DB-GPT

DB-GPT is an open-source agentic AI data assistant built for the next generation of AI + Data products. Released under the MIT License by the eosphoros-ai organization, it lets users connect to databases, CSV/Excel files, data warehouses, and knowledge bases, then ask questions in natural language while the AI autonomously writes SQL, executes Python analysis, and generates charts, dashboards, and HTML reports. The project has accumulated over 19,000 GitHub stars since its creation in April 2023 and is actively maintained, with the latest release being v0.8.1 in June 2026.

What It Is

DB-GPT is an open-source framework and platform that combines agentic task planning, autonomous SQL and code execution, multi-source data access, and reusable skills into a single AI-native data assistant. It is also a foundation for building custom AI data agents, workflows, and applications using its AWEL (Agentic Workflow Expression Language), RAG, and multi-model management components. Users can install it via a one-line shell script, pip (dbgpt-app), Docker, or from source, and access it through a local web UI at localhost:5670.

Core Architecture and Components

DB-GPT is organized around several interconnected subsystems:

• AWEL — Agentic Workflow Expression Language for orchestrating structured AI + data pipelines
• Agents — Autonomous task execution with tools, memory, planning, and multi-agent collaboration
• RAG — Knowledge-enhanced reasoning across documents, graphs, and private knowledge bases
• SMMF — Service-oriented Multi-model Management Framework for managing open-source and API-based LLMs
• DB-GPT-Hub — Text2SQL fine-tuning workflows for domain adaptation and model improvement
• Connections — Connectors to relational databases, CSV/Excel files, warehouses, and external data systems
• Sandbox — Isolated execution environments for safe code and tool use

Supported Models and Providers

DB-GPT's SMMF layer supports a wide range of LLM providers and models, including DeepSeek (R1, V3, Coder V2), Qwen (Qwen3, QwQ, Qwen2.5-Coder), GLM (GLM-Z1, GLM-4), Llama (Meta-Llama-3.1), Gemma, Yi, Mixtral, Phi-3, Starling, SOLAR, Baichuan, InternLM, BLOOM, Falcon, and XVERSE. The Text2SQL fine-tuning module (DB-GPT-Hub) supports LLaMA, LLaMA-2, BLOOM, BLOOMZ, Falcon, Baichuan, Baichuan2, InternLM, Qwen, XVERSE, and ChatGLM2.

What You Can Build and Do

• Analyze CSV and Excel files and generate visual reports
• Connect to relational databases and produce profiling reports
• Ask business questions in natural language and receive auto-generated SQL
• Perform financial report analysis with code, charts, and narrative summaries
• Create and reuse SQL analysis skills and domain-specific workflows
• Combine code, SQL, retrieval, and tool calls in a single agentic workflow
• Build AI + Data assistants for teams or products using the AWEL and agent APIs

Update: v0.8.1

The latest stable release is v0.8.1, published on June 18, 2026. The documentation site currently tracks v0.8.0 as the primary stable version, with a dev branch for upcoming changes and archived docs for v0.7.5. The repository was last pushed to on June 29, 2026, indicating active development. The project also has a published academic paper on arXiv (arXiv:2312.17449) describing the overall architecture, and a second paper (arXiv:2412.13520) covering the ROMAS role-based multi-agent system for database monitoring built on top of DB-GPT.