codegnipy 0.0.3

AI 原生的 Python 语言扩展 - 让非确定性 AI 成为 Python 一等公民

Codegnipy

AI-Native Python Language Extension

Codegnipy seamlessly integrates Large Language Models (LLMs) into Python, making non-deterministic AI capabilities a first-class citizen in your code. Write logic close to natural language while achieving production-grade performance and debuggability.

Features

• Syntactic Extension: ~ operator for natural language prompts directly in code
• Cognitive Decorator: @cognitive decorator to let LLMs implement functions
• Memory Management: Session-level memory with pluggable storage backends
• Reflection Loop: Built-in self-correction and quality assurance
• Async Scheduler: High-performance concurrent LLM calls with priority queuing
• Deterministic Guarantees: Type constraints, simulation mode, and hallucination detection

Installation

pip install codegnipy

For development:

pip install codegnipy[dev]

Quick Start

The ~ Operator

from codegnipy import CognitiveContext, cognitive_call

with CognitiveContext(api_key="your-api-key"):
    # Natural language prompts directly in code
    result = ~"Translate to English: Hello World"
    print(result)

The @cognitive Decorator

from codegnipy import cognitive, CognitiveContext

@cognitive
def summarize(text: str) -> str:
    """Summarize the key points of this text in no more than two sentences."""

with CognitiveContext(api_key="your-api-key"):
    summary = summarize("Python is a high-level programming language...")
    print(summary)

Async Batch Processing

import asyncio
from codegnipy import batch_call, CognitiveContext

async def main():
    prompts = [
        "Translate: Hello",
        "Translate: World", 
        "Translate: Python"
    ]
    results = await batch_call(prompts, max_concurrent=3)
    print(results)

asyncio.run(main())

With Memory Persistence

from codegnipy import CognitiveContext, FileStore

with CognitiveContext(
    api_key="your-api-key",
    memory_store=FileStore("session_memory.json")
):
    cognitive_call("My name is Alice")
    response = cognitive_call("What is my name?")
    # LLM will remember: "Alice"

With Type Constraints

from codegnipy import PrimitiveConstraint, deterministic_call

# Ensure LLM output is a valid integer between 0-100
constraint = PrimitiveConstraint(
    int,
    min_value=0,
    max_value=100
)

result = deterministic_call(
    "Generate a random number between 1 and 100",
    constraint
)

if result.status == "valid":
    print(result.value)  # Guaranteed valid integer

With Reflection

from codegnipy import CognitiveContext, with_reflection

with CognitiveContext(api_key="your-api-key") as ctx:
    result = with_reflection(
        "Explain quantum entanglement",
        context=ctx,
        max_iterations=2
    )
    
    if result.status == "passed":
        print(result.corrected_response or result.original_response)

Architecture

Python Source Code
        |
        v
  AST Preprocessing
        |
        v
Transformed Code + cognitive_call()
        |
        v
   Runtime Layer
        |
        v
  Scheduler (async)
        |
        v
    LLM APIs
        |
        v
  Validation Layer
        |
        v
 Deterministic Result

API Reference

Core Functions

Function	Description
cognitive_call(prompt, context=None, model=None, temperature=None)	Execute a cognitive call to LLM
deterministic_call(prompt, constraint, context=None)	Call LLM with type constraints
batch_call(prompts, max_concurrent=5)	Execute multiple prompts concurrently

Decorators

Decorator	Description
@cognitive	Decorate a function to be implemented by LLM
@cognitive(model="gpt-4")	With specific model selection

Context Manager

CognitiveContext(
    api_key=None,           # OpenAI API key (or use OPENAI_API_KEY env var)
    model="gpt-4o-mini",    # Default model
    base_url=None,          # Custom API endpoint
    temperature=0.7,        # Sampling temperature
    max_tokens=1024,        # Maximum response tokens
    memory_store=None       # Memory storage backend
)

Memory Backends

Class	Description
InMemoryStore	Volatile in-memory storage
FileStore(path)	Persistent file-based storage

