Erupt-AI use langchain4j by erupts · Pull Request #342 · erupts/erupt

erupts · 2026-03-04T11:20:20Z

Function Call change to @tool with Method

Copilot

Pull request overview

This PR migrates the erupt-ai module from a custom Function Call implementation to LangChain4j tools (@Tool / ToolExecutionRequest) and refactors the LLM integration accordingly, while also doing minor template/docs tweaks.

Changes:

Replaced custom AI Function Call framework (AiFunctionManager, AiFunctionCall, OpenAI streaming POJOs) with LangChain4j tool registration/execution (AiToolboxManager, @Tool).
Refactored LLM core + vendor adapters (OpenAI-compatible + Claude/Gemini/Ollama) to use LangChain4j chat/streaming models and new SSE streaming behavior.
Renamed chat entities (Chat → AiChat, ChatMessage → AiChatMessage) and adjusted controllers/services to use the new models.

Reviewed changes

Copilot reviewed 42 out of 44 changed files in this pull request and generated 9 comments.

Show a summary per file

File	Description
erupt-tpl-frame/vue/README.md	Adds Vue render README stub
erupt-tpl-frame/react/README.md	Adds React render README stub
erupt-tpl-frame/angular/README.md	Adds Angular render README + reference link
erupt-core/src/main/java/xyz/erupt/core/util/DateUtil.java	Adjusts ISO_8601 format constant
erupt-core/src/main/java/xyz/erupt/core/service/EruptCoreService.java	Comment formatting only
erupt-core/src/main/java/xyz/erupt/core/controller/EruptDataController.java	Line-ending/formatting change + method rename (`onChange` → `onchange`)
erupt-ai/src/main/java/xyz/erupt/ai/tool/EruptAiToolbox.java	New LangChain4j `@Tool` toolbox methods (schema/user/data/module list)
erupt-ai/src/main/java/xyz/erupt/ai/tool/AiToolboxManager.java	New toolbox scanner + tool invocation dispatcher
erupt-ai/src/main/java/xyz/erupt/ai/service/LLMService.java	Switches message model to LangChain4j + integrates tool execution + new SSE chunking
erupt-ai/src/main/java/xyz/erupt/ai/model/AiChat.java	Renames chat entity and drill link to `AiChatMessage`
erupt-ai/src/main/java/xyz/erupt/ai/model/AiChatMessage.java	Renames chat message entity + token type update
erupt-ai/src/main/java/xyz/erupt/ai/llm/OpenRouter.java	Updates OpenAI-compatible endpoint building (`chatApiPoint`)
erupt-ai/src/main/java/xyz/erupt/ai/llm/Qwen.java	Updates OpenAI-compatible endpoint building (`chatApiPoint`)
erupt-ai/src/main/java/xyz/erupt/ai/llm/GLM.java	Updates OpenAI-compatible endpoint building (`chatApiPoint`)
erupt-ai/src/main/java/xyz/erupt/ai/llm/Doubao.java	Adds new OpenAI-compatible provider adapter
erupt-ai/src/main/java/xyz/erupt/ai/llm/Ollama.java	Replaces OpenAI-compatible HTTP impl with LangChain4j Ollama models
erupt-ai/src/main/java/xyz/erupt/ai/llm/Gemini.java	Replaces OpenAI-compatible HTTP impl with LangChain4j Gemini models
erupt-ai/src/main/java/xyz/erupt/ai/llm/Claude.java	Replaces OpenAI-compatible HTTP impl with LangChain4j Anthropic models
erupt-ai/src/main/java/xyz/erupt/ai/core/SseListener.java	Refactors streaming listener payload to LangChain4j `AiMessage` + `TokenUsage`
erupt-ai/src/main/java/xyz/erupt/ai/core/OpenAi.java	Replaces custom OkHttp OpenAI client with LangChain4j OpenAI models
erupt-ai/src/main/java/xyz/erupt/ai/core/LlmRequest.java	Changes numeric config types (Float → Double)
erupt-ai/src/main/java/xyz/erupt/ai/core/LlmConfig.java	Changes numeric config types (Float → Double)
