pCompiler

pCompiler is a declarative prompt engineering framework that transforms high-level DSL definitions into optimized, model-specific LLM prompts. It bridges the gap between raw text prompting and structured, versioned, and secure prompt management.

🌟 What the project does

pCompiler allows developers to treat prompts as code. By defining prompts in a structured YAML Domain Specific Language (DSL), you can:

Generate starting specifications from natural language descriptions.
Advanced Templating: Support for Jinja2 in user input templates (loops, conditionals).
Validate prompts for ambiguities, contradictions, and security risks.
Optimize content for specific LLM backends (OpenAI, Anthropic, Gemini).
Estimate Costs before execution with real-time pricing and latency analytics.
Automate quality control with built-in evaluation suites.
Deploy versioned, reproducible prompt payloads to your applications.

✨ Why pCompiler?

In a world where LLM prompts are increasingly complex and critical to application logic, pCompiler provides the tools to make them robust:

Type Safety & Validation: Uses Pydantic to ensure your specifications are always valid.
Smart Optimization: Automatically applies semantic compression, Chain-of-Thought policies, and section reordering tailored to each model's strengths.
Security by Design: Native protection against prompt injection and system prompt leakage.
Collaborative Engineering: Versioned YAML files make it easy for teams to track changes and collaborate via Git.
RAG Ready: Built-in support for multiple context sources, including local files, vector stores, and web search.

🚀 How to get started

Installation

# Clone the repository
git clone https://github.com/marcosjimenez/pCompiler.git
cd pCompiler

# Install in editable mode with dev dependencies
pip install -e ".[dev]"

Quick Start

Auto-generate a DSL spec:

pcompile create "Summarize a medical report focusing on patient history." --output medical.yaml

Refine your DSL specification (medical.yaml):

task: summarize
model_target: gpt-4o
constraints:
  tone: professional
  cot_policy: auto
instructions:
  - text: "Highlight patient history and active medications."
    priority: 100

Compile to a model-ready payload:

pcompile compile medical.yaml --target gpt-4o

Run automated evaluations:
```
pcompile eval medical.yaml --mock
```

📖 Where to get help

DSL Reference: Detailed documentation of the YAML schema and attributes.
Advanced Templating: Using Jinja2 for loops, conditionals, and complex logic.
CLI Guide: Full list of commands and options.
Cost & Latency Estimator: How to estimate prompt costs before execution and manage pricing data.
Auto-Evals System: How to build and run automated prompt quality tests.
Context Engineering: Strategies for RAG and dynamic background information.
Integrated CI/CD: Automating validation and testing in your pipeline.
Packaging: How to build and distribute pCompiler on your own registry.
Plugin System: Guide on how to use, write and manage backend plugins and their associated models.

👥 Who maintains & contributes

Maintainer

Marcos Jiménez - GitHub Profile

Contributing

We welcome contributions! Please feel free to open an issue or pull request to suggest improvements or report bugs.

pCompiler — Optimize your prompts, automate your evaluations.

PCompiler

Install / Use

README