# LLM Check

> Free browser test bench for LLM APIs. Bring your own key, point the page at any OpenAI-compatible endpoint, the OpenCode Go gateway or the TypeSafe System One endpoint, then smoke-test models and new AI features: streaming chat, protocol detection, parameter overrides, raw stream inspection, model comparison and a TypeSafe Jev module.

LLM Check is a single-page web application deployed on Cloudflare Workers. It does not bundle, resell or proxy model access through a shared account: every request is signed with the API key you enter in the page and forwarded to the endpoint you configure. There is no signup, no account and no server-side storage of your key.

- Canonical URL: https://llm-check.dirtyrabbit.work/
- Interface language: English
- Cost of the site itself: free, no registration
- Models available: whichever models your own API key can access
- Contact: zweizhao@foxmail.com

## Core capabilities
- [Bring your own endpoint](https://llm-check.dirtyrabbit.work/#providers): any OpenAI-compatible base URL, Anthropic Messages style endpoints, the OpenCode Go gateway and the TypeSafe System One endpoint; OpenCode Go and TypeSafe ship as presets.
- [Protocol auto-detection](https://llm-check.dirtyrabbit.work/#features): Auto mode tries `/chat/completions`, then `/responses`, then `/messages` (OpenAI protocols first, Anthropic last) and remembers the working protocol per model so it is tried first next time.
- [Streaming chat](https://llm-check.dirtyrabbit.work/#features): responses render token by token over Server-Sent Events, and a response can be cancelled while it is still generating.
- [Parameter panel](https://llm-check.dirtyrabbit.work/#features): `temperature`, `top_p`, `max_tokens`, a streaming switch and additional request fields merged as JSON.
- [Inspector](https://llm-check.dirtyrabbit.work/#features): exact request, raw stream events, status code, time to first byte, total latency and token usage for every call, with API keys masked.
- [Compare module](https://llm-check.dirtyrabbit.work/#modules): one prompt against up to eight models in parallel, with per-model protocol, latency, usage and errors.
- [Jev module](https://llm-check.dirtyrabbit.work/#modules): TypeSafe System One state plus choice, score and noul questions, rendered with probabilities and confidence.
- [Client-side key handling](https://llm-check.dirtyrabbit.work/#privacy): API keys are kept in the browser's `localStorage` and sent only to the worker that forwards requests to the endpoint you configured.
- [Cost](https://llm-check.dirtyrabbit.work/#cost): the page itself is free with no account and no paid tier; only your own provider key incurs usage charges.
- [Examples](https://llm-check.dirtyrabbit.work/#examples): worked examples for protocol detection, side-by-side model comparison and the Jev decision module.

## Reference
- [How to use LLM Check](https://llm-check.dirtyrabbit.work/#how-to-use)
- [Providers and protocols](https://llm-check.dirtyrabbit.work/#providers)
- [Cost](https://llm-check.dirtyrabbit.work/#cost)
- [Examples](https://llm-check.dirtyrabbit.work/#examples)
- [API key and privacy](https://llm-check.dirtyrabbit.work/#privacy)
- [Frequently asked questions](https://llm-check.dirtyrabbit.work/#faq)
- [Sources and references](https://llm-check.dirtyrabbit.work/#sources)
- [Full page content as Markdown](https://llm-check.dirtyrabbit.work/llms-full.txt)
- [Sitemap](https://llm-check.dirtyrabbit.work/sitemap.xml)

## Optional
- [OpenCode Go error signatures](https://llm-check.dirtyrabbit.work/#providers): unavailable models, region restrictions, authentication failures and protocol mismatches are translated into readable messages.