Ask AI: Unable to create custom provider

[PumpkinJui]

Description

Although “Custom OpenAI-Compatible” displays as an option in the LLM provider list, it is unable to be used. Model name is grey and cannot be changed. An HTTP-500 error presents upon clicking the save button.

I understand if the dashboard or Ask AI are not part of this repo. If so, please let me know where I should file this issue instead.

Steps to reproduce

1. Login to the dashboard, go to “Data sources” and switch to “Ask AI”.
2. Click “Add assistant”. Choose “Custom OpenAI-Compatible” as the language model provider. After that, the system prompts to “select a custom provider” where the dropdown menu is empty. Then click “Create New” button beside it.
3. Set the name, base URL, etc. of the new custom provider. After that, the model name below is still grey and could not be specified. There is not a save button to save the new custom provider either. My settings: Name=SiliconFlow, Base URL=https://api.siliconflow.com/v1, and custom headers for authorization.
4. Enter a random name for the assistant and click “Save and exit” or “Save and add prompt”. The system notifys that it fails to create the custom provider, and that an error occurred with an HTTP request (500). UPDATE: This error will occur only when the custom headers is set. Without the headers, the assistant could be saved, but the model name could still not be specified, making it unusable.

Live reproduction:

I’m sorry, but doing so could leak my model API key. However, I can provide a screenshot for you.

Expected behavior

The custom provider should be saved. The model name should not be grey so I can specify it. And the assistant should be saved without any errors.

Environment

• OS: Windows
• Browser: Firefox 147.0.4, Ungoogled-Chromium 144.0.7559.96
• DocSearch version: v4.5.4

[jiangyangcreate]

i have a similar problem

Google Chrome 145.0.7632.111（64 ）

[vascobettencourt]

Hi @PumpkinJui ,

At the moment, Ask AI only supports model names that can be passed directly through the URL for custom OpenAI compatible providers. The dashboard does not yet support manually specifying or editing the model name field for custom providers, which is why it appears greyed out.

Regarding the HTTP 500 when custom headers are set, this is a known limitation in the current implementation of custom providers in the dashboard. Header based authentication is not fully supported yet, which is why the request fails when custom headers are included. Without headers, the provider can be created, but since the model name cannot be configured, it remains unusable for most external providers.

[Cp0204]

Same issue.

I don't know what kind of Provider (or which conditions must be met) will be recognized by Algolia as OpenAI-compatible. I've tried many Providers, even those that are fully compatible with OpenAI's /v1/chat/completions endpoint and offer models like gpt-4, gpt-5, etc., yet the Model Name field still cannot be edited.

This is extremely frustrating.