Applies to:
- Plan - Any
- Deployment - Any
- Use case - Configuring a custom AI provider when the prompt chat fails due to a malformed upstream URL
Summary
Issue: Chat with a prompt in a Conversation project fails when the AI provider’s base URL includes the/chat/completions path suffix, resulting in a duplicated upstream path like /v1/chat/completions/chat/completions that OpenAI rejects.
Cause: Braintrust automatically appends /chat/completions when constructing the upstream request URL, so including it in the base URL causes the path to be doubled.
Resolution: Remove /chat/completions from the provider’s base URL and retry.
Resolution steps
Step 1: Open the provider settings
Go to Settings > AI providers and open the provider backing the model (for example,GPT-5.2 Chat Provider). Select Setup and expand More settings if it is collapsed.
Step 2: Fix the base URL
Under Request configuration, update the API base URL field so it ends before/chat/completions.
For standard OpenAI providers, you can also leave the base URL field blank to use the default.