Type Constraints

Class	Description
PrimitiveConstraint(type, min_value=None, max_value=None, min_length=None, max_length=None, pattern=None)	Validate primitive types
EnumConstraint(values)	Validate enum values
SchemaConstraint(pydantic_model)	Validate against Pydantic schema
ListConstraint(item_constraint, min_items=None, max_items=None)	Validate list items

Scheduler

from codegnipy import CognitiveScheduler, Priority

scheduler = CognitiveScheduler(
    max_concurrent=5,
    default_timeout=30.0,
    retry_policy=RetryPolicy(max_retries=3, base_delay=1.0)
)

# Submit with priority
task_id = await scheduler.submit(
    my_coroutine,
    priority=Priority.HIGH,
    timeout=60.0
)

# Get result
result = await scheduler.get_result(task_id, timeout=10.0)

Hallucination Detection

from codegnipy import HallucinationDetector

detector = HallucinationDetector()
check = detector.detect(llm_response)

print(check.confidence)  # 0.0 - 1.0
print(check.issues)      # List of detected issues

Configuration

Environment Variables

Variable	Description
OPENAI_API_KEY	OpenAI API key
CODEGNIPY_MODEL	Default model to use
CODEGNIPY_TEMPERATURE	Default temperature
CODEGNIPY_MAX_TOKENS	Default max tokens

Programmatic Configuration

from codegnipy import CognitiveContext

ctx = CognitiveContext(
    api_key="sk-...",
    model="gpt-4",
    temperature=0.5,
    max_tokens=2048
)

CLI Usage

# Run a script with cognitive features
codegnipy run script.py

# Start interactive REPL
codegnipy repl

# Show version
codegnipy version

# With options
codegnipy run script.py --model gpt-4 --api-key sk-...

Testing

Unit Tests

# Run all tests
pytest tests/ -v

# With coverage
pytest tests/ --cov=codegnipy --cov-report=html

Simulation Mode

For testing without actual LLM calls:

from codegnipy import Simulator, SimulationMode

simulator = Simulator(mode=SimulationMode.MOCK)

# Mock responses
simulator.add_mock("Hello", "Hi there!")

# Or record real responses for replay
simulator = Simulator(mode=SimulationMode.RECORD)
# ... make real calls ...
simulator.save_recordings("recordings.json")

# Later, replay them
simulator = Simulator(mode=SimulationMode.REPLAY)
simulator.load_recordings("recordings.json")

Project Structure

Codegnipy/
  codegnipy/
    __init__.py          # Package exports
    runtime.py           # Core runtime (cognitive_call, CognitiveContext)
    decorator.py         # @cognitive decorator
    transformer.py       # AST transformer for ~ operator
    memory.py            # Memory storage backends
    reflection.py        # Reflection loop implementation
    scheduler.py         # Async scheduler with retry/timeout
    determinism.py       # Type constraints, simulator, hallucination detection
    cli.py               # Command-line interface
  tests/
    test_transformer.py
    test_memory.py
    test_scheduler.py
    test_determinism.py
  examples/
    demo.py
  pyproject.toml
  README.md

Contributing

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit changes (git commit -m 'Add amazing feature')
4. Push to branch (git push origin feature/amazing-feature)
5. Open a Pull Request

Development Setup

# Clone repository
git clone https://github.com/ChidcGithub/Codegnipy.git
cd Codegnipy

# Create virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install development dependencies
pip install -e ".[dev]"

# Run tests
pytest tests/ -v

# Run linting
ruff check codegnipy/
mypy codegnipy/

Roadmap

• Rust extension for high-performance scheduling
• Support for more LLM providers (Anthropic, local models)
• Enhanced hallucination detection with external verification
• Visual debugging tools
• Distributed execution support

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

Codegnipy is inspired by the vision of making AI a natural part of programming, bridging the gap between deterministic code and probabilistic intelligence.

Links

• Documentation
• Issue Tracker
• PyPI Package