AgentOps — Observability Dashboard for AI Agents

Core Features

Auto-Instrumentation

AgentOps automatically detects and instruments popular frameworks:

import agentops
agentops.init()

# These are automatically tracked:
from openai import OpenAI
from anthropic import Anthropic
from crewai import Agent, Task, Crew

Supported: OpenAI, Anthropic, LangChain, LlamaIndex, CrewAI, AutoGen, Cohere, Mistral, Groq, and more.

Session Replay

Every agent run is recorded. View the complete timeline of LLM calls, tool invocations, errors, and custom events.

Cost Tracking

Real-time token usage and cost breakdown by model, session, and user:

agentops.init(tags=["production", "customer-support"])

Error Analysis

Replay the exact sequence of events that led to any failure. Filter by error type, model, or time range.

Custom Events

agentops.record(agentops.ActionEvent(
    action_type="web_search",
    params={"query": "latest AI news"},
    returns="10 results found"
))

Key Stats

• 4,500+ GitHub stars
• 15+ auto-instrumented frameworks
• Free tier: 1,000 sessions/month
• Real-time dashboard with <1s latency
• SOC 2 compliant

FAQ

Q: What is AgentOps? A: AgentOps is a Python observability SDK that provides real-time dashboards, session replays, and cost tracking for AI agent applications with just two lines of code.

Q: Is AgentOps free? A: Yes, the SDK is open-source under MIT license. Free tier includes 1,000 sessions/month.

Q: Does AgentOps work with Claude? A: Yes, AgentOps auto-instruments the Anthropic Python SDK. Every Claude API call is automatically tracked.