Skip to main content
Enabling Patterns creates a scheduled Loop automation that investigates your project and records what it finds. This page covers turning it on, tuning it, and adding focused automations alongside it.

​
Before you start

Patterns runs Loop against your project on a schedule, so two things need to be in place first:
  • A way to pay for inference. Patterns runs on Loop’s models. By default these are Braintrust’s built-in models, which draw down your organization’s monthly model credits and need no setup. On the Starter plan, they require at least one organization owner with a work email address, or a payment method on file. See Requirements. To use your own key or keep Patterns running after your credits are exhausted, configure an OpenAI-compatible AI provider. Patterns uses only organization-level providers during setup. If your provider is configured at the project level, add it at the organization level before enabling Patterns.
  • Permission to create automations in the project. Members of the Owners and Engineers permission groups have this by default. Enable patterns is shown to everyone, so without permission you’ll see an error when you click it rather than a hidden button.
Patterns chooses a provider during setup according to these rules:
  • When built-in models are available, Patterns selects them automatically. On Enterprise, you can choose between built-in models and your compatible configured providers instead.
  • When built-in models are unavailable, Patterns selects your only compatible configured provider automatically, or asks you to choose if you have several.
You can change the provider after setup in the automation’s configuration. If you want to choose a Slack notification channel while enabling, connect a Slack workspace first. Otherwise the picker doesn’t appear, and you can add a destination later instead.

​
Enable pattern discovery

  1. Go to Patterns. The Get started with Patterns panel guides you through setup.
  2. If the panel shows Add provider, configure an organization-level AI provider before continuing. See Common issues for provider and model credit requirements.
  3. If the panel shows Continue, click it to open Configure Patterns. Choose a provider under Select a model provider if that field appears. If a Slack workspace is connected, you can also choose a workspace and channel for notifications.
  4. Click Enable Patterns. If no provider choice or Slack configuration is needed, this button appears directly on the initial panel.
Braintrust creates a Pattern discovery automation that runs daily at 9:00 AM in your default timezone, queries the last 30 days by default, and has permission to create and update patterns. It runs a default instruction that tells Loop what to look for, which you can edit later. See Tune the automation. Braintrust then queues an initial run immediately, so you don’t wait for the first scheduled one. While a run is in progress, the page shows an Analyzing traces… indicator. Patterns appear as the run finishes.
If a project has no patterns yet and no run is in progress, click Analyze patterns to queue another run without waiting for the schedule.

​
Tune the automation

The default automation is an ordinary Loop automation, so you can change its instruction, schedule, model, query range, and destinations at any time. To edit it, either click Settings on the Patterns page and select the automation, or go to Settings > Automations. The fields worth revisiting first:
  • Frequency. How often the run happens. A project with steady traffic rarely needs more than daily.
  • Default query range. How much recent data each run inspects. Each run inspects the full range, not only the traces created since the previous run, so a wide range costs more.
  • Instruction. What Loop should look at, how deeply, and when to stop.
  • Destinations. Where run digests are delivered.
For the full field reference, see Create a Loop automation. When you resume a paused automation, Patterns keeps its saved model and provider unless you select a different one.
Keep the instruction framed around finding or updating patterns, even when you narrow the automation to one concern. An instruction that asks for a summary produces a summary, not a pattern record.

​
Add a focused automation

You can run several pattern automations alongside the default one. Use a focused automation when one concern needs its own scope or schedule: tool-call errors, cost growth, latency for a specific set of customers, or quality regressions. To add one, select Settings > New automation on the Patterns page. Braintrust opens a prefilled automation for you to review and save. All of them write into the same patterns list.

​
Common issues

When neither a compatible organization-level AI provider nor usable built-in models are available, the setup panel shows Add provider instead of Enable Patterns. Click Add provider to configure a provider without leaving setup.
  • Built-in models are disabled. To use them instead of your own provider, enable them in Settings > AI providers. Your organization must meet the built-in model requirements.
  • Your model credits are exhausted. Add your own provider key with Add provider, or select Upgrade to raise your plan’s limits. See model credits for how usage is counted.
Configuring a provider and enabling built-in models are organization-level actions. Only members of the Owners permission group, or a custom permission group with the Manage settings organization permission, can do either. Without that permission, ask an organization owner or administrator to resolve it.
Enabling Patterns does not test your provider’s connection before creating the automation. If a run fails, an error banner appears below the Patterns page header.Open the automation’s Past runs tab and select the failed run to inspect its transcript. For provider errors, check that the configured credentials are valid and the provider can serve the automation’s selected model. After resolving the error, run the automation again with Run now.
Loop records a pattern only when it can attach supporting evidence, so a quiet run is a normal outcome rather than a failure.To work out what happened, open the automation, select the Past runs tab, and select the run to read the transcript. Select Continue to fork it into an editable thread and ask Loop directly why it recorded nothing.If runs are consistently empty, the query range or the instruction is usually too narrow.
Destinations belong to automation runs. Patterns that Loop records during an interactive chat appear in the list but do not trigger Slack or webhook destinations.Confirm the destination is configured on the automation itself, under Destinations in the configuration panel, and that the Slack workspace is still connected.
Patterns requires data plane v2.13 or later. To upgrade, see the routine upgrade guide.

​
Next steps