erupt-ai/src/main/java/xyz/erupt/ai/core/LlmCore.java	Adds LangChain4j chat/streaming integration + tool specification attachment
erupt-ai/src/main/java/xyz/erupt/ai/controller/ChatController.java	Switches controllers to `AiChat` / `AiChatMessage`
erupt-ai/src/main/java/xyz/erupt/ai/controller/McpController.java	Switches MCP tool listing/calls to LangChain4j tools
erupt-ai/src/main/java/xyz/erupt/ai/annotation/AiToolbox.java	Replaces old parameter annotation with type-level toolbox marker
erupt-ai/src/main/java/xyz/erupt/ai/config/AiProp.java	Adds SSE chunk size + delay config
erupt-ai/pom.xml	Replaces OkHttp dependency with LangChain4j modules
erupt-ai-web/src/App.vue	Adds error callback to messages fetch
erupt-ai/src/main/java/xyz/erupt/ai/pojo/ChatUsage.java	Deletes old OpenAI response POJO
erupt-ai/src/main/java/xyz/erupt/ai/pojo/ChatCompletionStreamResponse.java	Deletes old OpenAI response POJO
erupt-ai/src/main/java/xyz/erupt/ai/pojo/ChatCompletionResponse.java	Deletes old OpenAI response POJO
erupt-ai/src/main/java/xyz/erupt/ai/pojo/ChatCompletionMessage.java	Deletes old OpenAI request/role model
erupt-ai/src/main/java/xyz/erupt/ai/pojo/ChatCompletion.java	Deletes old OpenAI request POJO
erupt-ai/src/main/java/xyz/erupt/ai/constants/MessageRole.java	Deletes old role enum
erupt-ai/src/main/java/xyz/erupt/ai/call/AiFunctionManager.java	Deletes custom function-call manager
erupt-ai/src/main/java/xyz/erupt/ai/call/AiFunctionCall.java	Deletes custom function-call interface
erupt-ai/src/main/java/xyz/erupt/ai/call/ParamPromptTemplate.java	Deletes function-call param prompting helper
erupt-ai/src/main/java/xyz/erupt/ai/call/impl/EruptUserInfo.java	Deletes old function-call implementation
erupt-ai/src/main/java/xyz/erupt/ai/call/impl/EruptSchema.java	Deletes old function-call implementation
erupt-ai/src/main/java/xyz/erupt/ai/call/impl/EruptModuleInfo.java	Deletes old function-call implementation
erupt-ai/src/main/java/xyz/erupt/ai/call/impl/EruptList.java	Deletes old function-call implementation
erupt-ai/src/main/java/xyz/erupt/ai/call/impl/EruptDataQuery.java	Deletes old function-call implementation

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-03-04T11:29:37Z

erupt-ai/src/main/java/xyz/erupt/ai/core/LlmCore.java

+        chatMessages.add(0, SystemMessage.from(EruptSpringUtil.getBean(AiProp.class).getSystemPrompt()));
+        return chatModel.chat(chatMessages).aiMessage().text();
+    }
+
+    public void chatSse(LlmRequest llmRequest, List<ChatMessage> chatMessages, Consumer<SseListener> listener) {
+        StreamingChatModel streamingChatModel = this.buildStreamingChatModel(llmRequest, chatMessages, listener);
+        chatMessages.add(0, SystemMessage.from(EruptSpringUtil.getBean(AiProp.class).getSystemPrompt()));
+        List<ToolSpecification> specs = new ArrayList<>();
+        for (Method method : AiToolboxManager.getAiMethodMap().values()) {
+            specs.add(ToolSpecifications.toolSpecificationFrom(method));
+        }
+        ChatRequest request = ChatRequest.builder().messages(chatMessages).toolSpecifications(specs).build();


LlmCore.chat() mutates the provided chatMessages list by inserting the system prompt at index 0. If the same list is reused for subsequent calls (e.g., after tool execution), this will accumulate duplicate system prompts and change message ordering. Prefer creating a new list (or only prepending the system prompt when it is not already present) to avoid side effects.

