CentralMind/Gateway

--- title: 'CentralMind' description: 'Build a data platform for LLMs in one day. Securely connect any data source and let AI handle the rest.' --- <h1 align="center">CentralMind Gateway: AI-First Data Gateway</h1> <div align="center"> ## 🛸 Introduction </div> AI agents and LLM-powered applications need fast, secure access to data, but traditional APIs and databases aren’t built for this. We’re building an API layer that automatically generates secure, LLM-optimized APIs on top of your structured data. - Filters out PII and sensitive data to ensure compliance with GDPR, CPRA, SOC 2, and other regulations. - Adds traceability and auditing, so AI applications aren’t black boxes and security teams can control. - Optimizes for AI workloads, supports Model Context Protocol (MCP) with extra meta information to help AI agents understand APIs, caching and security. Our first users are companies deploying AI agents for customer support and analytics, where they need models to access the right data without security risks or compliance headaches.  ## Features - ⚡ **Automatic API Generation** – Creates APIs using LLM based on table schema and sampled data. - 🗄️ **Structured Database Support** – Works with PostgreSQL, MySQL, ClickHouse, and Snowflake. - 🌍 **Run APIs as Rest or MCP Server** – Easily expose APIs in multiple protocols. - 📜 **Swagger & OpenAPI 3.1.0 Documentation** – Automatically generated API documentation and OpenAPI spec. - 🔒 **Automatic PII Cleanup** – Uses regex rules, (Microsoft Presidio coming soon). - ⚡ **Configurable via YAML & Plugin System** – Extend API functionality effortlessly. - 🐳 **Run as Binary or Docker** – Comes with a ready-to-use Helm chart. - 🔑 **Row-Level Security (RLS)** – Restrict data access using Lua scripts. - 🔐 **Authentication** – Supports API keys and OAuth. - 👀 **Observability & Audit Trail** – Uses OpenTelemetry (OTel) for tracking requests. - 🏎️ **Caching** – Supports time-based and LRU caching for efficiency. ## How it Works <div align="center">  </div> ### Connect & Discover Gateway connects to your structured databases like PostgreSQL. Automatically analyzes the schema and samples data to generate an optimized API structure based on your prompt. Ensures security by detecting PII ### Deploy Runs as a standalone binary, Docker container, or Helm chart for Kubernetes. Configuration is managed via YAML and a plugin system, allowing customization without modifying the core code. Supports row-level security (RLS) with Lua scripts, caching strategies like LRU and time-based expiration, and observability through OpenTelemetry. Cleaning PII data using regex rules. ### Use & Integrate Exposes APIs through REST, and MCP with built-in authentication via API keys and OAuth. Designed for seamless integration with AI models, including OpenAI, Anthropic Claude, Google Gemini, and DeepSeek. Automatically provides OpenAPI 3.1.0 documentation for easy adoption and supports flexible query execution with structured access control. ## How to generate Gateway is LLM-model first, i.e. it's designed to be generated via LLM-models. To generate your gateway config simply run discover command with your connection info: 1. Connection info ```yaml hosts: - localhost user: postgres password: password database: mydb port: 5432 ``` 2. Discovery command ```shell gateway \ discover \ --connection-string "postgresql://my_user:my_pass@localhost:5432/mydb" \ --db-type postgres \ --tables "table_name_1,table_name_2" \ --open-ai-key $TOKEN \ --prompt "Generate for me awesome readonly api" ``` 3. Wait for completion ```shell INFO 🚀 API Discovery Process INFO Step 1: Read configs INFO ✅ Step 1 completed. Done. INFO Step 2: Discover data INFO Discovered Tables: INFO - payment_dim: 3 columns INFO - fact_table: 9 columns ... INFO ✅ Step 2 completed. Done. INFO Step 3: Sample data from tables INFO Data Sampling Results: INFO - payment_dim: 5 rows sampled INFO - fact_table: 5 rows sampled ... INFO ✅ Step 3 completed. Done. INFO Step 4: Prepare prompt to AI INFO Prompt saved locally to prompt_default.txt INFO ✅ Step 4 completed. Done. INFO Step 5: Using AI to design API Waiting for OpenAI response... Done! INFO OpenAI usage: Input tokens=3187 Output tokens=14872 Total tokens=18059 INFO API Functions Created: INFO - GET /payment_dim/{payment_key} - Retrieve a payment detail by its payment key INFO - GET /payment_dim - List payment records with pagination INFO - GET /payment_dim/count - Retrieve total count of payment records INFO - GET /fact_table/{payment_key} - Retrieve a transaction detail by its payment key INFO - GET /fact_table - List transaction records with pagination ..... INFO API schema saved to: gateway.yaml INFO ✅ Step 5: API Specification Generation Completed! INFO ✅ All steps completed. Done. INFO --- Execution Statistics --- INFO Total time taken: 2m12s INFO Tokens used: 18059 (Estimated cost: $0.0689) INFO Tables processed: 6 INFO API methods created: 18 INFO Total number of columns with PII data: 2 ``` 4. Explore results, the result would be saved in output file: ```yaml api: name: Awesome Readonly API description: "" version: "1.0" database: type: YOUR_DB_TYPE connection: YOUR_CONNECTION_INFO tables: - name: table_name_1 columns: ... // Columns for this table endpoints: - http_method: GET http_path: /some_path mcp_method: some_method summary: Some readable summary. description: 'Some description' query: SQL Query with params params: ... // List of params for query ``` ## How to run ```shell go build . ./gateway start --config ./example/gateway.yaml rest ``` ### Docker compose ```shell docker compose up ./example/docker-compose.yml ``` ### MCP Protocol Gateway implement MCP protocol, for easy access to your data right from claude, to use it 1. Build binary ```shell go build . ``` 2. Add gateway to claude integrations config: ```json { "mcpServers": { "gateway": { "command": "PATH_TO_GATEWAY_BINARY", "args": [ "start", "--config", "PATH_TO_GATEWAY_YAML_CONFIG", "mcp-stdio" ] } } } ``` 3. Ask something regards your data:  ## Roadmap - 🗄️ **Expand Database Support** – Add support for Redshift, S3, Oracle, MS SQL, Elasticsearch. - 🔐 **MCP with Authentication** – Secure Model Context Protocol with API keys and OAuth. - 🤖 **More LLM Providers** – Integrate Anthropic Claude, Google Gemini, DeepSeek. - 🏠 **Local & On-Prem Deployment** – Allow usage with self-hosted LLMs. - 📦 **Schema Evolution & Versioning** – Track changes and auto-migrate APIs. - 🚦 **Traffic Control & Rate Limiting** – Intelligent throttling for high-scale environments.