Braintrust API

aid: braintrust:braintrust-api
name: Braintrust API
description: The Braintrust REST API provides programmatic access to projects, experiments, datasets,
  prompts, functions, logs, and organization resources. It supports both US (api.braintrust.dev) and EU
  (api-eu.braintrust.dev) data planes, and is used to log production traces, run evaluations, manage prompt
  versions, and orchestrate scoring functions.
humanURL: https://www.braintrust.dev/docs
baseURL: https://api.braintrust.dev
tags:
- LLM
- Evaluation
- Observability
- Experiments
- Datasets
- Prompts
- Tracing
properties:
- type: Documentation
  url: https://www.braintrust.dev/docs
- type: GettingStarted
  url: https://www.braintrust.dev/docs/start
- type: Signup
  url: https://www.braintrust.dev/signup
- type: API
  url: https://www.braintrust.dev/docs/reference/api
- type: SDKs
  url: https://github.com/braintrustdata/braintrust-sdk
- type: SDKs
  url: https://pypi.org/project/braintrust/
- type: SDKs
  url: https://www.npmjs.com/package/braintrust
- type: GitHubRepository
  url: https://github.com/braintrustdata/autoevals
- type: Pricing
  url: https://www.braintrust.dev/pricing
- type: Authentication
  url: https://www.braintrust.dev/docs/reference/api
features:
- name: Experiments
  description: Create, run, and compare evaluation experiments across model and prompt versions to catch
    regressions before shipping.
- name: Datasets
  description: Manage versioned datasets of inputs, expected outputs, and metadata for repeatable evaluation
    runs.
- name: Production Logging
  description: Capture LLM spans, tool calls, and traces from production traffic via SDK or OTEL-compatible
    ingestion.
- name: Prompt Management
  description: Version, deploy, and A/B test prompts independent of application deploys.
- name: Autoevals and Custom Scorers
  description: Use built-in LLM-as-judge scorers or define custom Python and TypeScript scoring functions.
- name: Human Review
  description: Route traces and experiment runs to subject-matter experts for annotation and grading.
- name: Online Scoring
  description: Run scorers continuously against production logs to monitor drift and quality.
- name: Functions and Tools
  description: Register reusable tools, scorers, and workflows that can be invoked from prompts or experiments.
- name: Self-Hosting
  description: Deploy Braintrust inside your own AWS, GCP, or Azure account for data residency and compliance.
useCases:
- name: LLM Application Quality Gating
  description: Block deploys when evaluation scores regress against a golden dataset.
- name: Prompt Iteration
  description: Compare prompt and model variants side-by-side with traceable scores.
- name: Production Monitoring
  description: Detect hallucinations, latency spikes, and cost regressions in live AI traffic.
- name: Agent Evaluation
  description: Evaluate multi-step agent runs, tool calls, and retrieval performance.
- name: RAG Tuning
  description: Optimize retrieval and generation pipelines using dataset-driven experiments.
integrations:
- name: OpenAI
- name: Anthropic
- name: Google Gemini
- name: LangChain
- name: LlamaIndex
- name: Vercel AI SDK
- name: AutoGen
- name: CrewAI
- name: LangGraph
- name: Firebase Genkit
- name: OpenTelemetry
- name: AWS Bedrock
authentication:
- type: Bearer Token
  description: API keys and service tokens generated in the Braintrust dashboard are passed via the Authorization
    header as a Bearer token.