Braintrust provides flexible access control at multiple levels:
Organization level: Assign users to Owners, Engineers, or Viewers groups
Project level: Grant specific permissions to custom permission groups
Object level: Control access to individual experiments, datasets, or prompts
Create custom permission groups to match your team’s needs. Service accounts enable secure system integrations with granular permissions. See Control access for details.
Projects organize AI features in your application. Each project contains logs, experiments, datasets, and functions. Configure project settings including:
Tags: Organize and filter logs across your project
Human review scores: Define manual review criteria
Aggregate scores: Combine multiple metrics into single values
Online scoring: Automatically evaluate production logs
There are several ways to handle production vs. staging data:Use separate projects (recommended): Split production and staging into different projects so they’re isolated and code changes to staging cannot affect production. This also allows you to enforce access controls at the project level.Use tags within one project: If it’s easier to keep everything in one project (e.g., to triage issues in one place), use tags to separate environments. Filter by tags to view production or staging data independently.Use separate organizations: For physical isolation, create separate organizations for production and staging, each mapping to a different deployment.
Experiments, prompts, and playgrounds can use data across projects. For example, you can reference a prompt from your production project in your staging logs, or evaluate using a dataset from staging in a different project.
