Metadata-Version: 2.4
Name: ape-openai
Version: 0.1.2
Summary: OpenAI integration for the Ape AI-first programming language
Author-email: David Van Aelst <david@skyrah.be>
License: MIT
Project-URL: Homepage, https://github.com/Quynah/ape-openai
Project-URL: Repository, https://github.com/Quynah/ape-openai
Project-URL: Documentation, https://github.com/Quynah/ape-openai#readme
Project-URL: Issues, https://github.com/Quynah/ape-openai/issues
Keywords: ape,openai,gpt,ai,llm,deterministic,function-calling
Classifier: Development Status :: 3 - Alpha
Classifier: Intended Audience :: Developers
Classifier: License :: OSI Approved :: MIT License
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Topic :: Software Development :: Libraries :: Python Modules
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Requires-Python: >=3.11
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: typing-extensions>=4.0.0
Provides-Extra: dev
Requires-Dist: pytest>=7.0.0; extra == "dev"
Requires-Dist: pytest-cov>=4.0.0; extra == "dev"
Requires-Dist: black>=23.0.0; extra == "dev"
Requires-Dist: mypy>=1.0.0; extra == "dev"
Provides-Extra: openai
Requires-Dist: openai>=1.0.0; extra == "openai"
Dynamic: license-file

# ape-openai

**OpenAI function calling integration for the Ape AI-first programming language**

[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
[![Python 3.11+](https://img.shields.io/badge/python-3.11+-blue.svg)](https://www.python.org/downloads/)

## What is this?

`ape-openai` bridges the [Ape programming language](https://github.com/Quynah/Ape) with [OpenAI's function calling API](https://platform.openai.com/docs/guides/function-calling), enabling:

- **Deterministic function execution** - OpenAI decides *when* to call, Ape ensures *how* it executes
- **Zero hallucination in parameters** - Ape validates all inputs before execution
- **Automatic schema generation** - Convert Ape tasks to OpenAI function schemas
- **Type-safe execution** - Constraints enforced at runtime

## Why Use This?

### The Problem: OpenAI Function Calling is Unreliable

```python
# Traditional approach - GPT can hallucinate parameters
def calculate_price(base, tax):
    return base * (1 + tax)  # What if tax = "high"? What if base = -100?

tools = [{"type": "function", "function": {...}}]
# GPT might send invalid JSON, wrong types, missing params
```

### The Solution: Ape + OpenAI

```python
# Ape enforces determinism
from ape_openai import ApeOpenAIFunction
from openai import OpenAI

client = OpenAI()

# Wrap Ape task as OpenAI function
func = ApeOpenAIFunction.from_ape_file("pricing.ape", "calculate_total")

# Get OpenAI-compatible schema
tools = [func.to_openai_tool()]

# Chat with function calling
response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Calculate total for $99 with 21% tax"}],
    tools=tools
)

# Execute with Ape validation
if response.choices[0].message.tool_calls:
    tool_call = response.choices[0].message.tool_calls[0]
    result = func.execute(tool_call.function.arguments)
    # Result is validated, type-safe, deterministic
```

## Installation

```bash
pip install ape-openai
```

With OpenAI support:
```bash
pip install ape-openai[openai]
```

## Quick Start

### 1. Write an Ape Task

Create `pricing.ape`:

```ape
module pricing

task calculate_total
  inputs:
    base_price: String
    tax_rate: String
  outputs:
    total: String
  constraints:
    - base_price must be numeric and positive
    - tax_rate must be between 0 and 1
  steps:
    - Parse base_price as decimal
    - Multiply by (1 + tax_rate)
    - Return formatted total
```

### 2. Convert to OpenAI Function Schema

```python
from ape import compile
from ape_openai import ape_task_to_openai_schema
from ape_openai.task import ApeTask

# Compile Ape module
module = compile("pricing.ape")

# Extract task
task = ApeTask.from_module(module, "calculate_total")

# Generate OpenAI schema
schema = ape_task_to_openai_schema(task)

print(schema)
# {
#     "type": "function",
#     "function": {
#         "name": "calculate_total",
#         "description": "Deterministic Ape task: calculate_total",
#         "parameters": {
#             "type": "object",
#             "properties": {
#                 "base_price": {"type": "string"},
#                 "tax_rate": {"type": "string"}
#             },
#             "required": ["base_price", "tax_rate"]
#         }
#     }
# }
```

### 3. Execute OpenAI Function Calls with Ape

```python
from ape_openai import execute_openai_call
import json

# Simulate OpenAI function call response
arguments_json = '{"base_price": "99.99", "tax_rate": "0.21"}'

# Execute with Ape validation
result = execute_openai_call(
    module=module,
    function_name="calculate_total",
    arguments_json=arguments_json
)

print(result)  # "121.00" - validated and deterministic
```

