Vue d’ensemble
The project targets AI-native data products and combines agents, reusable skills, AWEL workflows, retrieval-augmented generation (RAG), tool use, and multi-model support. It can work with databases, CSV and Excel files, warehouses, documents, and knowledge bases. Repository metadata reports 19,622 stars, 2,853 forks, 421 open issues, and an active, non-archived main branch.
Fonctionnalités clés
- Agentic task planning, iterative reasoning, tool use, and step-by-step execution
- Natural-language SQL generation and autonomous database querying
- Python-based data cleaning, transformation, metric calculation, and visualization
- Connections to databases, warehouses, CSV and Excel files, documents, and knowledge bases
- RAG document parsing and knowledge-base workflows
- Reusable skills and domain-specific analysis workflows
- Agent orchestration and customizable AWEL execution flows
- Charts, dashboards, HTML reports, profiling reports, and narrative summaries
- Sandboxed code execution and controlled tool use
- Support for hosted APIs, OpenAI-compatible interfaces, and documented local-model deployment options
Prérequis, installation et démarrage rapide
Utilisation
Compatibilité des modèles et cas d’usage
The default package includes OpenAI-compatible LLM support and DashScope/Tongyi support. The README explicitly lists DeepSeek, Qwen, GLM, Llama, Gemma, Yi, Starling, SOLAR, Mixtral, and Phi model families, and references local serving through vLLM and llama.cpp. Text2SQL fine-tuning is listed for LLaMA, LLaMA-2, BLOOM, BLOOMZ, Falcon, Baichuan, Baichuan2, InternLM, Qwen, XVERSE, and ChatGLM2. Kimi 2.5 through Moonshot and MiniMax through an OpenAI-compatible API are included in quick-start examples. Compatibility details may vary by deployment path and model; exhaustive version and hardware requirements are not stated in the repository metadata.
Licence et notes sur les risques
The repository metadata identifies the license as MIT and links to https://api.github.com/licenses/mit. The README also labels the project as MIT licensed. Dependency, model-weight, dataset, and external-service licenses must be evaluated separately; their terms are not stated in the repository metadata.
Publication et maintenance
Release history is available at https://github.com/eosphoros-ai/DB-GPT/releases. Specific release versions and changes were not supplied.