NL Cache Framework

What is NL Cache Framework?

NL Cache Framework (ThinkForge) is a caching system that stores natural language queries and maps them to structured outputs such as SQL queries, API calls, URLs, or workflow templates. It uses embeddings for semantic similarity search to retrieve the most relevant cached entry for a given input query, improving response accuracy and speed for NLP applications.

How to use NL Cache Framework?

Clone the repository, set up the backend (install Python dependencies, initialize the database, start the server with python app.py) and the frontend (install npm dependencies, run npm run dev). Access the UI at http://localhost:3000 to manage cache entries, test queries, or bulk import CSV files. Use the REST API endpoints (e.g., POST /v1/complete) to process natural language queries programmatically.

Key features of NL Cache Framework

• Semantic similarity search using embeddings
• Supports SQL, API, URL, and workflow templates
• Entity extraction and substitution into templates
• Full CRUD REST API for cache management
• Interactive dashboard for managing and testing entries
• Reasoning trace capture and template validation
• Usage tracking for analytics

Use cases of NL Cache Framework

• Accelerating natural language to SQL query generation
• Caching and reusing API call templates from user questions
• Providing instant structured responses for frequently asked queries
• Enabling offline or low-latency retrieval of precomputed outputs
• Building a knowledge base of query–template pairs for domain-specific applications

FAQ from NL Cache Framework

What does NL Cache Framework do that alternatives don’t?

It combines semantic similarity search with entity substitution and supports multiple template types (SQL, API, URL, workflow) in a single caching framework, along with an interactive dashboard and reasoning trace capture.

What are the runtime dependencies?

Backend requires Python, FastAPI, SQLAlchemy, Sentence-Transformers, PostgreSQL with pgvector, and an LLM service for template generation. Frontend requires Node.js, Next.js, and npm.

Where are cached data stored?

All cache entries, embeddings, and usage logs are stored in a PostgreSQL database with pgvector extension for vector similarity search.

Does the framework support authentication or authorization?

The README does not mention any built-in authentication or authorization mechanisms. The API endpoints are exposed without security details.

What transport protocol does the API use?

The API is a RESTful HTTP service built with FastAPI, accessible via standard HTTP requests.