Sitelet https://www.braintrust.dev/docs/loop
Skip to main content
Loop is Braintrust’s AI agent for investigating logs, experiments, and datasets across your project. It can also create and edit prompts, scorers, datasets, facets, custom views, dashboards, and automations. Loop also powers Patterns and the Debugger, which investigate on your behalf without a chat.

​
Start a thread

Loop runs in a Braintrust-managed runtime, supporting long investigations and saving threads for you to return to. You can start without configuring an AI provider: Loop uses built-in models billed against your organization’s monthly model credits.
On the Starter plan, the built-in models require at least one organization owner with a work email address, or a payment method on file. See Requirements.
1

Open Loop

Go to Loop to start a thread or return to a saved one. Starting from the prompt box on the project overview page also opens your thread on the Loop page.You can also open Loop from the logs table, SQL sandbox, trace view, playgrounds, datasets, individual dashboards, and the chart editor. Each opens the same agent with the page’s data in context. See What Loop can do.
2

Choose a model and reasoning effort

Use the model picker to choose a model, reasoning effort, and provider.
3

Add context and attach files

Type @ followed by a name to reference a dataset, experiment, project log, playground, or SQL query. When other context is available, Add context opens the same list to browse and search.Select Attach files or drag files into the composer. Loop accepts images, audio, video, PDF, text, and JSON, with these limits:
  • 8 files per message.
  • 20 MB per file.
  • 40 MB per message.
4

Ask, approve, and iterate

Type a question and press Enter. Loop runs read-only work without interrupting you and pauses for approval before it creates or edits anything.Type / at the start of your message or after whitespace to see available commands. Depending on where you opened Loop, you can bootstrap a scorer, summarize a dataset, or file a support ticket.

​
Choose a provider and model

New threads start on GPT-6.1 Sol at medium reasoning effort, Braintrust’s recommended setting. Use the model picker to choose:
  • Model. GPT-6 Astra, GPT-6 Luna, GPT-6.1 Sol, GPT-6 Sol, GPT-5.6 Sol, GPT-5.6 Terra, or GPT-5.6 Luna.
  • Effort. Available levels depend on the model, and the menu marks its default. Loop remembers your choice for each thread.
  • Provider. Select Braintrust to use model credits, or a configured provider to bill inference to that provider. Supported providers include OpenAI, Azure, Bedrock, and OpenRouter. Loop remembers your choice for each thread and uses your most recent choice for new threads. See Models and providers for the full list.
Provider access and usage depend on your organization’s settings and credits:
  • Access. Allow built-in models on the AI providers page controls whether members can select Braintrust.
  • Credits. When model credits run out, Braintrust becomes unavailable and a banner above the message box offers an upgrade.
  • Logging. Loop logging can be turned off. See Privacy.
To monitor usage, turn on Show turn metrics in settings. The footer shows duration, time to first token, input and output token counts, cache hit rate, and estimated turn and total cost.

​
Approve changes

Before creating or modifying a Braintrust object, Loop describes the action and asks for your approval. For external MCP tools, approval depends on the server’s tool annotations and your approval settings. In a shared thread, Loop accesses Braintrust using the sender’s permissions. Only the sender can approve or cancel the work their message starts. To skip approvals for tools that support it, turn on Auto-accept edits in Settings. Some actions still require explicit approval.
Auto-accept can allow production changes, such as retagging an environment or creating an online scoring rule. Leave it off if you need to review those changes first.

​
Review results in the workspace

Results open in a workspace beside the chat. Keep several objects open in tabs while Loop works, including patterns, prompts, scorers, facets, preprocessors, dashboards, datasets, traces, automations, and experiment runs. You can open anything Loop creates, edits, or references. Hold Cmd, Ctrl, or Shift while selecting an object to open it in a new browser tab instead.

​
What Loop can’t do

Loop accesses Braintrust one project at a time, within your existing UI and API permissions. External MCP calls use the project’s shared credentials. Loop cannot:
  • Reach another Braintrust organization’s data. A thread is scoped to the project you opened it in, and Loop resolves objects only within your organization.
  • Delete Braintrust objects. Loop can create and update prompts, scorers, facets, preprocessors, datasets, views, and automations, but cannot delete them. It can delete dataset rows or close a pattern, marking it inactive rather than removing it.
  • Change organization settings. Loop cannot manage members, permission groups, API keys, AI providers, or billing. It can list and create environments for dataset versioning, but cannot delete them.
  • Access someone else’s private threads. Threads are private by default. Owners can share threads with project members.
  • Run tools that require approval in an interactive thread unless you approve them or enable auto-accept. Scheduled runs use separate permission rules.
Connected MCP tools can act on external systems within the shared credential’s permissions, including making changes if the server exposes tools for them.

​
Next steps