v0.4.1: Fix Codex Streaming

🐛 Bug Fixes

Fixed Streaming Requests for Codex Models

This patch release fixes a critical issue where streaming requests to Codex models were using the wrong API endpoint, causing 404 errors and forcing fallback to secondary providers.

What was broken:

• ❌ Streaming requests to Codex models used /v1/chat/completions
• ❌ OpenAI API returned 404 errors
• ❌ System fell back to zenmux provider

What's fixed:

• ✅ Streaming requests now use /v1/responses endpoint
• ✅ No more 404 errors from OpenAI API
• ✅ All requests succeed with primary provider

📝 Changes

• Modified send_message_stream() to detect Codex models
• Route streaming requests to correct endpoint based on model type
• Added debug logging for streaming endpoint selection

🔧 Technical Details

Code Changes:

// Before: Always used /v1/chat/completions
let url = format!("{}/chat/completions", self.base_url);

// After: Smart routing based on model
let (url, request_body) = if is_codex {
    // Codex models → /v1/responses
} else {
    // Standard models → /v1/chat/completions
}

✅ Testing

• Verified streaming works with gpt-5.1-codex model
• Confirmed no 404 errors in production logs
• Both streaming and non-streaming requests work correctly

📦 Files Changed

• src/providers/openai.rs: +17 lines, -4 lines

---

🤖 Generated with Claude Code