Tool profile
Braintrust: cost, fit and limits
Braintrust is an evaluation and observability platform made by Braintrust Data, Inc. in San Francisco that stores prompts as versioned functions next to datasets, scorers and experiments. Developers load prompts through TypeScript, Python, Go or Ruby SDKs, or call them over an HTTP endpoint.
What to know first
Braintrust makes the most sense when prompt changes should pass an experiment before release. Prompts, tools and scorers sit in one project, and unlimited users come with every plan. The free Starter plan keeps data for 14 days and includes 1 GB of processed data. Pro costs $249 a month for 5 GB and 30 days. The meter is data and scores, not requests.
Braintrust at a glance
| Maker | Braintrust Data, Inc., United States |
|---|---|
| Free plan | Yes |
| Entry price | $249 per month (Pro) |
| API access | Yes |
| Category | Best AI Prompt Management Software in 2026, rank 4 of 10 |
| Website | braintrust.dev |
Who Braintrust is best for
Teams that gate releases on scores
Starter includes 10K scores and Pro 50K, so a change to a prompt can be scored against a dataset before the environment points at the new version.
Polyglot engineering teams
Prompts load by pinned version or environment from TypeScript, Python and Go, and Ruby can load and build them locally, so a mixed stack shares one registry.
Companies that must keep data in their cloud
The docs describe a bring-your-own-cloud option where the data plane runs in your account while Braintrust manages the interface. Control of the key store stays with the customer.
Startups on a budget
The pricing page says Pro is free for 6 to 12 months for qualifying startups, which removes the $249 step during early evaluation.
Braintrust plans and pricing
| Plan | Price | What it includes |
|---|---|---|
| Starter | $0 per month | $10 in model credits, 1 GB processed data then $4 per GB, 10K scores then $2.50 per 1,000, 14-day retention, unlimited users. |
| Pro | $249 per month | $100 in credits, 5 GB processed data then $3 per GB, 50K scores then $1.50 per 1,000, 30-day retention then $0.50 per GB monthly, RBAC listed. |
| Enterprise | Custom | Custom data, scores and retention, SAML SSO, custom RBAC, and hosted or on-prem deployment for high-volume or privacy-sensitive data. |
Checked October 2026.
What Braintrust does
- Versions and environments
- A prompt can be pinned to a version or loaded for an environment, and a CLI command assigns a version to an environment, which is how a release is done.
- HTTP invocation
- Prompts can be invoked through a function endpoint over REST with the same parameters as the SDKs, including versioning, environments and streaming. That lets non-Python services call a prompt directly.
- Tools attached to prompts
- When a prompt includes tools, the response contains tool calls that your code handles. Tools can be deployed alongside the prompt as functions. Evaluating tool calls therefore needs code in your application.
- Template variables
- Values passed in the input fill placeholders in the prompt text, so the same prompt serves many inputs and test cases. A test dataset can feed many inputs through one stored prompt.
- Scheduled and on-demand export
- Data can be exported to cloud storage on a schedule as JSON Lines or Parquet, or on demand through the API, which keeps an exit route open.
How Braintrust handles your data
Braintrust's security documentation says it is SOC 2 Type II compliant with reports available through its Trust Center after a mutual NDA, can sign HIPAA BAAs and DPAs, and encrypts data at rest and in transit, with AES-256 for provider API keys. It describes SaaS and bring-your-own-cloud options and recommends BYOC with EU residency for full GDPR needs. Retention policies can delete logs, experiments or datasets after a chosen period. SaaS hosting regions and whether customer data trains models are not stated, and the privacy notice excludes customer-uploaded content.
Where Braintrust falls short in practice
- Short retention by default. Starter keeps data 14 days and Pro 30 days, so a comparison of prompt versions across a quarter means paying $0.50 per GB monthly for longer storage.
- Three meters to forecast. Platform fee, processed data and scores each bill separately, so a team that scores every response can pass the Pro $249 floor and pay more in score charges.
- SSO is not on Pro. The pricing table shows SAML SSO as not included on Pro, so identity-provider rules force an Enterprise conversation even for a small team. Plan on Enterprise if single sign-on is a hard requirement.
- Prompts are not the product's center. The docs page read covers loading and invoking, and says nothing about the playground. Teams that want a prompt editor for non-engineers should test the interface first.
Works with
Alternatives to Braintrust
Braintrust: common questions
How do Braintrust environments release a prompt?
A prompt version is assigned to an environment through the bt prompts CLI, and application code loads whichever version that environment points to. Code can also pin a specific version. Both paths work through SDKs and the REST endpoint, with versioning and streaming supported.
What does the free Braintrust Starter plan allow?
Starter costs $0 and includes $10 in model credits, 1 GB of processed data, 10K scores, 14-day retention and unlimited users, projects, datasets, playgrounds and experiments. Overage is $4 per GB and $2.50 per 1,000 scores. It is hosted only, with no SSO listed.
Can Braintrust keep data in the EU?
The security docs do not list SaaS regions. They say full GDPR needs are best served by a bring-your-own-cloud deployment with EU data residency, where the data plane runs in your own account while Braintrust keeps the managed interface. The pricing page lists on-prem or hosted deployment under Enterprise.
Which languages can load Braintrust prompts?
TypeScript, Python and Go support both invoking and loading prompts. Ruby supports loading and building prompts locally but not server-side invocation. Anything else can reach the function endpoint over HTTP, which accepts the same parameters as the SDKs. Check the SDK page for the language you use.
Does Braintrust train on my prompts or logs?
The security documentation and the privacy notice read for this page do not say. The privacy notice states it does not apply to content customers upload to the service. Ask for a written answer on model training before sending production logs.