What is hipdash?

Hipdash is a live monitoring dashboard built for MTPLX, a multi-token-prediction runtime for large language models. It watches an MTPLX server while it runs and shows what is happening in real time: how many tokens it is generating per second, how long it takes before the first token appears, memory and CPU use, and the state of its session cache. Everything it collects is saved to a local SQLite database, so past runs and restarts stay visible even after you close the dashboard. The dashboard has several pages. The Overview page gives a quick snapshot of decode speed, time to first token, CPU and memory percentages, and when the service last restarted. The MTPLX page is the most detailed, showing the live generation pipeline stage by stage along with twelve history charts covering things like prefill time, cached tokens, and acceptance rates. There is also a page for OpenCode agent activity, a Computer page showing general host health like CPU, memory, and disk, and a Restarts page that lists every detected restart with details. Under the hood, hipdash is a small full stack app. The backend is written in Node with Express and the better-sqlite3 library, and it polls the MTPLX service and a companion sidecar service every few seconds to record snapshots and metrics. The frontend is built with Vite, React, and TypeScript, and the backend serves the finished frontend so the whole thing runs behind a single web address. It can be started directly with npm for local development or run continuously with PM2, a process manager. The dashboard also exposes its own small HTTP API so other tools can pull the same history data. This project is aimed squarely at developers already running or building on MTPLX. It is not a general purpose dashboard, it assumes MTPLX and its sidecar service are already running on the same machine. The README does not mention a license, so its usage terms are unclear.