ape-openai 1.0.5

OpenAI integration for APE (AI Programmatic Execution)

ape-openai

OpenAI integration for APE (AI Programmatic Execution).

What is ape-openai?

ape-openai bridges APE's deterministic validation layer with OpenAI's function calling API. It prevents hallucinations in function parameters by enforcing strict type checking and constraints before execution.

Why ape-openai?

OpenAI's function calling is powerful but can be unreliable:

• Function parameters can be incorrectly formatted
• Type mismatches cause runtime errors
• Missing required fields break execution
• No validation before calling your code

ape-openai solves this by adding APE as a validation layer:

OpenAI → JSON parameters → APE validation → Deterministic execution ✓

Installation

# Core package (schema conversion + execution)
pip install ape-openai

# With OpenAI SDK (for code generation)
pip install ape-openai[openai]

# Development dependencies
pip install ape-openai[dev]

Prerequisites:

• Python >= 3.11
• ape-lang >= 0.2.0

Test Coverage

✅ All tests passing

• Total tests: 49
• Last verified via pytest discovery

See ../ape/docs/APE_TESTING_GUARANTEES.md for details on what these tests guarantee.

The test suite covers:

• Schema conversion (APE → OpenAI)
• Executor (OpenAI → APE runtime)
• Utils (error formatting, validation)
• End-to-end integration
• Generator (NL → APE code)

To verify test counts:

pytest packages/ape-openai/tests --collect-only -q

Quick Start

from openai import OpenAI
from ape_openai import ApeOpenAIFunction

# 1. Create Ape task file
# calculator.ape:
# task add:
#     inputs: a: Integer, b: Integer
#     outputs: sum: Integer
#     constraints: a > 0, b > 0
#     steps: sum = a + b

# 2. Load as OpenAI function
func = ApeOpenAIFunction.from_ape_file("calculator.ape", "add")

# 3. Get OpenAI function schema
function_schema = func.to_openai_function()

# 4. Use with OpenAI
client = OpenAI()
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Add 5 and 3"}],
    functions=[function_schema],
    function_call="auto"
)

# 5. Execute with APE validation
if response.choices[0].message.function_call:
    func_call = response.choices[0].message.function_call
    result = func.execute(func_call.arguments)
    print(f"Result: {result}")  # 8

API Reference

Schema Conversion

ape_task_to_openai_schema(task: ApeTask) -> dict

Converts APE task to OpenAI function schema.

from ape_openai import ape_task_to_openai_schema, ApeTask

task = ApeTask(
    name="calculate_tax",
    inputs={"amount": "float", "rate": "float"},
    output="float",
    description="Calculate tax amount"
)

schema = ape_task_to_openai_schema(task)
# {
#     "name": "calculate_tax",
#     "description": "Calculate tax amount",
#     "parameters": {
#         "type": "object",
#         "properties": {
#             "amount": {"type": "number"},
#             "rate": {"type": "number"}
#         },
#         "required": ["amount", "rate"]
#     }
# }

Execution

execute_openai_call(module, function_name, arguments) -> Any

Execute OpenAI function call with APE validation.

from ape import compile
from ape_openai import execute_openai_call

module = compile("calculator.ape")
result = execute_openai_call(module, "add", '{"a": 5, "b": 3}')

High-Level Wrapper

ApeOpenAIFunction

Complete integration wrapper.

func = ApeOpenAIFunction.from_ape_file("calc.ape", "multiply")

# Get schema
schema = func.to_openai_function()

# Execute
result = func.execute({"a": 4, "b": 7})

Features

• ✅ Schema conversion: APE → OpenAI function format
• ✅ Type validation: Strict parameter checking
• ✅ Error handling: Clear error messages
• ✅ Code generation: Natural language → APE (experimental)
• ✅ Full type support: String, Integer, Float, Boolean, List, Dict

Type Mapping

Ape Type	OpenAI Type
String	string
Integer	integer
Float	number
Boolean	boolean
List	array
Dict	object

Advanced Usage

Code Generation

Generate APE code from natural language:

from ape_openai import generate_ape_from_nl

code = generate_ape_from_nl(
    "Create a function that calculates compound interest",
    model="gpt-4o"
)
print(code)

Error Handling

from ape_openai import ApeOpenAIFunction
from ape import ApeExecutionError

func = ApeOpenAIFunction.from_ape_file("calc.ape", "divide")

try:
    result = func.execute({"a": 10, "b": 0})
except ApeExecutionError as e:
    print(f"Execution failed: {e}")

Examples

See the examples/ directory for complete examples:

• Basic function calling
• Multi-function conversations
• Error handling patterns
• Code generation workflows

Development

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

# Run tests
pytest

# Format code
black .

# Type checking
mypy src/

License

MIT License - see LICENSE file for details.

Links

• APE Language
• Documentation
• Issues