Remote Embedding fails with CORS error when using local Ollama

Environment

• ReadAny version: v1.3.3
• OS: Windows 11
• Ollama version: 0.30.7

Problem Description

When configuring a Remote Embedding model pointing to a local Ollama instance, ReadAny reports "failed to fetch". The root cause is a CORS policy violation: ReadAny runs under the origin http://tauri.localhost, which Ollama does not allow by default.

Root Cause Analysis

Opening DevTools (via environment variable) revealed the actual error:

Access to fetch at 'http://localhost:11434/v1/embeddings' from origin
'http://tauri.localhost' has been blocked by CORS policy:
Response to preflight request doesn't pass access control check.

Ollama rejects the preflight OPTIONS request from http://tauri.localhost.
Even after setting OLLAMA_ORIGINS=*, a duplicate Access-Control-Allow-Origin header issue appeared (*, http://localhost:11434), which also caused the request to fail.

Workaround: Caddy Reverse Proxy

The following Caddy configuration resolves all three issues:

1. Handles OPTIONS preflight requests correctly
2. Rewrites the Origin header before forwarding to Ollama
3. Removes Ollama's own Access-Control-Allow-Origin header to prevent duplicates

Caddyfile:

{
    auto_https off
}

:11435 {
    @options method OPTIONS
    handle @options {
        header Access-Control-Allow-Origin "*"
        header Access-Control-Allow-Headers "*"
        header Access-Control-Allow-Methods "*"
        respond 204
    }

    reverse_proxy 127.0.0.1:11434 {
        header_up Origin "http://localhost:11435"
        header_up Host "localhost:11434"
        header_down -Access-Control-Allow-Origin
    }

    header Access-Control-Allow-Origin "*"
    header Access-Control-Allow-Headers "*"
    header Access-Control-Allow-Methods "*"
}

Then configure ReadAny with:

• API Endpoint: http://localhost:11435/v1/embeddings
• Model ID: bge-m3
• API Key: ollama (any value)

Suggestion

It would be great if ReadAny could handle this more gracefully, either by:

1. Adding OLLAMA_ORIGINS setup instructions to the documentation for local Ollama users
2. Using Tauri's native HTTP client (@tauri-apps/plugin-http) instead of browser fetch for embedding requests — this would bypass WebView2 CORS restrictions entirely and allow direct access to local services without a reverse proxy

Steps Others Can Follow to Reproduce

1. Install Ollama, pull any embedding model (e.g. ollama pull bge-m3)
2. In ReadAny v1.3.3, go to Settings → Vector Model → Remote Models → Add Model
3. Set APIendpoint to http://localhost:11434/v1/embeddings, model ID bge-m3
4. Click Save, and then click the Test button → "failed to fetch"
5. Open DevTools to see the actual CORS error