DB-GPT
Open-source framework for building AI-native data apps: text-to-SQL, RAG, and multi-agent data workflows.
About
DB-GPT is an open-source framework for building applications where large language models talk to databases. It bundles the pieces such a system needs: text-to-SQL generation with effect optimization, a RAG framework for knowledge-base Q&A, a service-oriented multi-model management layer (SMMF) so you can run private models and keep data in-house, and a multi-agent collaboration framework with AWEL, its agent workflow expression language. On top of the framework sit ready-made apps for chatting with databases, spreadsheets, and visualizations. The project is backed by published research papers on arXiv and had over 20,000 GitHub stars as of October 2026, with commits landing within days of this writing.
The honest caveats matter here. In October 2026, security researchers disclosed two critical remote-code-execution CVEs in DB-GPT's agent data-access layer, and at the time of the report no patched release was listed. Anyone deploying it should pin a fixed version and harden the default configuration before exposing it anywhere. It is also a genuinely complex system: AWEL, SMMF, and the agent layers are powerful but carry real learning curve, so teams that only need plain text-to-SQL may be better served by something smaller.
What it can do
- Text-to-SQL and SQL-to-text with optimization
- Chat interfaces for databases, Excel files, and data visualization
- RAG framework for knowledge-base question answering
- AWEL: agent workflow expression language for multi-agent data pipelines
- SMMF: multi-model management with private-model support
- Generative data analysis and dashboard apps
Setup
- Clone the repository:
git clone https://github.com/eosphoros-ai/DB-GPTand check the install docs in the README. - Choose your model backend: a local/private model via the SMMF layer, or an API-based model.
- Configure your database connections (MySQL, PostgreSQL, and others are supported).
- Start the server and use the web UI for chat-to-database, chat-to-Excel, or build custom data apps with AWEL workflows.
- Before any network exposure, apply the latest security patches and review default credentials and permissions.
Prerequisites
Python 3.10+; a supported database; an LLM backend (local or API); Docker optional for containerized setup.
Use cases
Works with
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: 20,150 GitHub stars (GitHub API), MIT license, pushed 2026-10-11; two arXiv papers document the architecture; framework surface (AWEL, SMMF, agents) corroborated by the README and docs. Strengths: huge adoption, active maintenance, permissive license, research-backed. Docked for: (1) two critical RCE CVEs disclosed Oct 2026 with patches pending at report time; (2) high complexity for teams that only need simple text-to-SQL. License: MIT.
★ 20,150 GitHub stars
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 9.0/10
- Maintenance · recency, releases, issue load
- 9.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · 20,150 GitHub stars (GitHub API), pushed 2026-10-11
- · Two arXiv papers document architecture (AWEL, SMMF, multi-agent framework)
- · Two critical RCE CVEs disclosed Oct 2026 (Forkast); patches pending at report time
- · High complexity for teams needing only simple text-to-SQL
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.