v1.1.2.2

AgentScope Integration

AgentScope Java is an agent-oriented programming framework for building LLM-powered applications.

• AgentScope – AgentScopeAgent wraps AgentScope ReActAgent as a BaseAgent for use in graph workflows.

# Add the following dependency to ochestrate AgentScope using SAA Graph
<dependency>
  <groupId>com.alibaba.cloud.ai</groupId>
  <artifactId>spring-ai-alibaba-starter-agentscope</artifactId>
  <version>1.1.2.2</version>
</dependency>

Multiagent Patterns

• Subagent – Main orchestrator delegates tasks to specialized sub-agents (codebase explorer, web researcher, etc.) via Task/TaskOutput tools; supports both Markdown and API-defined sub-agents.
• Supervisor – Central supervisor agent wraps calendar and email agents as tools (AgentTool), invokes them on demand, and synthesizes results.
• Skills – Single agent uses read_skill to load skill content on demand; system prompt shows only skill descriptions for progressive disclosure and smaller context.
• Routing
  • Routing (simple) – LlmRoutingAgent classifies the user query, invokes specialist agents (GitHub, Notion, Slack) in parallel, then synthesizes a single answer.
  • Routing (graph) – LlmRoutingAgent as a StateGraph node with preprocess/postprocess and an internal merge node for routing and result synthesis.
• Handoffs
  • Handoffs (single-agent) – One ReactAgent advances steps via state (e.g. current_step); a ModelInterceptor injects step-specific system prompt and tools per turn.
  • Handoffs (multi-agent) – Sales and support agents as graph nodes; handoff tools update active_agent and conditional edges route between agents.
• Workflow – Custom workflow examples: RAG (rewrite → retrieve → prepare → agent) and SQL agent (list_tables → get_schema → run_query) as graph-based flows.

Multimodal & Voice Agent

• Voice Agent – Sandwich architecture (STT → ReactAgent → TTS): WebSocket-based real-time voice with DashScope ASR and CosyVoice TTS, plus text input; agent uses sandwich-order tools and streams events (stt_chunk, agent_chunk, tts_chunk).
• Multimodal – Vision (image in/out) and TTS: DashScope vision models for image understanding and ReactAgent with media input; image generation via tools (Wanx), TTS via DashScopeAudioSpeechModel; ToolMultimodalResult for structured tool responses (url/base64).

v1.1.2.1

Please use 1.1.2.2 instead. 1.1.2.1 was reported with several bugs that have been fixed in 1.1.2.2.

v1.1.2.0

Summary

This release upgrades Spring AI to 1.1.2, adds Agent Skills support for Agent Framework, adds parallel sub-agent execution for workflow agents (LlmRouting, Supervisor), extends the graph with parallel conditional edges, parallel aggregation strategy (allOf and anyOf), async tool execution, and tool returnDirect. It also includes serialization and MergeStrategy fixes, admin/UI fixes, and documentation updates.

Please see the documentation for more details.

Features

Agent

• Agent Skills in ReactAgent (#3975) — Integrated skills capabilities in ReactAgent, supports skill and too progressive disclosure.
• Built-in multi-agent pattern support — Supports routing, supervisor, supervisor(agent as tool), handoffs, skills, etc., multi-agent patterns; added parallel execution support for routing and supervisor patterns.
• Flow agent hook support (#3983) — All flow agent (routing, supervisor, loop, sequential, etc) supports customized Hooks.
• New Agent API (#4031) — New streamMessages API for simplified message streaming; New overload invokeand callmethods for extra inputs.
• Async and parallel tool execution (#4037) — Support tool execution in parallelization and asynchronous tool execution .
• Tool returnDirect (#4139) — Support for tool returnDirect behavior.
• ToolContextHelper (#4163) — Helper class for metadata access in tool context (fixes #3110).

Graph

• Parallel conditional edges (#4027) — Conditional edges support parallel execution (fixes #3917).
• Aggregation strategy for parallel branches (#4028) — New aggregation strategy (AllOf and AnyOf) for parallel branches.
• Batch addEdge for parallel branches (#3938) — Convenient batch addEdge methods for parallel branch definition in StateGraph.
• interruptAfter hook (#3864) — InterruptableAction supports interruptAfter hook (fixes #3447, #3904).
• Async tool execution for AgentToolNode (#4037 / #3988) — Async tool execution support for AgentToolNode.
• Full output for streaming node (#3941) — Streaming nodes now emit full output including _FINISHED OutputType.

