Status: [UNTESTED] Endpoints found: http://localhost:3000/api/checks http://localhost:3000/api/checks/ Source: https://github.com/syaikhipin/fake-llm-checker # πŸ•΅οΈ fake-llm-checker A tiny SaaS that answers one question: **is that cheap LLM API endpoint you bought actually serving the real model β€” or a masked substitute?** It wraps an MIT-licensed open-source scan engine (vendored in `vendor/engine/`, see `LICENSE` there) in a web service: submit an endpoint + key, get a verdict (🟒 real / 🟑 unverified / 🟠 suspicious / πŸ”΄ masked) plus the full HTML evidence report the scan generates. 100% free stack: Node.js, SQLite, Docker β€” deployable on Render's free tier. ## How it works ``` browser ──POST /api/checks──▢ Express API ──SQLite queue──▢ in-process runner (1 scan at a time) β–² β”‚ └── poll GET /api/checks/:id ◀── verdict β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ └── GET /api/checks/:id/report β†’ upstream HTML report ``` - **Scans run strictly sequentially** (`MAX_CONCURRENT_SCANS=1`): the fingerprint module fires many API calls; this keeps it polite and cheap. - **Your LLM key is never stored.** It lives in process memory only, is passed to the scan via env, and is deleted when the scan ends. A restart orphans queued jobs rather than persisting keys. - **SSRF guard**: the target hostname must resolve to a public IP (no localhost, private ranges, or `169.254.x.x` metadata). - **Rate limiting**: 60 requests/minute per IP, plus a cap on concurrent scans. ## Quickstart (local, Docker) ```bash docker compose up --build # β†’ http://localhost:3000 ``` Or without Docker (needs node β‰₯ 22, bash, curl, openssl, python3): ```bash npm install DATA_DIR=./data npm start ``` ## Free deploy (Render) 1. Push this repo to GitHub. 2. Render β†’ New β†’ Blueprint β†’ point at the repo (`render.yaml`). 3. Open the service URL. Note: the free tier has no persistent disk, so SQLite results reset on redeploy β€” fine for a checker, docum