Droid Configuration - Lessons Learned

Overview

Configuration of Factory Droid to work with Claude Max, Codex, and Antigravity models via CLIProxyAPI.

Key Lessons

1. Port Configuration

• CLIProxyAPI runs on port 8317 (not 8318)
• Initially configured port 8317 correctly, then accidentally changed to 8318
• Always verify which port the proxy server is actually listening on using lsof -i :<port>

2. Model Discovery

• CLIProxyAPI provides a /v1/models endpoint to list all available models
• Use curl http://127.0.0.1:8317/v1/models | jq to see all available models
• Not all models listed in the API work with authentication

3. Authentication Methods

• Claude: Uses /v1/messages endpoint with x-api-key header and dummy key
• GPT/Codex/Gemini: Uses /v1/chat/completions endpoint with Authorization: Bearer header and dummy key
• CLIProxyAPI handles OAuth token management automatically when configured with --claude-login or similar auth commands

4. Model Testing Strategy

Test models individually before adding to config:

# Claude models
curl -X POST http://127.0.0.1:8317/v1/messages \
  -H "Content-Type: application/json" \
  -H "x-api-key: dummy" \
  -H "anthropic-version: 2023-06-01" \
  -d '{"model": "claude-opus-4-5-20251101", "max_tokens": 10, "messages": [{"role": "user", "content": "hi"}]}'

# GPT/Codex/Gemini models
curl -X POST http://127.0.0.1:8317/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer dummy" \
  -d '{"model": "gpt-5.2", "messages": [{"role": "user", "content": "hi"}], "max_tokens": 10}'

5. Working Models (as of 2026-01-01)

Claude Models (Anthropic)

• claude-opus-4-5-20251101 ✓
• claude-sonnet-4-5-20250929 ✓
• claude-haiku-4-5-20251001 ✓
• claude-3-7-sonnet-20250219 ✓
• claude-opus-4-1-20250805 ✓

GPT/Codex Models (OpenAI)

• gpt-5.2 ✓
• gpt-5.2-codex ✓
• gpt-5.1-codex-max ✓
• gpt-5-codex-mini ✓
• gpt-5-mini ✓
• gpt-4.1 ✓

Gemini/Antigravity Models

• gemini-2.5-flash ✓
• gemini-claude-opus-4-5-thinking ✓

Other Models

• grok-code-fast-1 ✓

6. Non-Working Models

• antigravity-1 ✗ Unknown provider (token exists but not mapped)
• antigravity-1-preview ✗ Unknown provider
• gemini-3-pro ✗ Model not supported
• raptor-mini ✗ Model not supported
• All -high, -medium, -low variants ✗ Unknown provider

7. Droid Configuration File

Location: ~/.factory/config.json

Example configuration:

{
    "custom_models": [
        {
            "model": "claude-opus-4-5-20251101",
            "base_url": "http://127.0.0.1:8317",
            "api_key": "sk-dummy",
            "provider": "anthropic"
        },
        {
            "model": "gpt-5.2",
            "base_url": "http://127.0.0.1:8317/v1",
            "api_key": "sk-dummy",
            "provider": "openai"
        }
    ]
}

Important notes:

• Claude models use provider "anthropic" without /v1 in base_url
• All other models use provider "openai" with /v1 in base_url
• api_key is required by Factory but ignored by CLIProxyAPI (use "sk-dummy")
• Validate JSON with python3 -m json.tool ~/.factory/config.json

8. Common Issues

Issue: Models not showing in Droid

Solution: Restart Droid completely (not just tmux session). Droid caches config on startup.

Issue: "402 status code (no body)" error

Solution: This is a Factory billing error, not a configuration issue. Check your Factory.ai subscription.

Issue: "Unknown provider for model" error

Solution: The model name is not mapped to a provider in CLIProxyAPI authentication. Check available models via /v1/models endpoint.

Issue: "OAuth token has expired" error

Solution: Re-authenticate with CLIProxyAPI:

/Applications/VibeProxy.app/Contents/Resources/cli-proxy-api-plus --claude-login

9. Testing Workflow

1. List available models from API
2. Test each model individually via curl
3. Add only working models to config.json
4. Validate JSON syntax
5. Restart Droid completely
6. Use /model command to select model in Droid

10. CLIProxyAPI Management

• Process name: cli-proxy-api-plus
• Running from: /Applications/VibeProxy.app/Contents/Resources/
• Auth tokens stored in: ~/.cli-proxy-api/
• Config file: /Applications/VibeProxy.app/Contents/Resources/config.yaml

View running processes:

ps aux | grep cli-proxy
lsof -i :8317

Verification Commands

# Check proxy is running
lsof -i :8317

# List all available models
curl -s http://127.0.0.1:8317/v1/models | python3 -m json.tool

# Validate config JSON
python3 -m json.tool ~/.factory/config.json > /dev/null

# Test a model
curl -s -X POST http://127.0.0.1:8317/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer dummy" \
  -d '{"model": "gpt-5.2", "messages": [{"role": "user", "content": "test"}], "max_tokens": 5}'

References

• CLIProxyAPI GitHub: https://github.com/luispater/CLIProxyAPI
• Factory Droid: https://factory.ai/
• VibeProxy (CLIProxyAPI wrapper): /Applications/VibeProxy.app/