Sandbox & Studio

• SAA Sandbox (#4069) — New SAA Sandbox module that provides safe and isolated tool execution.
• Built-in RenderTemplate (#3935) — Built-in customized RenderTemplate implementation.
• Studio CORS optional (#4154) — CORS configuration is now optional in Studio.

---

Fixes

DashScopeChatModel

• Connection reset reported in #3942 (spring-ai-alibaba/spring-ai-extensions#140) — Set default HTTP client for DashScopeApi to better manage connections, avoid Connection Reset issue when the connection is closed by model server.

Graph

• Serialization (#3969) — Fixed class bug in serialization (fixes #3895, #3931)..
• MergeStrategy (#4129) — Corrected Optional handling and enforced strict type checks in MergeStrategy (fixes #3909).

Agent

• Instruction (#3974) — Added instruction handling; fixed CI and tests.
• AgentLlmNode model options (#3986) — Prevent AgentLlmNode from overriding default model options; ensure internalToolExecution(false) is set for ChatOptions (fixes #3903).
• selectorModel (#4045) — Added selectorModel support (fixes #4026).
• systemPrompt (#4067) — Fixed system prompt handling.
• ParallelAgent maxConcurrency (#4053) — Ensured maxConcurrency function of ParallelAgent is implemented.
• jumpTo (#4081) — Ensured jumpTo operates correctly.
• ModelRequest builder systemMessage (#4138) — ModelRequest builder no longer loses systemMessage.
• NacosReactAgentBuilder (#4093) — Fixed NacosReactAgentBuilder to properly merge local tools.
• AgentLlmNode tool callbacks (#4112) — Corrected tool callbacks filtering logic in AgentLlmNode.

Admin & UI

• Admin package rename (#3977) — Renamed @agentscope-ai/flow to @spark-ai/flow (fixes #3949).
• Admin startup/compilation (#4123) — Fixed spring-ai-alibaba-admin startup and compilation issues.
• CustomInputsControl (#4106) — Added variableList prop to CustomInputsControl.

Dependencies

• Spring AI version — Spring Ai upgraded to 1.1.2
• Spring AI Alibaba Extensions version — Spring Ai Alibaba Extensions upgraded to 1.1.2.1

Upgrade Notes

• Spring AI: This release uses Spring AI 1.1.2. Align your project's Spring AI version accordingly.
• Admin: If you use the admin flow UI, update the dependency from @agentscope-ai/flow to @spark-ai/flow.

v1.1.0.0

The official Spring AI Alibaba release that built on Spring AI 1.1.0 and Spring AI Alibaba Extensions 1.1.0.0.

Please check the documentation on our website for usage and upgrade guide.

Please submit issues and follow the updates on the official website. We will try our best to keep the documentation up to date in the next few weeks.

1.1.0.0-RC1

v1.1.0.0-RC1

1.1.0.0 rc1 release

v1.1.0.0-M5

Spring AI Alibaba 1.1.0.0-M5 Release Notes

Spring AI Alibaba 1.1.x provides the core features for building enterprise-level agent applications, built upon the practical experience learned from building agent with 1.0.x in Alibaba and numerous enterprises. 1.1.x offers various agent development modes such as Agentic, Multi-agent, and Workflow orchestration, helping developers construct enterprise-grade applications.

The official website and documentation have also been upgraded following 1.1.x.

Agent Framework

ReactAgent

• ReAct paradigm (Reasoning + Acting) implementation with reasoning-action loop iteration
• Tool calling capabilities with FunctionToolCallback and ToolContext for state access
• Streaming output support with real-time execution progress and token streaming
• Structured output with outputSchema and outputType for format definition

Context Engineering

• Model context: Dynamic system prompts, message history management, tool selection, model configuration
• Tool context: Access state, store, and runtime configuration through ToolContext
• Lifecycle context: Hook mechanism (before/after agent/model) for context injection and modification

Memory Management

• Short-term memory: Session-level persistence based on Checkpointer (MemorySaver, RedisSaver, etc.)
• Long-term memory: MemoryStore supports cross-session data storage with namespace and key-value management
• Message management: Strategies for message trimming, deletion, and summarization to address context window limitations

