PromptLab is a browser-first prompt notebook for people who iterate on prompts a lot.
Instead of managing everything in chat threads, PromptLab gives you a structured workspace where you can edit prompt cells, run them, and compare output quickly.
Most chatbox-based tools have friction for serious prompt work:
-
Prompt iteration is painful. When tuning prompts, you usually need to adjust wording while looking at responses. In chat UIs, this often means creating many new conversations just to test small changes.
-
Temporary workflow prompts are awkward. For tasks like paper writing, you may need a temporary system-style instruction. It is not stable enough to create a dedicated agent/bot, but copying it into every new chat is repetitive. Keeping it in one chat also keeps accumulating irrelevant context.
-
LLM APIs are fragmented. Different providers have different API styles and setup steps. PromptLab tries to unify this into one workflow: choose provider/model and run.
- Build prompts with multiple editable cells.
- Include/exclude specific cells for each run.
- Stream responses in real time.
- View output as raw text or rendered markdown.
- Save model/provider configurations.
- Save provider credentials per provider.
- Export/import notebook snapshots as JSON.
- Go to Settings and set provider API info (
Base URL+API Key). - Create an LLM Bot config (name, provider, model).
- Write your prompt across multiple cells.
- Toggle which cells should be included in the current run.
- Click Run to stream output.
- Refine cells and run again without creating new chat threads.
Open PromptLab directly:
- Live URL: https://niujuxin.github.io/promptlab/
- Repository: niujuxin/promptlab
Build and host the static dist/ output on any static platform.
Supported options include:
- GitHub Pages
- Netlify
- Vercel (static)
- Nginx / S3 + CDN
See DEPLOYMENT.md for deployment details.
- Node.js 20+
- npm 10+
npm install
npm run devnpm run typecheck
npm run build
npm run preview- Poe
- OpenAI-compatible APIs
- DeepSeek
PromptLab currently stores provider credentials in browser local storage. Use caution on shared/untrusted devices.
MIT License. See LICENSE.
