The AI observability platform for tracing production, running evals, and catching regressions before they reach users.
This is the broad commercial model reported for Braintrust. Product-level terms are listed below when available.
Trust profile
Braintrust has updated its OpenTelemetry (OTel) ingestion process to prevent data duplication. Attributes previously parsed into structured fields (like input, output, and metrics) were historically duplicated in span metadata. The system now strips these parsed attributes from metadata by default to ensure each payload is stored only once. This update applies to Braintrust-hosted environments, while self-hosted customers can opt in via a configuration variable. Developers can override this behavior on a per-span basis for debugging purposes using the `braintrust.otel.preserve_attributes` attribute.
Braintrust now supports configurable data retention policies, allowing users to extend data retention windows and automate data deletion. Base retention windows apply organization-wide, with options for Pro and Enterprise plans to extend retention up to 180 or 365 days for an additional storage charge. Enterprise users can additionally set up retention automations to permanently delete specific object types—such as logs, experiments, or datasets—per project once they exceed a set age, aiding in compliance requirements.
Braintrust launched a centralized AI provider management page, enabling users to configure credentials at both the organization and project levels. The update supports advanced authentication methods, including workload identity federation for major providers such as OpenAI, Anthropic, Google Vertex AI, and Azure AI Foundry, as well as AWS AssumeRole for Bedrock. New governance features allow administrators to control project-level overrides and restrict permission groups, reducing reliance on long-lived API keys. Additionally, Braintrust now facilitates temporary access to the open-source model GLM-5.2.