Edge Function vision-ai v14 and vision-ai-v2 v1 both fail - app always shows "No AI provider is configured", no real AI reply

[prathmeshgudadhe95-byte]

Hi, I need help with Edge Functions on my project (ref: fduwpiblmmrqunmqctsi). Support ticket: SU-537892 (free plan, only an auto-acknowledgement so far).

Problem: my desktop app (Tauri + React, supabase-js) calls my Edge Function for AI chat. The app always shows "No AI provider is configured" and never gets a real AI reply. The POST to /functions/v1/vision-ai fails with 400 and no CORS header, status returns 0 providers.

What I already checked:

• User is logged in, session is present, so this is not a login problem.
• Edge Function Secrets have the correct key names set (no values shared here).
• OPTIONS preflight returns 200.
• I rewrote the function so createClient is lazy inside the handler, the whole handler is in try/catch, and every response (including errors) returns CORS headers. Deployed as vision-ai v14 — same result.
• I deployed the exact same code as a NEW function, vision-ai-v2 v1, and pointed the app at it. Fresh app build and reinstall — same result.

So both vision-ai v14 and vision-ai-v2 v1 fail the same way. I have not redeployed again and I have not deleted vision-ai-v2.

Note: I am testing a Cloudflare Worker as a workaround, but the app still shows no real provider reply.

Has anyone seen a 400 with no CORS header even after the CORS-on-every-error fix, on both an old and a brand-new function? What logs or checks should I look at next?

Thank you.

[g4lb]

CORS works: POST with no token returns your 401 JSON with CORS headers. The 400 likely occurs only with a signed-in request. Run it and read the body:

curl -i -X POST "$SUPABASE_URL/functions/v1/vision-ai" -H "Authorization: Bearer $ACCESS_TOKEN" -H "apikey: $ANON_KEY" -d '{}'

[fedyfsgerana]

I would try to isolate whether the 400 response is coming from the Edge Function handler itself or from the Supabase platform before making another deployment.

A few checks that may help:

1. Check Invocations and Logs separately. In Dashboard → Edge Functions → vision-ai → Invocations, look for the exact failed request. If there is no corresponding invocation, the request may be rejected before reaching your handler. If there is one, inspect the response body and function logs.

2. Inspect the actual error response. Check the response body and the sb-error-code header. A browser or Tauri frontend reporting status 0 does not necessarily mean the server returned HTTP 0; it can indicate that the client could not access the response because of a network or CORS failure.

3. Investigate the 0 providers result. Since both the original and newly deployed functions return the same provider count, I would verify that the function reads the expected environment variable names and that the provider-discovery logic recognizes them. Check the secret names and configuration paths, but never expose secret values in logs or screenshots.

4. Test the function independently of the desktop app. Use curl or another HTTP client with the same URL, Authorization header, apikey header, and request body. This can help distinguish a Tauri/WebView CORS issue from a server-side failure.

5. Check authentication separately. Supabase performs platform-level JWT verification before the handler executes when verify_jwt is enabled. Since the user has a valid session, verify that the actual request sends the user access token as Authorization: Bearer <access_token> and the publishable key in the apikey header. I would not disable JWT verification just to troubleshoot this.

One particularly useful detail would be whether the failed 400 requests appear in the function's Invocations view.

If they do not appear, I would investigate the gateway, function metadata, and platform-level authentication. If they do appear, I would focus on the handler's provider configuration and error path.

Has anyone encountered this exact combination: a 400 response without CORS headers, no configured AI providers reported by the handler, and identical behavior after deploying the same code under a new function name?

Thanks!