Vllora

vLLora is a debugging and observability tool for AI agents that traces every model call as it happens. It connects to agent frameworks such as LangChain, Google ADK and the OpenAI Agents SDK, and presents latency, cost and model output in one unified trace view. The goal is to let developers see what an agent actually did, find silent failures such as invisible retries that inflate spend, and fix them quickly. It works through OpenAI compatible endpoints, so an existing setup can be pointed at it without rewriting code. Users bring their own provider keys and can reach more than 300 models, mixing and benchmarking them side by side. The project publishes an MCP server so traces can be inspected and agents debugged directly from an IDE, and a built in assistant called Lucy reads traces, diagnoses agent failures and suggests concrete fixes. The tool targets developers and teams building agent workflows who need runtime visibility across every model call. Installation is a Homebrew command, brew tap vllora/vllora followed by brew install vllora. The site states that it is free for personal and work use under its published license, and it lists no paid plans or prices.