Promptos
◐ PromptOS
A local-first, privacy-first operating system for Prompt Engineering. Analyze, optimize, compare and learn — entirely on your machine.
The DNA Analyzer — a deterministic, explainable prompt score. No LLM required.
Not another cloud prompt manager. PromptOS is a native desktop app that scores your prompts with a transparent, deterministic engine (no LLM required), helps you cut token costs, recommends the best AI model for each task, and teaches you prompt engineering — all offline. An optional local LLM (via Ollama) augments the analysis when you want it.
✨ Why it's different
- 🔒 Offline & private — everything runs locally; no data leaves your machine. API keys (if ever used) stay in the OS keychain, never in the repo.
- 🔍 Deterministic analysis — prompt scoring is a transparent, inspectable rule engine, not an opaque LLM judge. Instant, free, explainable.
- 🧠 Hybrid by design — the deterministic core is always on; an optional local LLM enriches the analysis and degrades gracefully when absent.
- 🧩 Everything is content — AI models, techniques, scoring rubrics, personas and lessons live in editable Content Packs (Markdown + JSON). Extend the knowledge without touching code.
🧰 Features — 10 modules
| Module | What it does | |
|---|---|---|
| ◐ | DNA Analyzer | Decompose a prompt into dimensions → 0–100 score, weaknesses, concrete fixes |
| ⇄ | Token ROI | Compare two prompts: quality per token, with a plain-language verdict |
| ⊚ | AI Router | Recommend the best model for a task — and why — from the editable registry |
| ⇲ | Compression | 3 levels of token reduction with measured quality impact |
| ✦ | Persona Generator | Build expert system-prompts from configurable blueprints |
| ⇡ | Prompt Evolution | Watch a prompt grow across 5 levels (quality vs. tokens) |
| ⊗ | Prompt Battle | Model-tailored prompt variants (Claude / ChatGPT / Gemini …) side by side |
| ⎇ | Versioning | Git-like versions, line diff, rollback, tags |
| ❖ | Knowledge Center | AI registry (6 models) + prompting techniques, searchable |
| ✎ | Learn | Interactive lessons + quizzes with progress tracking |
➕ an optional Ollama layer for an AI-powered review of your prompt.
📸 Screenshots
| Token ROI | AI Router |
|---|---|
| Knowledge Center | Learn |
🏛️ Architecture in one breath
UI (Svelte)
└─ L3 augmented intelligence (optional LLM, degrades offline)
└─ L2 knowledge (content-driven)
└─ L1 deterministic analysis (offline, testable)
└─ L0 foundations (tokenizer · search · versioning)
Providers: Ollama (optional) · content: Markdown + JSON
- Deterministic core — tokenizer, composable detectors, and a declarative rubric runner. Pure, testable, framework-agnostic TypeScript.
- Declarative scoring — rubrics (JSON) recombine reusable detectors; a new scoring system is a new file, not new code.
- Honest metrics — token, cost and quality are estimates carrying
confidenceandlast_verified. No false precision.
🛠️ Tech stack
Tauri 2 · Svelte 5 · TypeScript · Vite · (optional) Ollama. Zero AI dependencies for the core.
🚀 Quick start
git clone https://github.com/<your-username>/promptos
cd promptos
npm install
npm run dev # web app → http://localhost:5173
Build the native desktop app (Windows needs Rust + MSVC Build Tools):
npm run tauri build # produces PromptOS.exe + installers
Zero-dependency headless demos (run with Node 22+):
node scripts/demo-dna.ts
node scripts/demo-roi.ts
node scripts/demo-router.ts
node scripts/demo-compress.ts
🧩 Extend it
Add an AI model, a technique, a persona or a lesson by dropping a Markdown file into content/packs/official/. The app discovers it automatically — no rebuild of the logic required.
🌍 Language
The UI and learning content are currently in Italian. The content-pack model makes localization straightforward — translations welcome.
🤝 Contributing
Issues and PRs are welcome — especially new Content Packs (models, techniques, lessons) and translations.
📄 License
MIT.
