Quick Start

# Option 1: Interactive OAuth login (recommended for Anthropic)
agentuse auth login

# Option 2: Use environment variables
export ANTHROPIC_API_KEY="sk-ant-..."
export OPENAI_API_KEY="sk-proj-..."

---
model: anthropic:claude-3-5-haiku-latest
description: "Simple greeting agent that welcomes users"
---

Greet the user.

agentuse run hello.agentuse

# Or run with additional instructions
agentuse run hello.agentuse "speak like a pirate"

# Run with options
agentuse run hello.agentuse --verbose
agentuse run hello.agentuse --quiet
agentuse run hello.agentuse --timeout 600

# Override the model at runtime
agentuse run hello.agentuse --model openai:gpt-5-mini
agentuse run hello.agentuse -m anthropic:claude-3-5-sonnet-20241022

---
model: anthropic:claude-3-5-haiku-latest
description: "Analyzes project structure and identifies key components"
mcp_servers:
  filesystem:
    command: "npx"
    args: ["-y", "@modelcontextprotocol/server-filesystem", "."]
    # Optional: Environment variables to pass to the MCP server
    env:
      DEBUG: "true"
    # Optional: Required environment variables
    requiredEnvVars: []
    # Optional: Allowed environment variables to pass through
    allowedEnvVars: ["HOME", "PATH"]
    # Optional: Tools to disable from this server
    disallowedTools: []
  
  # HTTP MCP Server example
  remote_api:
    url: "https://api.example.com/mcp"
    sessionId: "session-123"
    auth:
      type: "bearer"
      token: "${API_TOKEN}"
    headers:
      X-Custom-Header: "value"
    allowedEnvVars: ["API_TOKEN"]
---

You are a code analyzer. When asked about a project:
1. List the main files and folders
2. Identify the project type (Node.js, Python, etc.)
3. Summarize what the project does based on README or package.json

# Navigate to a project
cd /path/to/your/project

# Run the analyzer
agentuse run project-analyzer.agentuse

# Original agent uses claude-3-5-haiku
cat analyzer.agentuse | grep model
# model: anthropic:claude-3-5-haiku-latest

# Test with a more powerful model
agentuse run analyzer.agentuse --model anthropic:claude-3-5-sonnet-20241022

# Compare with OpenAI
agentuse run analyzer.agentuse --model openai:gpt-5

# Use cheaper model for development
agentuse run analyzer.agentuse -m openai:gpt-5-mini

# Use different API key via environment suffix
agentuse run analyzer.agentuse --model openai:gpt-4o:dev
# This uses OPENAI_API_KEY_DEV instead of OPENAI_API_KEY

---
model: anthropic:claude-3-5-haiku-latest
description: "Orchestrates tasks by delegating to specialized sub-agents"
subagents:
  - path: "./helper-agent.agentuse"
    name: "helper"
    maxSteps: 50
---

You can call the helper agent using the sub-agent tools.

# AgentUse will prompt for confirmation before executing remote agents
agentuse run https://example.com/path/to/agent.agentuse

# Preview the agent content first
# Choose 'p' when prompted to preview before execution

# Set maximum execution steps (default: 1000)
export MAX_STEPS=2500

# API keys for different providers
export ANTHROPIC_API_KEY="sk-ant-..."
export OPENAI_API_KEY="sk-proj-..."
export OPENROUTER_API_KEY="sk-or-..."

# Custom API key variables
export ANTHROPIC_API_KEY_WORK="sk-ant-..."
export OPENAI_API_KEY_PERSONAL="sk-proj-..."