### 4. Full OpenAI Integration Example

```python
from openai import OpenAI
from ape import compile
from ape_openai import ApeOpenAIFunction

# Initialize OpenAI client
client = OpenAI(api_key="your-api-key")

# Create Ape-backed function
pricing_func = ApeOpenAIFunction.from_ape_file(
    "pricing.ape",
    "calculate_total",
    description="Calculate total price including tax"
)

# Chat with function calling
messages = [
    {"role": "user", "content": "What's the total for $149.99 with 21% tax?"}
]

response = client.chat.completions.create(
    model="gpt-4",
    messages=messages,
    tools=[pricing_func.to_openai_tool()],
    tool_choice="auto"
)

# Handle function call
message = response.choices[0].message
if message.tool_calls:
    for tool_call in message.tool_calls:
        if tool_call.function.name == "calculate_total":
            # Execute with Ape - validated and deterministic
            result = pricing_func.execute(tool_call.function.arguments)
            
            # Add result to conversation
            messages.append(message)
            messages.append({
                "role": "tool",
                "tool_call_id": tool_call.id,
                "content": str(result)
            })
            
            # Get final response
            final_response = client.chat.completions.create(
                model="gpt-4",
                messages=messages
            )
            
            print(final_response.choices[0].message.content)
```

## API Overview

### `ape_task_to_openai_schema(task: ApeTask) -> dict`

Convert an Ape task to OpenAI function schema format.

```python
from ape import compile
from ape_openai import ape_task_to_openai_schema
from ape_openai.task import ApeTask

module = compile("calculator.ape")
task = ApeTask.from_module(module, "add")
schema = ape_task_to_openai_schema(task)
```

### `execute_openai_call(module, function_name, arguments_json) -> Any`

Execute an OpenAI function call with Ape validation.

```python
from ape_openai import execute_openai_call

result = execute_openai_call(
    module=compiled_module,
    function_name="add",
    arguments_json='{"a": "5", "b": "3"}'
)
```

### `ApeOpenAIFunction` (High-level Helper)

Convenient wrapper combining schema generation and execution.

```python
from ape_openai import ApeOpenAIFunction

func = ApeOpenAIFunction.from_ape_file("module.ape", "task_name")

# Get OpenAI tool definition
tool = func.to_openai_tool()

# Execute function call
result = func.execute(arguments_json)
```

### `generate_ape_from_nl(prompt, model="gpt-4") -> str` (Experimental)

Generate Ape code from natural language description.

```python
from ape_openai import generate_ape_from_nl

ape_code = generate_ape_from_nl(
    "Create a task that calculates compound interest"
)
print(ape_code)
# Returns Ape code as string
```

## Features

- ✅ **Automatic schema generation** - Ape tasks → OpenAI function schemas
- ✅ **Type-safe execution** - All inputs validated before execution
- ✅ **Zero hallucination** - Invalid parameters rejected instantly
- ✅ **JSON error handling** - Graceful handling of malformed arguments
- ✅ **OpenAI compatible** - Works with GPT-4, GPT-3.5, and fine-tuned models
- ✅ **Streaming support** - Compatible with OpenAI streaming responses

## Type Mapping

Ape types are mapped to OpenAI JSON Schema types:

| Ape Type | OpenAI Type |
|----------|-------------|
| String   | "string"    |
| Integer  | "integer"   |
| Float    | "number"    |
| Boolean  | "boolean"   |
| List     | "array"     |
| Any      | "string" (fallback) |

## Error Handling

```python
from ape_openai import execute_openai_call, ApeExecutionError

try:
    result = execute_openai_call(module, "calculate", '{"invalid": "params"}')
except ApeExecutionError as e:
    print(f"Execution failed: {e}")
    # Handle gracefully
```

## Requirements

- Python 3.11+
- ape-lang (installed automatically)
- openai >= 1.0.0 (optional, for OpenAI API calls)

## Development Status

**v0.1.0** - Initial release

Currently supports:
- Function schema generation
- Function call execution
- Basic type mapping
- Error handling

Roadmap:
- Advanced type system (nested objects, arrays)
- Streaming function call support
- Multi-function orchestration
- Cost tracking and optimization

## Examples

See `examples/` directory for:
- Basic function calling
- Multi-turn conversations
- Error handling patterns
- Streaming responses

## Contributing

Contributions welcome! See the [main Ape repository](https://github.com/Quynah/Ape) for guidelines.

## License

MIT License - see LICENSE file for details.

## Links

- [Ape Language](https://github.com/Quynah/Ape)
- [OpenAI Function Calling](https://platform.openai.com/docs/guides/function-calling)
- [Documentation](https://github.com/Quynah/Ape/tree/main/docs)

---

**Built with Ape 🦍 - Making AI function calls deterministic, not approximate.**
