What is ai_monitoring?

AI-Monitoring is a self-hosted dashboard for keeping an eye on the cost and health of a homegrown AI system built around large language models. If you are running your own AI models instead of paying for a service like OpenAI, this tool watches how much you are spending, how much data is flowing through your models, and whether the computers and software running everything are healthy. It works with a piece of software called LiteLLM to track spending and usage per API key, and it also checks on GPU, host machine, and container health for tools like Ollama, llama.cpp, and vLLM. Everything runs from one small application, and it checks each system's own built in reporting instead of adding extra monitoring software on top. The main dashboard, called Spend and Quota, puts money and usage front and center. It shows total real world cash spent on external paid models next to an estimate for self-hosted costs, daily and monthly usage charts, cost broken down by each individual key or by team, and a list of budgets ranked by how close each one is to running out. Beyond spending, there are ten different dashboard views covering an overview, LiteLLM, GPU, Ollama, llama.cpp, vLLM, network activity, alerts, spend and quota, and settings. The interface has a collapsible sidebar, a day and night viewing mode, adjustable time windows, and the ability to export data to CSV. Data is stored for up to a year by default, with detailed recent data and coarser summaries for older history. The tool can send alerts when spending spikes or when unusual activity is detected, including webhook notifications and a way to test that alerts work. It is built to run safely, using a minimal Alpine Linux base, running as a non-root user, and supporting multiple computer architectures. Getting started involves copying an example settings file, filling in a few details, and starting it with Docker. It works even without any backend configured, showing basic host statistics until you connect it to your AI setup. It is released under the Apache 2.0 license. The full README is longer than what was shown.