Braintrust JS SDK v3.5.0: Agent Tracing and Experiment Parameters

Here at industry sources, we tracked the release of Braintrust's JavaScript SDK v3.5.0, which introduces two focused capabilities for teams building with AI. First, the SDK now supports Agent tool call tracing for AI SDK v5 and v6 with auto instrumentation - meaning you get visibility into how your agents are calling external tools without manual integration work. Second, you can now attach saved parameters directly to experiments, reducing the friction between testing configurations and deployment.

These aren't flashy features, but they address real friction points in the AI development workflow. Tool call tracing gives you the observability layer that's essential when debugging agent behavior at scale. Parameter attachment simplifies experiment management by letting you bundle configuration with test results, making reproduction and iteration faster.

• Agent tool call tracing for AI SDK v5/v6 with auto instrumentation enabled
• Ability to attach saved parameters to experiments
• Improved observability without manual instrumentation overhead
• Tighter integration between experimentation and production configuration

Tool call tracing is the foundation of trustworthy agent systems. When your agent makes decisions by calling external tools - APIs, databases, search engines - you need to know what it's actually doing. The auto instrumentation approach here saves you from hand-rolling observability code, which is where most teams lose momentum. You get traces for free, which means faster debugging and better confidence in production behavior.

The saved parameters feature addresses a specific workflow problem: experiment results are useless if you can't remember or reproduce the exact configuration that produced them. By coupling parameters to experiments, Braintrust makes it easier to track what you've tested and compare configurations side-by-side. This is particularly valuable for teams iterating rapidly on prompt engineering, model selection, or tool selection.

For builders using AI SDK, this release tightens the integration between your development framework and your observability/experimentation platform. Less glue code means faster iteration and fewer points of failure.

• Reduces debugging time by providing automatic visibility into agent tool calls
• Eliminates manual instrumentation overhead - auto tracing is enabled by default
• Improves experiment reproducibility by bundling configuration with results
• Strengthens the AI SDK ecosystem integration for standardized development workflows

If you're running agents with AI SDK v5 or v6, update to v3.5.0 immediately. The auto instrumentation means you get Agent tracing without refactoring. Spend 30 minutes confirming that traces are flowing into Braintrust - check your dashboard to see tool calls appearing. This is low-risk, high-value work.

For teams doing active experimentation, the saved parameters feature is worth integrating into your workflow. Document which experiments are candidates for production, and ensure you're capturing parameters for all variations you want to track. This compounds the value of your test suite over time.

Audit your current observability setup. If you're missing visibility into what tools your agents are calling, this update addresses that gap. If you're already instrumenting manually, you can simplify. The momentum in this space continues to accelerate.

• Upgrade to v3.5.0 and verify Agent traces appear in your Braintrust dashboard within 30 minutes
• Adopt saved parameters for all experiments you intend to evaluate for production deployment
• Audit gaps in your current agent observability and use auto tracing to close them
• Document parameter strategies so team members know what to capture and why