Draft

ai-generator — AI Software Factory

Internalactive

Human-on-the-loop, agent-driven software factory. English requirements in → production-grade software out. Specialized AI agents handle each SDLC phase — requirements analysis, architecture, implementation, testing, review, delivery — communicating via structured protocols and shared context.

flowchart TD
    HUMAN["Human
intent + acceptance criteria"] --> GEN["generator CLI
coordinator agent"]
    GEN --> REQ["Requirements
analysis agent"]
    GEN --> ARCH["Architecture
agent"]
    GEN --> IMPL["Implementation
agent"]
    GEN --> TEST["Testing
agent"]
    GEN --> REV["Review
agent"]
    IMPL --> CODE["production-grade
software output"]
    style GEN fill:#EDEDFF,stroke:#3B54F7
    style CODE fill:#E4E4FF,stroke:#23233A

Principles: correctness over speed · self-repair before escalation · observable by default · composable agents · spec fidelity over convenience.

Install: pipx install 'git+https://github.com/propel-ai/propel.ai-generator.git'
Usage: drop requirements docs in a directory → generator init-config → generator run --coordinator
State: lives in coordinator-workspace/ (context, manifest.yaml, run.log, signals, locks) — inspectable and git-trackable.
Providers: Fireworks → OpenRouter → OpenAI → Anthropic (precedence by key availability).
Research foundation: LLM code gen, ReAct/Reflexion/LATS agentic architectures, automated testing, formal verification, neurosymbolic program synthesis, RAG for code.