AI Assistant returns 404 "Service Request Failed" with Custom AI Provider (BYOK, Mistral) — other AI features work

1

Hi,

I’m using the Custom AI Provider (BYOK) option to connect eM Client to the Mistral API, but the AI Assistant fails with a 404 error while all other AI features work fine. I believe this is a bug in the AI Assistant’s request handling, not in my configuration.

My setup

  • eM Client 11.0.1267 (Windows)
  • Settings → Artificial Intelligence → General → Custom AI Provider
  • URL: https://api.mistral.ai/v1
  • API key: my own Mistral API key (valid, works outside eM Client)
  • Primary model: mistral-small-latest
  • Models for AI Assistant: mistral-large-latest (also tested with mistral-small-latest)

What works

  • “Test connection” succeeds (“The connection to the custom AI provider was successful”).
  • AI Summarize, Reply with AI and the AI writing tools in the compose window all work correctly through the custom provider.

What fails

  • Any prompt sent to the AI Assistant in the sidebar returns: Service Request Failed. Status 404 (Not Found).
  • This happens with both mistral-large-latest and mistral-small-latest set as the Assistant model, so it is model-independent.
  • Restarting eM Client does not help.
  • All models I’m using are current and generally available on Mistral’s API (mistral-large-latest points to Mistral Large 3, GA).

My analysis

Since the same key, URL and models work fine in the other AI features (which use the standard OpenAI-style POST /v1/chat/completions endpoint that Mistral serves), the AI Assistant appears to be calling a different endpoint or API shape that Mistral’s OpenAI-compatible API does not expose — which would explain the consistent 404 while everything else works.

Questions

  1. Does the AI Assistant support Custom AI Providers (BYOK) at all, or is it limited to the built-in provider?
  2. If it is supposed to work: is there a specific endpoint or API format it requires, and could the 404 be caused by the Assistant calling something other than /v1/chat/completions?

Happy to provide logs if you tell me which logging to enable.

2

Quick follow-up with technical details that pinpoint the cause.

I enabled the AI Assistant logging option (Settings → Advanced → Logging), and the log shows exactly where the request goes:

POST https://api.mistral.ai/v1/responses

So the AI Assistant sends its requests to the OpenAI Responses API endpoint (/v1/responses), built on Microsoft.Agents.AI / the OpenAI .NET SDK (User-Agent: OpenAI/2.13.0 (.NET 10.0.13; Microsoft Windows 10.0.26300),MEAI/10.10.0).

Mistral does not serve a /v1/responses endpoint — their API is OpenAI-compatible only at the /v1/chat/completions level. A request to a non-existent path returns 404 Not Found, which matches the error and explains why it happens with every model I tried (mistral-small-latest and mistral-large-latest).

The other AI features (Summarize, Reply with AI, compose) use /v1/chat/completions and work fine with the same provider URL and key.

To reproduce:

  • eM Client 11.0.1267 (Windows)
  • Custom AI Provider (BYOK): https://api.mistral.ai/v1 + Mistral API key
  • Connection test: OK — Summarize / Reply with AI / compose: OK
  • AI Assistant: “Service Request Failed. Status 404 (Not Found)” on every prompt

It looks like the fix would be for the Assistant to fall back to the chat completions endpoint for providers that don’t support the Responses API.

Happy to share the full AIChat.log if useful — the API key is redacted in it, so it’s safe to post.

3

Thank you for your effort. We are fully aware of that and we’ve already fixed that. The fix will be included in future update just a few days from now.