
Braintrust Data builds a platform for evaluating, observing, and shipping LLM applications and AI agents.
The platform is an active observability layer for AI agents, organized around three capabilities: Observe for real-time inspection of agent traces, Evaluate for evals with code, LLM, and human scoring, and Discover for automatic pattern discovery. Native SDKs cover Python, TypeScript, Go, Ruby, and C#, plus an MCP server for IDE integration.
Underneath sits Brainstore, a purpose-built trace database the company says is dramatically faster than general-purpose stores on large trace workloads. Deployment options include self-serve cloud, hybrid, and on-premises for enterprise.
Management frames 2026 as the year AI moved from experimentation to production, with agents embedded in nearly every engineering workflow. Traces are now long-running multi-step interactions that can produce hundreds of megabytes per interaction, outpacing traditional observability tooling.
The company positions AI observability as a category of core infrastructure in its Series B announcement. SiliconANGLE characterized the $80M round as Braintrust's push to become the observability layer for AI.
Named production customers include Notion, Stripe, Vercel, Cloudflare, Ramp, Dropbox, Replit, Box, and OpenAI. The company claims large query-speed gains from its purpose-built Brainstore trace database over general-purpose stores.
Compliance covers SOC 2 Type II, HIPAA, and GDPR with SSO and RBAC plus hybrid deployment. A free tier serves open-source and academic use, and the company published an open-source evaluation library with a published methodology in August 2026.
On May 4, 2026 the company discovered that hackers had accessed an internal AWS account, exposing org-level AI provider API keys stored by customers. All customers were emailed on May 5 to rotate keys after the incident.
At least one customer confirmed impact, with three reported suspicious AI-provider usage spikes. Nudge Security's CTO said potentially exposed keys belonged to customers including Box, Cloudflare, Dropbox, Notion, Ramp, and Stripe.
Pricing is self-serve SaaS with usage metering: Starter is $0/month with limited data and retention, Pro is $249/month with 5 GB data and RBAC, and Enterprise is custom with on-prem or hosted deployment. Metered units include model credits priced by token rate, processed data at roughly $3-4 per GB, and scores at $1.50-2.50 per 1,000.
Retention storage runs about $0.50 per GB per month. Qualifying startups receive 6-12 months free, and the platform is free for open-source and academic research.