Deploy and monitor AI agents
Agent Orchestration
Launch AI agents into sandboxed desktop environments, monitor their actions through a real-time command center, and capture full execution traces for evaluation and debugging.
Key Capabilities
Lifecycle Management
Start, pause, resume, and terminate agents through a single API. Agents run in isolated environment sessions with configurable timeouts, step limits, and resource quotas.
Event Streaming
Subscribe to a server-sent event stream for real-time agent activity. Events include screenshots, action descriptions, tool calls, and error states, delivered with sub-second latency.
Execution Traces
Every agent run produces a complete trace: screenshots at each step, action metadata, timing information, and model inputs/outputs. Traces are stored durably for post-hoc analysis.
Command Center
A live dashboard displays all running agents in a grid view. Click into any agent to watch its screen in real time, inspect recent actions, or intervene manually if needed.
Usage
curl -X POST https://api.commandagi.com/v1/agents \
-H "Authorization: Bearer $COMMANDAGI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"environment_id": "env_x7k2m",
"model": "claude-sonnet-4-20250514",
"task": "Open the settings app and enable dark mode",
"max_steps": 50,
"stream": true,
"callbacks": {
"events": "https://your-app.com/hooks/agent-events"
}
}'
# Stream response (SSE)
# event: step
# data: { "step": 1, "action": "screenshot", "screenshot_url": "..." }
# event: step
# data: { "step": 2, "action": "click", "target": [512, 340] }