Developers
AI agent monitoring and uptime tracking with two lines of code. Python, Node.js, CLI, LangChain, CrewAI, AutoGen. Pick your flavor. Everything ships with your stack.
SDKs & CLI
2 lines to start monitoring.
Register your agent and it's being validated from real devices worldwide within seconds. No configuration, no infrastructure, no code changes to your agent.
# pip install agentstatus import agentstatus # Initialize with your API key agentstatus.init(api_key="as_xxx") # Register your agent. Now monitored 24/7 from real devices worldwide agentstatus.register( endpoint="https://api.mycompany.com/chat", name="Customer Support Bot", auth={"type": "bearer", "token": "..."}, ) # Check status anytime status = agentstatus.status(agent_id="abc-123") print(status.latency, status.gold_passing) # 234ms, 3/3 # One-off validation run result = agentstatus.run( endpoint="https://api.mycompany.com/chat", prompts=["What is 2+2?", "Return JSON: {x:1}"] )
Framework Integrations
Works with your stack.
LangChain, CrewAI, AutoGen, OpenAI, Anthropic. Agent Status integrates with how you already build. One line to expose any local agent to user-side monitoring.
LangChain
Callback handler or expose() for user-side monitoring of chains and agents. No deployment required.
CrewAI
Agent Status EventListener for multi-agent crews. Track which agents in your crew are healthy and fast.
AutoGen
Agent Status AgentWrapper and ConversationTracker for multi-agent conversation monitoring.
OpenAI
Drop-in StatusOpenAI client. Use it exactly like the official OpenAI client. Latency, errors, and usage tracked automatically.
Anthropic
Drop-in StatusAnthropic client for Claude-based agents. Identical API surface, automatic observability layer.
expose(): The Key Primitive
One function call. Your local agent is now reachable by Agent Status's distributed test network worldwide. No deployment. No port forwarding. No ngrok.
Tunnel Relay
Monitor before you deploy.
Testing locally? Agent Status can still test your agent from real devices worldwide. No deployment, no port forwarding, no ngrok. Powered by agentstatus-tunnel.carmel.so.
from agentstatus.integrations.langchain import expose # Your agent is running on localhost chain = LLMChain(...) # One line: tunnel opens to agentstatus-tunnel.carmel.so expose(chain, api_key="as_xxx", agent_name="Dev Agent" ) # Now Agent Status tests it from: # Sao Paulo, Tokyo, Lagos, Frankfurt, Mumbai...
CLI Reference
Command-line first.
Integrate into CI/CD pipelines, run post-deploy checks, debug from your terminal. Identical feature parity between Python and Node CLIs.
| Command | Description |
|---|---|
| agentstatus register | Register a new agent for monitoring |
| agentstatus list | List all monitored agents and their status |
| agentstatus status <id> | Get current health, latency, and gold prompt results |
| agentstatus run <id> | Trigger a manual validation run from 5 nodes |
| agentstatus delete <id> | Remove an agent from monitoring |
rora commandMCP Server
AI-native from day one.
Agent Status speaks MCP. Add agent monitoring capabilities to Claude Desktop, Cursor, or any MCP-compatible AI assistant. Ask in plain English.
Hey Claude, is my production agent healthy right now?
Let me check that for you.
Your agent is UP. Latency: 234ms. Eval checks: 3/3 passing. Last check: 47 seconds ago.
Run a full validation and tell me if anything fails.
Running validation across all nodes now.
All 5 nodes passed. No failures. P95 latency: 445ms.
Setup in 60 seconds
Start the MCP server, drop the config into Claude Desktop or Cursor, and Agent Status tools appear instantly.
{ "servers": { "agentstatus": { "command": "python", "args": ["-m", "agentstatus.mcp"], "env": { "AGENTSTATUS_API_KEY": "as_xxx" } } } }
Compatible with
Open Source
Integration layer is open source.
Fork it, extend it, contribute back. The SDKs, CLIs, integrations, and MCP server are all public under the Carmel Labs GitHub org.
What's open source
Private infrastructure