spring-ai-alibaba/dataagent

1. Purpose and deployable entry points

DataAgent is an enterprise AI data analyst: conversational NL-to-SQL, optional sandboxed Python analysis, and report/chart generation. The backend is a Java 17+/Spring Boot application rooted at data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/DataAgentApplication.java; its API surface is exposed by controllers such as data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/controller/ChatController.java and data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/controller/GraphController.java.

A separate Nuxt frontend provides the browser UI, entering through data-agent-frontend-nuxt/app/app.vue and route pages including data-agent-frontend-nuxt/app/pages/chat.vue. Container deployment is defined by docker-file/docker-compose.yml, docker-file/Dockerfile-backend, and docker-file/Dockerfile-frontend.

2. API, security, and application boundary

Controllers separate HTTP/WebFlux concerns from domain services for agents, data sources, knowledge, models, prompts, sessions, and graph execution. Responses use API/streaming-oriented value objects, including data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/vo/ChatResponse.java and data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/vo/SessionUpdateEvent.java.

Reactive API-key authentication is configured at data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/config/WebFluxSecurityConfiguration.java, implemented through data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/security/AgentApiKeyReactiveAuthenticationManager.java, and backed by credential lifecycle handling in data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/security/ApiKeyCredentialService.java.

3. Conversational graph orchestration

GraphServiceImpl is the orchestration boundary from chat requests into a stateful Spring AI Alibaba Graph workflow: data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/service/graph/GraphServiceImpl.java. Streaming and multi-turn state are managed by data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/service/graph/Context/StreamContext.java and data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/service/graph/Context/MultiTurnContextManager.java.

The workflow decomposes requests into graph nodes and dispatchers: recognize intent/feasibility, enhance the query, recall schema/evidence, validate semantics, plan and execute, then produce a report. Core node implementations live in data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/workflow/node/PlannerNode.java, data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/workflow/node/SqlGenerateNode.java, and data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/workflow/node/ReportGeneratorNode.java. Human feedback is an explicit control-flow branch via data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/workflow/node/HumanFeedbackNode.java.

4. Data access and NL2SQL execution

Agent-specific datasource configuration and schema metadata drive SQL generation and execution. data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/service/nl2sql/Nl2SqlServiceImpl.java provides the NL2SQL service boundary; prompt templates are externalized under data-agent-management/src/main/resources/prompts/, including new-sql-generate.txt and planner.txt.

The connector layer abstracts database access, dialect DDL, pools, and result conversion through data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/connector/SqlExecutor.java, data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/connector/accessor/AccessorFactory.java, and data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/connector/pool/DBConnectionPoolFactory.java. Concrete support includes MySQL, PostgreSQL, Oracle, SQL Server, Hive, H2, and Dameng. MySQL is the primary management-store runtime expectation, with schema at data-agent-management/src/main/resources/sql/schema.sql.

5. Knowledge, retrieval, and model runtime

Business knowledge, agent knowledge, semantic models, logical relations, and datasource schema are persisted as domain entities and used as retrieval context. The ingestion lifecycle is centered on data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/service/knowledge/AgentKnowledgeServiceImpl.java and asynchronous embedding events in data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/event/AgentKnowledgeEventListener.java.

Vector retrieval is abstracted behind data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/service/vectorstore/AgentVectorStoreServiceImpl.java; configurations indicate Milvus and Elasticsearch alternatives in data-agent-management/src/main/resources/application-milvus.yml and data-agent-management/src/main/resources/application-elasticsearch.yml. Hybrid Elasticsearch retrieval and reciprocal-rank fusion are implemented in data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/service/hybrid/retrieval/impl/ElasticsearchHybridRetrievalStrategy.java and data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/service/hybrid/fusion/impl/RrfFusionStrategy.java.

LLM and embedding providers are dynamically registered/configured via data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/service/aimodelconfig/AiModelRegistry.java and data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/service/aimodelconfig/DynamicModelFactory.java.

6. Analysis output, integrations, and frontend flow

Workflow plans may execute generated Python after SQL extraction; the isolation boundary is data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/service/code/sandbox/SaaSandboxPythonCodeExecutorService.java and its task runtime data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/service/code/sandbox/runtime/SaaSandboxRuntime.java. Docker is materially required for this sandbox path.

The backend can also operate as an MCP tool server through data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/service/mcp/McpServerService.java, and emits Langfuse-oriented node tracing through data-agent-management/src/main/java/com/alibaba/cloud/ai/dataagent/service/langfuse/NodeTracingLifecycleListener.java.

The Nuxt UI calls backend service modules, maintains chat state, and renders streamed workflow progress/results: data-agent-frontend-nuxt/app/services/chat/index.ts, data-agent-frontend-nuxt/app/stores/chat.ts, and data-agent-frontend-nuxt/app/components/chat/ChatWorkflowTimeline.vue. Markdown reports and ECharts rendering are handled at data-agent-frontend-nuxt/app/components/chat/ChatMarkdownReport.vue and data-agent-frontend-nuxt/app/utils/markdown/markdown-plugin-echarts.ts.