Manage Loop threads and limits

Loop threads persist, so you can leave a page and come back to the work in progress. This page covers cleaning them up, how sandboxes behave, and how many threads can be active at once.

​

View thread history

When you open a long thread, Loop initially loads its recent history. To view earlier messages, scroll to the top of the conversation and click Load earlier messages when available. Wait for Loop to finish working before loading earlier messages.

​

Rename a thread

On the Loop page, select the menu on the thread header or on any thread in the sidebar, select Rename thread, enter a name of up to 256 characters, then click Save.

​

Delete a thread

On the Loop page, select the menu on the thread header or on any thread in the sidebar, select Delete thread, then confirm. You can’t delete a thread while Loop is actively working in it. A thread whose sandbox has gone to sleep isn’t running, so you can delete it. Automation run threads are read-only and can’t be deleted this way. See Review past runs.

​

Sandboxes

Loop runs in an isolated, per-user sandbox, which is what lets it inspect trace files directly and run commands against your data. When a sandbox starts. Loop starts one the first time you send a message in a thread, not when you open the Loop page or revisit a saved thread. When a sandbox sleeps. After a period without activity, the sandbox goes to sleep. Send another message in that thread and it resumes automatically, with the thread’s history intact. When a sandbox reaches its maximum lifetime, Loop starts a fresh one for your next message. Your thread and its history are unaffected. A sandbox has a maximum lifetime of 8 hours, including time spent asleep.

​

Thread concurrency

Braintrust-hosted Loop limits how many threads can be active at the same time. A thread counts as active while Loop is working in it and for a short idle period afterward. The limit applies separately to each member, service token, and automation within each organization. Sending another message in a thread that is already active does not consume another slot. Starting work in another thread while at the limit returns an HTTP 429 error. Wait for an active thread to release its slot, then try again. Existing threads and their histories are unaffected.

​

Models and providers

For any of Loop’s models, the Provider setting in the model picker decides who pays for the inference:

• Braintrust. Braintrust serves the model from its own AI providers, and usage draws down your organization’s monthly model credits, shared with Topics. No setup is required. On the Starter plan, this option requires at least one organization owner with a work email address, or a payment method on file. See Requirements.
• Your configured provider. The model runs through one of your organization’s own OpenAI-compatible AI providers, and inference is billed to that provider.

Loop remembers the selected provider for each thread and applies your most recent choice to new threads. Compatible provider types are OpenAI, Azure, Bedrock, OpenRouter, and Azure AI Gateway. Your configured provider must serve the Loop model you select. You need one of these once your model credits are exhausted. To configure a custom endpoint from Loop, click Add provider in the model picker, then click New under Custom providers. The form offers compatible provider types. Adding a custom endpoint does not expand Loop’s supported models. To keep members off the built-in models, turn Allow built-in models off on the AI providers page. Loop then runs only on your organization’s own providers, and Braintrust does not appear as a provider option.

​

Privacy

Threads are private to the person who created them. Loop cannot see another user’s threads, and sharing a thread with a teammate is not available. By default, Braintrust logs frontend Loop interactions for product improvement and support. Loop works normally with this turned off, and self-hosted organizations have it off by default. See Control Loop logging. When you select your own provider, Loop calls the model through that AI provider, using your organization’s own key. Your agreement with that provider, including any zero data retention agreement, governs the prompt and the response. Loop automations and Patterns use the model and provider saved in each automation’s configuration. See Data handling. Topics is separate. Facet summarization, embeddings, and cluster naming always run on Braintrust-served models with Zero Data Retention, whatever model you pick in Loop.

​

Next steps

• What Loop can do for the full range of tasks.
• Run Loop on a schedule for work that shouldn’t depend on an open chat.
• Plans and limits for the rest of the system limits.