Human-in-the-Loop

• HumanInTheLoopHook supports manual approval workflow for tool calls
• Three decision types: approve, edit, and reject
• Interrupt recovery mechanism based on Checkpointer

Hooks & Interceptors

• Built-in Hooks: SummarizationHook, ModelCallLimitHook, PIIDetectionHook, HumanInTheLoopHook
• Built-in Interceptors: ToolRetryInterceptor, TodoListInterceptor, ToolSelectionInterceptor, ContextEditingInterceptor
• Custom extensions: ModelHook, AgentHook, ModelInterceptor, ToolInterceptor

Multi-agent Support

Flow Agents

• SequentialAgent: Sequential execution of multiple agents with state passing
• ParallelAgent: Parallel execution of multiple agents with custom merge strategies
• LlmRoutingAgent: LLM-based intelligent routing for dynamic selection of the most suitable sub-agent
• FlowAgent abstraction: Support for custom multi-agent collaboration patterns

Agent as Tool

• AgentTool: Encapsulates ReactAgent as a tool for other agents to call
• Input/output control: Supports inputSchema, inputType, outputSchema, outputType
• Context isolation: Sub-agents execute independently, results returned to the controller agent

A2A (Agent-to-Agent)

• A2A protocol support: Enables distributed agent communication
• Nacos integration: Agent registration and discovery with load balancing support
• A2aRemoteAgent: Remote agent invocation capability

Enhanced Graph Engine

StateGraph

• Core concepts: State, Node, Edge
• Custom nodes: Support for NodeAction and NodeActionWithConfig interfaces
• Conditional routing: addConditionalEdges supports state-based dynamic branching
• Parallel execution: Supports multi-node parallel execution and result aggregation

ReactAgent Integration

• ReactAgent.asNode(): Embeds agent as SubGraph node in workflows
• Context passing: Controls message history and reasoning content transmission
• Hybrid orchestration: Mix agent nodes with regular nodes

Streaming & Performance

• Streaming response: Real-time streaming output of node execution status and tokens
• Performance optimization: 15x+ throughput improvement compared to Dify platform (150 RPS vs 10 RPS)

RAG (Retrieval-Augmented Generation)

• Two-step RAG: Fixed retrieval-generation pipeline, suitable for FAQ and document bots
• Agentic RAG: Agent-driven dynamic retrieval with multi-tool access
• Hybrid RAG: Complete workflow combining query enhancement, retrieval validation, and answer verification
• Knowledge base construction: Supports document loading, text splitting, and vector store integration

Upgrade Spring AI to 1.1.0 latest milestone version

• MCP (Model Context Protocol) upgrade, including authentication, connection retry, streamable, stateless, etc.
• Bugfixes and enhancements

Breaking Changes and Migration Guide

Please check documentation on the official website for breaking changes and how to migrate to 1.0.x.

v1.1.0.0-M4

1.1.0 M4 release

v1.0.0.4

This release marks a significant evolution of our AI framework, featuring a rebuilt Agent Graph engine and the introduction of Agent-to-Agent (A2A) communication.

• Rebuilt Agent Graph Engine: We've refactored the core Agent API and shifted to a Flux-based reactive stream architecture.
• Agent-to-Agent (A2A): A2A client and server communication with Nacos integration for seamless remote agent discovery.
• Spring AI 1.0.1: The project is now based on Spring AI 1.0.1 and includes widespread stability fixes.

v1.0.0.3

release 1.0.0.3

v1.0.0.2

We are excited to announce the release of Spring AI Alibaba 1.0.0.2, the first GA version! Built on Spring AI and deeply integrated with Bailian, Spring AI Alibaba supports development of ChatBot, Workflow, and Multi-agent applications.

Key Features

1. Comprehensive Framework Integration

• Integration with Alibaba Cloud: Seamlessly connect to Alibaba services, including Bailian's Dashscope, supporting various mainstream AI models like Qwen and Deepseek.

2. Multi-agent Framework

• Graph-Based Multi-agent Framework: Inspired by Langgraph, Spring AI Alibaba Graph makes building workflows and multi-agent applications effortlessly.

