sahilleth/promptlens

PromptLens is a monitoring and debugging tool built for people who develop applications and agents powered by large language models. When you build something that calls an AI model, it can be hard to see exactly what was sent to the model, what came back, what tools it called, and what it cost. PromptLens captures all of that activity, prompts, responses, tool calls, memory operations, document retrievals, and cost data, and shows it in a live dashboard, similar in spirit to the developer tools built into a web browser. The project is split into a few parts: a backend server built with FastAPI that stores the data and handles migrations, a frontend dashboard built with React and Vite, a Python SDK you install into your own application to send trace data, and a plugin system for adding new event types. The dashboard updates in real time over a WebSocket connection as new events come in, and includes pages for browsing sessions, inspecting prompts, viewing tool and memory events, and looking at aggregated analytics like throughput, latency, and cost over time. Getting started means running the backend with Python, running the frontend separately with Node.js, and then installing the SDK into whatever application you want to trace, with optional integrations for OpenAI, Ollama, and LangChain. A handful of demo scripts are included to generate sample traffic so you can see the dashboard working without writing your own instrumentation first. By default the tool does not require an API key while in development mode, but you can generate your own secret key and configure it on both the backend and any client so the API is protected, and production mode refuses to start without one being set. The backend can use SQLite by default or be pointed at a PostgreSQL database instead. This is aimed at developers building AI agents or LLM powered applications who need visibility into what their systems are actually doing under the hood.