. For a deployed Worker, use manual instrumentation above.1
Initialize Braintrust and define your agent
2
Run with the import hook
node --import can run the file directly. The Braintrust APIs work the same in TypeScript projects — compile your TypeScript to JavaScript, then run the compiled file with the import hook.If you’re using a bundler, see Trace LLM calls for plugin and loader setup.
What Braintrust traces
Braintrust captures:- Tool spans for each
runAgentToolcall, named after the agent class (for example,MyAgent). - Tool input passed to the agent tool call.
- Tool output returned by the agent, or an error if the call fails.