Full-Code · Orchestration Frameworksactive

DB-GPT

Type: full-code · Vendor: Eosphoros AI · Language: Python · License: MIT · Status: active · Status in practice: mature

Links: homepage repo

Open-source AI-native data platform built around the AWEL agent workflow language, multi-model support, and natural-language access to databases (Text2SQL, vector + relational + graph stores) for end-to-end data agents.

Description. DB-GPT (17k+ stars, MIT) is Eosphoros AI's open-source data agent platform. It positions itself as the AI-native infrastructure for data products: a unified agent runtime over multiple LLMs (open and proprietary), the AWEL (Agentic Workflow Expression Language) for composing agent pipelines, native multi-store support (vector, relational, graph), Text2SQL with rigorous schema-aware prompting, GraphRAG with both triplet-and-structure indexing (50% token savings vs. Microsoft GraphRAG per their published numbers), and the dbgpts package ecosystem for sharable agents and data apps.

Agent loop shape. AWEL defines the orchestration: each agent is a node, edges declare flow, and the runtime executes the DAG with explicit operator semantics. Each agent runs a tool-augmented loop with access to SQL execution, vector retrieval, graph queries, and Python code. Multi-model serving means routing per-task to the best-fit LLM.

Primary use cases

natural-language data analytics over relational databases (Text2SQL)
agentic data workflows composed in AWEL
AI-native data products combining vector, relational, and graph storage
domain-specific Text2SQL fine-tuning via DB-GPT-Hub

flowchart TD USER[User question] --> AWEL[AWEL workflow runtime] AWEL --> NODE1[Agent node 1] AWEL --> NODE2[Agent node 2] NODE1 --> ROUTE{Tool} ROUTE -->|SQL| SQL[(Relational store<br/>Text2SQL)] ROUTE -->|vector| VEC[(Vector store)] ROUTE -->|graph| GRAPH[(GraphRAG store<br/>triplet + structure)] ROUTE -->|code| PY[Python executor] SQL --> NODE2 VEC --> NODE2 GRAPH --> NODE2 PY --> NODE2 NODE2 --> MODELS{Multi-model serving} MODELS --> OUT[Answer] DBHUB[(DB-GPT-Hub<br/>Text2SQL fine-tuning)] -.train.-> MODELS

Key concepts

AWEL (Agentic Workflow Expression Language) → orchestrator-workers — DAG-based agent orchestration language.
GraphRAG with triplet + structure → graphrag — Combined triplet KG and document-structure graph; 50% token savings vs. Microsoft GraphRAG.
DB-GPT-Hub — Text2SQL model fine-tuning toolkit.
dbgpts → agent-skills — Package ecosystem for sharable data agents and apps.