← All companies

Agents / Company profile

Braintrust

What Braintrust does

Braintrust (Braintrust Data, Inc.) is an AI observability and evaluation platform for teams building LLM-enabled applications and AI agents. The company’s core idea is that agent/LLM behavior is non-deterministic and changes over time, so teams need an end-to-end workflow that (1) captures real production behavior (traces, prompts/responses, and tool calls), (2) defines what “good” means via evaluations (including versioned datasets and scoring), and (3) turns production signals into continuous improvements and release gates. Braintrust positions its system as “active observability” for agents and emphasizes capabilities designed for long-running, multi-step traces rather than only single-shot prompt/response testing.

In product terms, Braintrust is presented as a platform with three main pillars—Observe, Evaluate, and Discover—plus supporting developer workflows and data infrastructure. “Observe” focuses on ingesting and inspecting agent traces and instrumenting applications so teams can see what actually happened in production. “Evaluate” centers on running experiments and scoring outputs using LLMs, code, or humans, alongside versioned datasets and comparisons across prompts/models. “Discover” focuses on turning production patterns into evals/automation (including continuous online scoring and quality gates) so teams can block regressions before they ship. Additional platform components highlighted by Braintrust include Loop (to generate prompts/scorers/datasets), Topics (automated pattern discovery), an MCP server for connecting coding agents/IDE workflows, and Brainstore, a database designed for querying large, nested agent traces.

Business model: Braintrust sells an enterprise developer platform (with plans referenced publicly in its content and program pages, and with security/compliance and deployment options described as selling points). It serves cross-functional engineering and product teams who need measurable reliability for LLM features and agents in production. Examples of customers and partner ecosystems referenced by Braintrust include teams like Notion, Stripe, Cloudflare, Zapier, Airtable, and others (cited on its website and in funding/feature posts).

News

Company record

Aug 05, 2026 · Official · BraintrustTrace and improve Cloudflare Agents

Braintrust describes how Cloudflare launched native agent tracing built on OpenTelemetry, and how teams can route those traces to Braintrust for behavior inspection and evaluation over time.

Jul 29, 2026 · Official · BraintrustBehavior specs, an open standard for supervising long-horizon agents

Braintrust and Basis release “behavior specs” as an open standard for defining and evaluating intended agent behavior across long trajectories, with an open source entry point for judging behavior specs on traces.

Jul 07, 2026 · Official · BraintrustFaster phrase search with shingled bloom filters in Brainstore

Braintrust engineering post describing improvements to Brainstore’s phrase-search performance for agent debugging, including an approach using shingled bloom filters and reported efficiency gains.

Jun 30, 2026 · Official · BraintrustAgent observability for startups

Braintrust launches a “Braintrust for Startups” program designed to provide early-stage teams access to the platform (including connecting evals and observability in one workflow), plus program benefits like access duration and credits.

Jun 26, 2026 · Official · BraintrustHow to eval stateful agents

Braintrust publishes a best-practices guide explaining why stateful agents require different evaluation approaches than single-turn prompt-and-response evals, and describes workflows involving instrumentation, trace-based evaluation, and dataset promotion.

Jun 04, 2026 · Official · BraintrustHow we made continuous trace intelligence possible at scale

Braintrust engineering post describing architecture for Topics-style continuous trace intelligence at scale, including requirements like minimum trace counts and unique facet summaries for topic map generation.

Jun 01, 2026 · Official · BraintrustAutomate pattern discovery with Topics, now generally available

Braintrust announces Topics as generally available, describing how it works within the Braintrust workflow to filter traces by topic and run experiments against surfaced failure modes.

Jun 01, 2026 · Official · BraintrustAI observability is active observability

Braintrust explains the shift from passive monitoring to active observability for AI/agents, emphasizing topic clustering at scale limitations and the “active observability” framing.

Show 3 earlier updates
Source map · 2 recurring channels · 15 references

Still resolving: Newsroom · Official X

Funding

Latest disclosed valuation$800M
Tracked capital$121.1M3 sourced rounds

What it builds

Braintrust (Observe / Evaluate / Discover)

Braintrust is positioned as an end-to-end “active observability” platform for agents, combining agent trace observability (Observe), evaluation and experimentation workflows (Evaluate), and automated pattern discovery and release-quality gates (Discover).

source ↗
Loop

Loop is described as an AI workflow that helps teams improve agents by generating prompts, scorers, and datasets automatically based on what they want to optimize.

source ↗
Topics

Topics is described as an automated pattern-discovery capability for AI traces, intended to surface patterns and help teams turn them into evals and improvements.

source ↗
MCP server (Braintrust MCP)

Braintrust’s MCP server connects coding agents to the user’s AI stack so they can query logs, run evals, and update prompts from the IDE environment.

source ↗
Brainstore

Brainstore is described as the database built for AI data at scale, designed specifically for querying large, nested agent traces efficiently.

source ↗
Braintrust CLI

Braintrust’s CLI (“bt”) is presented as a terminal interface for running evals and inspecting results (including a workflow described for agentic debugging loops around eval failures).

source ↗

Key metrics

Milestones & partnerships