Mobile app uses hardcoded model (chatgpt-5.5) instead of routing chat through tunnel to server

[akhildhanuka-euka]

Problem

The mobile app connects to the OpenCode server via tunnel (ngrok/cloudflare) and can read sessions (SSE works fine), but when sending a new message, the app uses its own cloud backend with a hardcoded model (chatgpt-5.5) instead of routing the chat request through the tunnel to the OpenCode server.

This means the server's configured model (e.g. openai/gpt-5.4, anthropic/claude-sonnet-4-6, etc.) is never used for mobile-originated messages.

The result: chatgpt-5.5 model not found error, since that model doesn't exist.

Expected behavior

When a tunnel is active and the OpenCode server is reachable, the mobile app should route chat requests through the tunnel to the server, which then uses the model configured in opencode.json. This way mobile and desktop use the same LLM.

Current behavior

• ✅ Session loading via tunnel: works
• ✅ SSE/push notifications: works
• ❌ Sending messages: uses app's own cloud backend with hardcoded chatgpt-5.5 model → fails with "model not found"

Environment

• opencode-mobile: v1.4.0
• Tunnel: ngrok with static domain
• OpenCode server: running on port 9020, confirmed healthy
• OS: macOS (Apple Silicon)

Suggestion

1. Add a setting in the mobile app to choose between "Cloud" and "Tunnel/Server" for chat routing
2. When tunnel is connected, default to routing chat through the tunnel
3. Optionally, add a model selector in the app settings

Or at minimum, allow users to configure which model the cloud backend uses.