ProductOS turns a one-line product idea into a full engineering plan in under two minutes. Four AI agents run in sequence, each building on the previous one's output, streamed live to the browser.
Your Idea → PM → System Architect → Software Architect → Launch Engineer
PRD HLD LLD Build Plan
| Agent | Artifact | What it produces |
|---|---|---|
| Product Manager | PRD | Problem, users, competitors, monetization, core tech bet, privacy risks |
| System Architect | HLD | Architecture overview, API audit, cost projection, failure modes |
| Software Architect | LLD | Tech stack, data model, folder structure, security & rate limiting |
| Launch Engineer | Build Plan | Build sequence, Day 1 prompt, hard problems, deployment path |
Each agent receives only the artifacts it needs — keeping token usage tight without losing context.
# 1. Clone
git clone https://github.com/Mugeshgithub/ProductOS
cd ProductOS
# 2. Install dependencies
pip install flask flask-cors anthropic openai python-dotenv
# 3. Set your API key (or enter it in the UI settings panel)
echo "ANTHROPIC_API_KEY=sk-ant-..." > .env
# 4. Run
python app.py
# Open http://localhost:5002Set ANTHROPIC_API_KEY in .env or paste it in the settings panel.
Models: claude-sonnet-4-5, claude-opus-4-5, claude-haiku-4-5
Set OPENAI_API_KEY in .env or paste it in the settings panel.
Models: gpt-4o, o4-mini, o3
# Install: https://ollama.com
ollama serve
ollama pull llama3.2Select Ollama in the settings panel and pick your model.
Do this before deploying. The prompts define what each agent produces. Tuning them for your domain makes a significant difference in output quality.
Prompts live in prompts/ as plain Markdown files:
prompts/
pm_agent.md ← Product Manager instructions
system_architect.md ← System Architect instructions
software_architect.md ← Software Architect instructions
launch_engineer.md ← Launch Engineer instructions
What to change:
- Add domain constraints (e.g. "always include HIPAA compliance notes for health apps")
- Adjust mandatory sections (add regulatory requirements, remove sections you don't need)
- Change output format (more tables, less prose, specific section order)
- Tune tone (startup-lean vs enterprise-formal)
No code changes needed — edit the .md files directly.
Vercel serverless functions have a 10-second timeout. For SSE streaming, deploy to a platform that supports long-running connections: Railway, Render, or Fly.io.
# Procfile (Railway / Render)
web: python app.pySet API keys as environment variables in the platform dashboard. Never commit .env.
For production, replace SQLite with a hosted database and set SECRET_KEY for session signing.
ProductOS/
app.py Flask server + SSE streaming
agents.py Agent and Orchestrator logic
index.html Single-page UI (no build step)
prompts/ Agent system prompts — customise these
build.py CLI runner (terminal mode)
- Backend: Python / Flask with SSE (
text/event-stream) - Frontend: Vanilla JS — no framework, no build step
- DB: SQLite (sessions + artifacts)
- AI: Anthropic, OpenAI, or Ollama — switchable at runtime
MIT