enable-ai 1.0.15

AI-powered natural language to REST API translation. v2: Claude native tool calling, zero hardcoding, works with any OpenAPI spec.

Enable AI

Natural language to REST API translation using LLM tool calling.

Zero hardcoding - works with any OpenAPI spec. Parent module provides all configuration.

Installation

pip install enable-ai

# For Anthropic/Claude support:
pip install enable-ai[anthropic]

Quick Start

from enable_ai_v2 import Orchestrator, Config, JWTAuth

config = Config(
    openapi_schema="path/to/openapi.json",  # or dict
    base_url="https://api.example.com",
    # llm_provider="openai",  # default
    # llm_provider="anthropic",  # optional
)

ai = Orchestrator(
    config=config,
    auth=JWTAuth(token="your-jwt-token"),
)

result = ai.process("show all pending orders")
print(result.message)

Features

• LLM tool calling - OpenAPI spec converted to LLM tools (OpenAI default, Anthropic optional)
• Server-side scoping - doesn't inject user filters; server's get_queryset() handles permissions
• Resource hints - status values, enums from your API for better accuracy
• Status synonyms - map natural language ("pending") to actual values ("DRAFT")
• Query caching - exact match + pattern match for fast repeat queries
• Progress tracking - real-time callbacks for UI updates

Configuration

Parent module provides everything:

from enable_ai_v2 import (
    Config,
    ResourceHint,
    UserContext,
    build_resource_hints_from_api,
    build_status_synonyms,
)

# Fetch from your APIs (parent module's responsibility)
openapi_schema = your_api.get_schema()
statuses = your_api.get_service_order_statuses()
priorities = your_api.get_service_order_priorities()

config = Config(
    # Required
    openapi_schema=openapi_schema,
    base_url="https://api.example.com",

    # Recommended (for accuracy)
    resource_hints=build_resource_hints_from_api(
        openapi_schema,
        statuses,
        priorities,
    ),
    status_synonyms=build_status_synonyms(statuses),

    # Optional
    # Optional
    model="gpt-4o",  # default; or "claude-sonnet-4-20250514" with llm_provider="anthropic"
    temperature=0.0,
    cache_enabled=True,
    include_trace=False,  # True for debugging
)

User Context

Parent module provides user context (doesn't fetch from APIs):

from enable_ai_v2 import UserContext

user_ctx = UserContext(
    user_id=123,
    username="john.doe",
    role="Technician",
    company_id=456,
    is_admin=False,
)

result = ai.process("show my orders", user_context=user_ctx)

Authentication

from enable_ai_v2 import JWTAuth, APIKeyAuth, BasicAuth, NoAuth

# JWT (most common)
auth = JWTAuth(token="your-jwt-token")

# API Key
auth = APIKeyAuth(api_key="your-api-key", header_name="X-API-Key")

# Basic Auth
auth = BasicAuth(username="user", password="pass")

# No auth
auth = NoAuth()

ai = Orchestrator(config=config, auth=auth)

Progress Tracking

def on_progress(message: str, progress: float):
    print(f"[{int(progress*100)}%] {message}")

config = Config(
    openapi_schema=schema,
    base_url="https://api.example.com",
    progress_callback=on_progress,
)

Response

result = ai.process("show pending orders")

result.message   # Natural language response
result.data      # Raw API data
result.success   # True/False
result.trace     # Debug info (if include_trace=True)

Key Principle

Server handles data scoping. The system prompt instructs Claude:

Do NOT add user ID filters (like technician=, customer=, created_by=) unless the user explicitly asks to filter by a specific person.

Your backend's get_queryset() permissions handle which data each user sees.

Requirements

• Python 3.9+
• OPENAI_API_KEY environment variable (default provider)
• Or ANTHROPIC_API_KEY if using llm_provider="anthropic"

License

MIT