Agentic PRD Generation

Alpha-stage MVP for generating Product Requirements Documents with a small, honest surface area.

Current Scope

• Generate PRDs from a single project idea.
• Stream state updates from FastAPI to Streamlit over SSE.
• Use either the OpenAI adapter or the Google GenAI adapter.
• Persist run state in Redis when available, with automatic fallback to in-memory storage for local development and tests.

Not Shipped Yet

• Framework comparison adapters such as CrewAI or AutoGen
• Supervised checkpoints or stop controls
• Technical specification generation

Architecture

• backend/: FastAPI API, runtime setup, adapters, pipeline, and state backends
• frontend/: Streamlit UI for starting runs and following live updates
• tests/: unit and integration coverage for runtime selection, pipeline behavior, streaming, and CLI wiring

The backend uses lifespan-managed shared resources: AppSettings, one StateStore, and one StreamerService per process.

Quickstart

1. Install dependencies:

pip install -e ".[dev,test]"

2. Create local configuration:

cp .env.example .env

3. Add at least one provider key to .env.

4. Run the backend:

agentic-prd --host 0.0.0.0 --port 8000 --reload

5. Run the frontend:

streamlit run frontend/app.py --server.port 8501

6. Open http://localhost:8501.

Quality Gates

ruff check .
ruff format --check .
mypy backend frontend
pytest

Docker

Runtime images install only the application package and its runtime dependencies.

docker-compose up --build

Docs

• docs/PRD.md: current MVP product requirements
• docs/TECH_SPEC.md: current architecture and API contract