AgentOps
agentops.ai
- Free plan
- Open source
- Web
- API
- Self-hosted
What AgentOps is
AgentOps is a developer platform for building AI agents and LLM applications. It provides observability for OpenAI, CrewAI, AutoGen, and more than 400 LLMs and frameworks, with tracing for LLM calls, agents, and retrieval. Users can view tools and multi-agent interactions, rewind and replay runs, and keep trails of logs, errors, and prompt-injection attacks. Token counts, agent spending, and prices across 400+ LLMs are tracked. AgentOps supports Python and TypeScript/JavaScript applications, with integrations for frameworks including LangChain, Google ADK, OpenAI Agents, and Smolagents, plus providers such as Anthropic, OpenAI, and xAI. Dashboard sessions show summaries and waterfall views of calls, tools, and errors. It is available through a web service, API, or self-hosting. Basic is free up to 5,000 events; Pro costs $40 a month, billed Pay as you go.
Who it suits
It suits engineers building and scaling AI agents who need tracing, replay, cost tracking, and execution records. Teams can use its web service, API, or self-hosted deployment.
Good to know, and what to watch
- Tracing for LLMs, agents, and retrieval
- Replay debugging for agent runs
- Cost monitoring across 400+ LLMs
- Python and TypeScript/JavaScript SDKs
- Self-hosting option
- Basic free plan limited to 5,000 events
- Pro costs $40 a month
- Enterprise pricing is custom
Plans
Every plan and priceEnterprise
Basic
Pro
Free plan
The verdict
AgentOps brings tracing, replay, audit trails, and spending visibility to AI-agent development. Its broad framework and provider integrations, plus API and self-hosting options, support both individual developers and enterprise teams.
What the maker publishes
- Audit trail
- The site says it keeps a data trail of logs, errors, and prompt injection attacks from prototype to production. source
- Audit trail
- AgentOps records logs, errors and prompt-injection attacks across development and production. source
- Company identity
- Crunchbase lists the legal company name as Staf, Inc., also known as Agency AI and AgentOps.ai. source
- Company location
- LinkedIn lists Agency headquarters at 525 Market Street, San Francisco, California. source
- Company size
- LinkedIn lists Agency as a privately held software-development company with 11–50 employees. source
- Cost monitoring
- The service monitors agent spending using current model-price data. source
- Cost tracking
- The site says AgentOps tracks token counts and agent spend, with price monitoring across 400+ LLMs. source
- Dashboard
- The documentation says recorded program executions appear as sessions in the AgentOps Dashboard, with session summaries and a waterfall view of calls, tools, and errors. source
- Dashboard modules
- The live dashboard exposes projects, traces, metrics, MCP, Deploy Alpha, patch notes, support and documentation links. source
- Documentation
- Documentation includes quickstarts, integration guides, examples and self-hosting instructions. source
- Early reliability problem
- TechCrunch reported that the founders' initial agents failed 30–40% of the time, motivating the debugging tools that became AgentOps. source
- Enterprise security
- The homepage lists SOC-2, HIPAA, and NIST AI RMF among Enterprise plan offerings. source
- Fine-tuning
- Saved completions can be used to fine-tune specialized models at a claimed cost reduction of up to 25 times. source
- Founded
- LinkedIn and G2 list the company as founded in 2023. source
- Framework integrations
- The documentation lists integrations including AG2, Agno, AutoGen, CrewAI, Google ADK, Haystack, LangChain, OpenAI Agents, and Smolagents. source
- Framework integrations
- The SDK integrates with CrewAI, AG2/AutoGen, Agno, LangGraph, OpenAI Agents SDK and other agent frameworks. source
- Funding news
- TechCrunch reported on August 28, 2024 that Agency had raised $2.6 million in pre-seed funding led by 645 Ventures and Afore Capital. source
- G2 review availability
- G2 says there are too few AgentOps reviews to provide buying insight. source
- Headquarters
- Delray Beach, Florida, United States source
- Independent product assessment
- OpenClaw describes AgentOps as a practical observability layer for Python-heavy teams needing structured traces, replayability and cost-aware debugging. source
- Installation
- The homepage gives `pip install agentops` as the installation command. source
- Instrumentation controls
- Developers can use decorators and manual traces to define sessions, agents, operations, tasks and workflows. source
- Intended users
- The site presents AgentOps for engineers building AI agents and offers help to teams building and scaling enterprise-grade agents. source
- Licence
- The AgentOps application and repository are released under the MIT licence. source
Checked against the maker's pages on 15 Oct 2026.