3. Enterprise-ready AI Agent Solutions

• Nacos MCP Registry
  • Distributed MCP discovery and load balance
  • Dynamic API to MCP proxy
• NL2SQL for transforming natural language queries into SQL statements.
• Observability

Developer Experience

Quick Start Guide

Kick-start your first AI application with Spring AI Alibaba by adding the following dependency to your Spring Boot project:

<dependencyManagement>
    <dependencies>
        <dependency>
            <groupId>com.alibaba.cloud.ai</groupId>
            <artifactId>spring-ai-alibaba-bom</artifactId>
            <version>1.0.0.2</version>
            <type>pom</type>
            <scope>import</scope>
        </dependency>
    </dependencies>
</dependencyManagement>

<dependencies>
    <dependency>
        <groupId>com.alibaba.cloud.ai</groupId>
        <artifactId>spring-ai-alibaba-starter-dashscope</artifactId>
    </dependency>
</dependencies>

Spring Boot Starters

• spring-ai-alibaba-starter-dashscope - Integrated model service adaptation
• spring-ai-alibaba-graph-core - Core components of the AI Graph framework
• spring-ai-alibaba-starter-nl2sql - Natural language to SQL conversion component
• spring-ai-alibaba-starter-memory - Session memory component
• spring-ai-alibaba-starter-nacos-mcp-client - Nacos MCP client; recommended version: Nacos 3.0.1. For Nacos 2 Server users, please use the older version (spring-ai-alibaba-starter-nacos2-mcp-client).
• spring-ai-alibaba-starter-nacos-mcp-server - Nacos MCP server; recommended version: Nacos 3.0.1. For Nacos 2 Server users, please use the older version (spring-ai-alibaba-starter-nacos2-mcp-server).
• spring-ai-alibaba-starter-nacos-prompt - Nacos Prompt management
• spring-ai-alibaba-starter-arms-observation - ARMS observability
• Community components
  • spring-ai-alibaba-starter-tool-calling-* - Tool calling component
  • spring-ai-alibaba-starter-document-reader-* - Document reading component
  • spring-ai-alibaba-starter-vector-store-* - Vector Store component

Example Projects

The official community has developed a Playground example that includes a complete Frontend UI and Backend based on Spring AI Alibaba. Playground allows users to experience all core framework capabilities such as chat, observability, multi-round conversations, image generation, multimodality, tool calling, MCP, and RAG.

For more usage cases, see Spring AI Alibaba Examples Repository.

Agent Products built with Spring AI Alibaba

• JManus
• DeepResearch

Merged Contributors

• chickenlj
• yuluo-yx
• answeropensource
• brianxiadong
• zhangshenghang
• Aias00
• PolarishT
• jianyi-gronk
• disaster1-tesk
• sincerity-being
• xiaohai-78
• conghuhu
• Cirilla-zmh
• LoongYun
• GTyingzi
• VLSMB
• AlbumenJ
• Yeaury
• wxbty
• kevinlin09
• ALSoryu
• big-mouth-cn
• Sunrisea
• robocanic
• fuyou-lxm
• tonycody
• Helltab
• SCMRCORE
• CoderSerio
• future0923
• wblu214
• rainerWJY
• CZJCC
• Wangzy455
• robinyeeh
• jackxu2011
• SDUWYS
• kone-net
• Sparkle6979
• xixingya
• flystar32
• ppbcn
• erasernoob
• zuoc1993
• ikun-Lg
• inlines10
• cxhello
• wstever
• zxuexingzhijie
• w9lsky
• ev1lQuark
• Vishvesh-Codehunt
• archiesingla
• finefuture
• dolphin57
• ericql
• drawwon
• psxjoy
• hongshuo-wang
• superhandsomeg
• stillmoon
• skippy-404
• sixiyida
• shiyiyue1102
• ran222
• qnnn
• timothy-huo
• wanlihang
• kaori-seasons
• xxsc0529
• ZeroVd
• zhengbiyu0218
• heliang666s
• StudiousXiaoYu
• xuanmiss
• liu-zhen-wei
• logicwu0
• lambochen
• jenspapenhagen
• gzlicanyi
• daichangya
• littleahri
• NewGK
• Matthew-Miao
• LofiSu
• LjyYano
• JunJieLiu51520
• HY-love-sleep
• DomKing
• CandyLedge

Thank you to all community contributors and users for your support! We look forward to your feedback and continued participation as we advance the Spring AI Alibaba framework together!