braintrust@3.40.0

braintrust

3.40.0

Minor Changes

• feat(workers-ai): Add Clef and Clairic tracing (#2654)
• feat(logger): batch log rows with a short flush delay Batch nearby log rows into fewer HTTP requests in long-lived Node processes. Rows wait up to 1s before flushing, so call flush() before shutting down on signals or process.exit(). Configure the delay with BRAINTRUST_FLUSH_DELAY_MS; serverless and other runtimes continue to flush immediately by default. (#2653)

Patch Changes

• fix(ai-sdk): Log toModelOutput results for tools that define it When an AI SDK tool defines toModelOutput, its tool span and the parent span's tool results now log the value the model saw instead of the raw execute result. (#2652)
• fix(pi-coding-agent): preserve system prompts and tool declarations Capture system instructions and declared tool schemas from Pi 1.4 transcript messages, including section updates and tool additions and removals. Wrap refreshed tool executors from agent state and enable auto-instrumentation for Pi 1.x while preserving legacy context support. (#2673)
• fix(pi-coding-agent): avoid double-counting token usage and cost Keep token usage on child LLM spans instead of repeating it on AgentSession.prompt task spans, so trace rollups count each provider call once. (#2672)
• fix(instrumentation): Support TypeScript loaders (#2647)
• fix(livekit): Preserve native lifecycle spans (#2649)