Braintrust is an observability and evaluation platform for AI agents. Teams trace prompts, responses and tool calls in real time, track latency, cost and quality, run experiments on datasets, and score outputs with language models, code or human reviewers. Production traces can be turned into evaluation datasets with one click. SDKs are available for Python, TypeScript, Go, Ruby and C#, among others, and an MCP server lets developers query logs and run evaluations from an IDE. The platform is not open source.
Braintrust Data, Inc. is a US company. Besides the managed cloud service, Braintrust offers a hybrid model: the data plane holding logs, datasets and evaluation results runs in your own AWS, GCP or Azure environment, while Braintrust hosts the web application and metadata. According to the vendor, browser requests go directly to your data plane. A fully self-hosted variant is also available for organisations that need complete isolation. For a European organisation, this choice determines where data resides; the vendor itself is subject to US jurisdiction.