Skip to main content

Droid

Droid is Friday Dev's flexible execution backend that supports multiple AI model providers.

Overview​

Droid provides:

  • Multi-model support - Connect any compatible model
  • Unified interface - Same tools across all models
  • Flexible autonomy - Configure permissions per model
  • Custom profiles - Create your own agent configurations

When to Use Droid​

Ideal For​

  • ✅ Custom model configurations
  • ✅ Self-hosted models
  • ✅ Experimental models
  • ✅ Specific autonomy needs
  • ✅ Cost optimization

Less Ideal For​

  • ⚠️ Quick start (use preset agents)
  • ⚠️ Beginners (more configuration)

How Droid Works​

┌─────────────────────────────────────────────────────────────────┐
│                         Friday Dev                               │
├─────────────────────────────────────────────────────────────────┤
│                                                                  │
│     Task → Droid Executor → AI Model → Tools → Output           │
│                    │                      │                      │
│                    ▼                      ▼                      │
│              ┌─────────────┐       ┌─────────────┐              │
│              │ Model API   │       │ File System │              │
│              │ (Any)       │       │ Git         │              │
│              │             │       │ Terminal    │              │
│              │ - Gemini    │       │ MCP         │              │
│              │ - OpenAI    │       │             │              │
│              │ - GLM       │       │             │              │
│              │ - Local     │       │             │              │
│              └─────────────┘       └─────────────┘              │
│                                                                  │
└─────────────────────────────────────────────────────────────────┘

Setup​

Default Profiles​

Droid comes with pre-configured profiles:

{
  "GEMINI_2_5_PRO": {
    "DROID": {
      "autonomy": "workspace-write",
      "model": "gemini-2.5-pro-preview-06-05"
    }
  },
  "GLM_4_7": {
    "DROID": {
      "autonomy": "workspace-write",
      "model": "glm-4.7"
    }
  }
}

Custom Profile​

Create your own profile:

{
  "MY_CUSTOM_AGENT": {
    "DROID": {
      "autonomy": "workspace-write",
      "model": "your-model-id",
      "apiKey": "YOUR_API_KEY",
      "baseUrl": "https://your-api-endpoint.com/v1"
    }
  }
}

Configuration​

Profile Options​

OptionDescriptionDefault
modelModel identifierRequired
autonomyPermission levelworkspace-write
apiKeyAPI key (or env var)-
baseUrlCustom API endpoint-
maxTokensMax output tokens8192
temperatureRandomness (0-1)0.7

Autonomy Levels​

LevelCapabilities
workspace-writeRead/write workspace files
skip-permissions-unsafeFull system access

Environment Variables​

# Set API keys via environment
export GEMINI_API_KEY="your-key"
export OPENAI_API_KEY="your-key"
export GLM_API_KEY="your-key"

Usage​

From CLI​

# Use specific profile
friday-dev run --task 123 --agent droid --profile GEMINI_2_5_PRO

# Use custom profile
friday-dev run --task 123 --agent droid --profile MY_CUSTOM_AGENT

From UI​

  1. Open task
  2. Click "Run Agent"
  3. Select "Droid"
  4. Choose profile
  5. Start

Supported Models​

Cloud Providers​

ProviderModels
GoogleGemini 2.5 Pro, Gemini 2.5 Flash
OpenAIGPT-5, GPT-4 Turbo, GPT-4o
AnthropicClaude 3.5 Sonnet, Claude 3 Opus
ZhipuGLM-4.7, GLM-4
AlibabaQwen-Coder

Self-Hosted​

SystemModels
OllamaLlama, Mistral, CodeLlama
vLLMAny compatible model
LocalAIAny GGUF model

Self-Hosted Setup​

Ollama​

# Install Ollama
curl https://ollama.ai/install.sh | sh

# Pull a coding model
ollama pull codellama

# Configure Friday Dev
{
  "LOCAL_CODELLAMA": {
    "DROID": {
      "autonomy": "workspace-write",
      "model": "codellama",
      "baseUrl": "http://localhost:11434"
    }
  }
}

vLLM​

# Start vLLM server
python -m vllm.entrypoints.openai.api_server \
  --model codellama/CodeLlama-34b \
  --port 8000
{
  "VLLM_CODELLAMA": {
    "DROID": {
      "autonomy": "workspace-write",
      "model": "codellama/CodeLlama-34b",
      "baseUrl": "http://localhost:8000/v1"
    }
  }
}

MCP Integration​

Droid supports Model Context Protocol:

{
  "mcp": {
    "enabled": true,
    "servers": [
      {
        "name": "filesystem",
        "command": "mcp-filesystem",
        "args": ["--root", "/workspace"]
      }
    ]
  }
}

Creating Custom Agents​

Step 1: Define Profile​

{
  "MY_AGENT": {
    "DROID": {
      "autonomy": "workspace-write",
      "model": "my-model",
      "systemPrompt": "You are a helpful coding assistant..."
    }
  }
}

Step 2: Add to Configuration​

Edit ~/.friday-dev/profiles.json:

{
  "profiles": {
    "MY_AGENT": { ... }
  }
}

Step 3: Use the Agent​

friday-dev run --task 123 --agent droid --profile MY_AGENT

Troubleshooting​

Model Not Responding​

  1. Check API endpoint is reachable
  2. Verify API key is valid
  3. Check model name is correct
  4. Review error logs

Poor Performance​

  1. Try a different model
  2. Increase maxTokens
  3. Adjust temperature
  4. Add more context in task

Connection Issues​

  1. Check network/firewall
  2. Verify base URL
  3. Test API endpoint directly
  4. Check rate limits

Best Practices​

  1. Start with presets - Use built-in profiles first
  2. Test thoroughly - Verify custom profiles work
  3. Use appropriate autonomy - Don't over-permission
  4. Monitor costs - Different models have different pricing
  5. Keep profiles versioned - Track configuration changes

Next Steps​