# vitalops

> Open-source tools for running LLMs in production, in pure Python with no dependencies. `vitalops plan`: will the model fit, how fast will it run, how many GPUs, what will it cost, for any Hugging Face model on any GPU. `vitalops bench`: measure an OpenAI-compatible endpoint and compare it with the plan.

## Prompt

- [Prompt for coding agents](https://vitalops.ai/oss/agent-prompt.txt): install, the API, the command line, what it is and is not

## Getting started

- [Introduction](https://vitalops.ai/oss/docs/content/introduction.md): What vitalops does and does not do
- [Install](https://vitalops.ai/oss/docs/content/install.md): pip, the one-file build, checking it works
- [vitalops plan](https://vitalops.ai/oss/docs/content/cli.md): every flag, the report line by line, exit codes, JSON output, the Hub
- [vitalops bench](https://vitalops.ai/oss/docs/content/bench.md): measuring an endpoint across concurrency, against the plan, in CI
- [Python API](https://vitalops.ai/oss/docs/content/python-api.md): plan, fit, perf, the Plan, Layout, Perf and Scenario dataclasses, the catalogues, bench

## What goes in

- [Models](https://vitalops.ai/oss/docs/content/models.md): the built-in catalogue, the Hub loader, mixtures of experts, latent attention, caveats
- [GPUs](https://vitalops.ai/oss/docs/content/gpus.md): the hardware catalogue, how each figure is used, hardware not in it

## How it works

- [The physics](https://vitalops.ai/oss/docs/content/physics.md): every formula and constant, what is not modelled, calibrating against a measurement
- [vitalops ui](https://vitalops.ai/oss/docs/content/ui.md): vitalops ui and the web page, which run vitalops's Python (Pyodide on the site, the local package under vitalops ui)
- [Contributing](https://vitalops.ai/oss/docs/content/contributing.md): layout, tests, adding a GPU or a model, the one-file build

## Source

- [GitHub](https://github.com/vitalops/vitalops)
- [PyPI](https://pypi.org/project/vitalops/)