"What were our top 10 customers last quarter by revenue, as a bar chart?" DB-GPT translates that to SQL, runs it against your database, and renders the chart. No SQL knowledge required. Fully local. MIT licensed. 17K GitHub stars — almost no English content. DB-GPT is an open-source framework that puts a natural language interface on top of your databases. You connect PostgreSQL, MySQL, SQLite, or others — then ask questions in plain English. It generates the SQL, executes it, and can visualize results automatically. Think Metabase meets AI, but fully self-hosted and free. PostgreSQL · MySQL · MariaDB · SQLite · ClickHouse · DuckDB · Spark SQL Clone the repo, copy .env.example to .env , add your database connection string, then docker compose up -d . Open localhost:5670 → admin / admin → Settings → Database → Add → paste connection string. Full docker-compose: chinese-ai-tools-english-guide/tools/db-gpt Once your database is connected: DB-GPT generates SQL for each, runs it, returns results. Charts render automatically in the UI. Settings → LLM Provider → Ollama → Base URL: http://ollama:11434/v1 → Model: llama3 For best SQL accuracy use sqlcoder — fine-tuned specifically for SQL generation. Pull it with docker exec -it ollama ollama pull sqlcoder . Both let you query databases with natural language: DB-GPT if you want a complete self-contained app. Vanna.ai if you're embedding the capability programmatically into your own product.
A ready-to-import workflow JSON is in the repo (integration/n8n-workflows/db-gpt-query.json
). POST {question, db_name}
→ returns {answer, sql, data}
. → chinese-ai-tools-english-guide Previous articles in this series: Your data never leaves your machine. No API keys, no cloud, no SQL knowledge needed.