Connect a model
Choose a provider for AI conversations and understand where your requests go.
Yenpo separates the local reading workspace from the model used for AI conversations. You can browse your code without a model connection and configure a provider when you want help exploring it.
Choose a provider
Open Settings and configure the provider and model you want to use. Yenpo supports OpenAI-compatible Chat Completions, OpenAI Responses, and Anthropic Messages API families. Choose the API family that matches your provider’s endpoint; similar model names do not imply the same protocol.
Enter the model identifier supplied by your provider. If the provider uses a custom endpoint, use the endpoint documented by that provider.
Authentication options depend on the provider. Enter API keys through Yenpo’s settings interface. Do not put credentials into a chat message, a repository file, or a screenshot you plan to share.
Understand the data flow
AI requests are processed by the model service you configure. A conversation may include your question, relevant code, file references, and other context selected for the request. Running the application locally does not make a remote model request local.
Review your provider’s data handling terms before using it with code that has access or confidentiality restrictions. Your organization may have a specific approved endpoint or model.
Images are disabled by default for each provider. If you enable them, only attach images you intend that model service to process.
Try a focused question
Open a small source file and ask a question whose answer you can verify:
What does this function return when the input is empty? Point to the relevant branch.
Check both the explanation and the referenced source. A successful response establishes that the model connection works; it does not establish that every answer will be correct.
If the request fails
Check the API family, endpoint, model identifier, and authentication settings. Confirm that your provider account can access that model and that the service is reachable from your network.
If a response is interrupted, inspect the conversation state before retrying. Do not assume the absence of a final answer means an authorized edit did not occur. See privacy and control for the distinction between reading, model requests, and workspace edits.