📦 Install

# Via npm (meta-package, no postinstall scripts)
npm install -g @kleosr/pe2-cli

# From source (requires Rust 1.81+)
cargo install --git https://github.com/kleosr/PE2-CLI

# Or grab a binary from the releases page
# https://github.com/kleosr/PE2-CLI/releases

🚀 Usage

# Interactive mode — just run it
pe2

# One-shot
pe2 "Write a blog post about AI"

# From a file
pe2 path/to/prompt.txt

# With overrides
pe2 "Explain quantum computing" --provider openai --model gpt-4o --iterations 5

# Config menu
pe2 --config

No arguments = interactive mode. First run will prompt you to set up your provider and API key.

⚙️ Configuration

Settings live in ~/.kleosr-pe2/config.json. Change them through the interactive menu (/config) or edit the file directly.

| Flag | What it does | |------|-------------| | -p, --provider | LLM provider: openai, anthropic, google, openrouter, ollama | | -m, --model | Model identifier (check your provider's docs) | | --api-key | Your API key (or set the env var below) | | -o, --output-file | Where to save the result | | -i, --iterations | Refinement pass count (auto-detected by default) | | --max-tokens | Max response tokens (default: 1024) | | --temperature | Sampling temperature (default: 0.3) |

Environment Variables

| Provider | Variable | |----------|----------| | OpenAI | OPENAI_API_KEY | | Anthropic | ANTHROPIC_API_KEY | | Google | GOOGLE_API_KEY | | OpenRouter | OPENROUTER_API_KEY | | Ollama (local) | OLLAMA_BASE_URL |

🔄 How It Works

1. Complexity analysis — scores your prompt on 5 factors (tech, domain, structure, logic, special chars), maps to difficulty tiers
2. LLM call — sends it with a structured JSON template
3. Refinement loop — auto-detects complexity; simple prompts get 1 pass, technical deep-dives get up to 5
4. Output — writes to ./pe2-prompts/ as markdown with full history and metrics

{
  "context": "…",
  "role": "…",
  "task": "…",
  "constraints": "…",
  "output": "…"
}

Every prompt is returned as a structured PE² JSON document.

🏗️ Architecture

crates/
├── pe2-core/       — config, analysis, engine/pipeline, templates
├── pe2-providers/  — 5 adapters: OpenAI, Anthropic, Google, Ollama, OpenRouter
├── pe2-tui/        — banner, spinner, themed display, interactive REPL (crossterm)
├── pe2-cli/        — binary entry: clap args, single-prompt + interactive modes
└── pe2-bindings/   — napi-rs bridge (optional, Node.js native addon)

Clean dependency chain — no circular deps. ~3,200 lines of Rust across 36 source files.

🛠️ Development

git clone https://github.com/kleosr/PE2-CLI.git
cd PE2-CLI
cargo build
cargo test
cargo run -- --help

Tests live in crates/*/tests/. Run with cargo test.

🤖 CI/CD

Push a v* tag and GitHub Actions:

1. Runs cargo test
2. Matrix builds across 6 platforms (linux x64/arm64, darwin x64/arm64, windows x64/arm64)
3. Uploads release tarballs

🦀 Why Rust?

I wrote the first version in Node.js. It worked, but:

• ~300ms startup waiting for the runtime to warm up
• GC pauses during prompt refinement
• Callbacks and promises for concurrency

Rust fixed all of it. Single static binary, ~2ms startup, zero GC, and tokio async everywhere. Five concurrent provider calls? No problem.

Built with Cursor.

📄 License

ISC