AgentOps

Two lines of code after pip install agentops start recording an agent's full execution: every LLM call, tool invocation, cost, and error lands in a session that can be replayed step by step to debug multi-agent runs. The MIT-licensed Python SDK builds on OpenTelemetry, so spans export to standard OTLP backends, and instrumentation ships for CrewAI, AG2, the OpenAI Agents SDK, LangChain, LlamaIndex, CamelAI, and LiteLLM, often enabled from the framework side with a single flag. Token spend tracking across Anthropic, OpenAI, Mistral, Cohere, and other providers, latency breakdowns, and benchmarking round out the feature set. The dashboard at agentops.ai runs as a hosted service, the application code sits in the open repository for self-hosting, and an alpha TypeScript SDK extends coverage beyond Python. With about 5,700 GitHub stars it ranks among the most adopted agent-specific observability SDKs, and docs.agentops.ai walks through each integration end to end, making it a common first stop for teams moving agent prototypes toward production.