Suggested change

chatMessages.add(0, SystemMessage.from(EruptSpringUtil.getBean(AiProp.class).getSystemPrompt()));

return chatModel.chat(chatMessages).aiMessage().text();

}

public void chatSse(LlmRequest llmRequest, List<ChatMessage> chatMessages, Consumer<SseListener> listener) {

StreamingChatModel streamingChatModel = this.buildStreamingChatModel(llmRequest, chatMessages, listener);

chatMessages.add(0, SystemMessage.from(EruptSpringUtil.getBean(AiProp.class).getSystemPrompt()));

List<ToolSpecification> specs = new ArrayList<>();

for (Method method : AiToolboxManager.getAiMethodMap().values()) {

specs.add(ToolSpecifications.toolSpecificationFrom(method));

}

ChatRequest request = ChatRequest.builder().messages(chatMessages).toolSpecifications(specs).build();

List<ChatMessage> messagesWithSystem = new ArrayList<>(chatMessages.size() + 1);

messagesWithSystem.add(SystemMessage.from(EruptSpringUtil.getBean(AiProp.class).getSystemPrompt()));

messagesWithSystem.addAll(chatMessages);

return chatModel.chat(messagesWithSystem).aiMessage().text();

}

public void chatSse(LlmRequest llmRequest, List<ChatMessage> chatMessages, Consumer<SseListener> listener) {

StreamingChatModel streamingChatModel = this.buildStreamingChatModel(llmRequest, chatMessages, listener);

List<ChatMessage> messagesWithSystem = new ArrayList<>(chatMessages.size() + 1);

messagesWithSystem.add(SystemMessage.from(EruptSpringUtil.getBean(AiProp.class).getSystemPrompt()));

messagesWithSystem.addAll(chatMessages);

List<ToolSpecification> specs = new ArrayList<>();

for (Method method : AiToolboxManager.getAiMethodMap().values()) {

specs.add(ToolSpecifications.toolSpecificationFrom(method));

}

ChatRequest request = ChatRequest.builder().messages(messagesWithSystem).toolSpecifications(specs).build();

Copilot · 2026-03-04T11:29:38Z

erupt-ai/src/main/java/xyz/erupt/ai/controller/McpController.java

+                .name(code)
+                .arguments(GsonFactory.getGson().toJson(params))
+                .build();
+        return (String) AiToolboxManager.invoke(request);


mcpCall() casts the invoked tool result to String. If any @Tool method returns a non-String (e.g., an object or list), this will throw ClassCastException and fail the MCP request. Consider returning String.valueOf(AiToolboxManager.invoke(request)) or enforcing a String-only return contract when registering tools.

Suggested change

return (String) AiToolboxManager.invoke(request);

return String.valueOf(AiToolboxManager.invoke(request));

Copilot · 2026-03-04T11:29:38Z

erupt-ai/src/main/java/xyz/erupt/ai/tool/EruptAiToolbox.java

+    @Tool("Query erupt model data")
+    public String eruptDataQuery(@P("HQL (Hibernate Query Language)") String hql) {
+        List<?> result = eruptDao.getEntityManager().createQuery(hql).getResultList();
+        return GsonFactory.getGson().toJson(result);


eruptDataQuery executes arbitrary HQL received from the model/client and returns the full result list without limits. This enables data exfiltration and potentially expensive queries (cartesian products / full table scans) if tool calling is enabled. Consider restricting to a safe query subset, enforcing tenant/permission checks, and applying maxResults/timeouts (or removing this tool from MCP/LLM exposure by default).

Copilot · 2026-03-04T11:29:38Z

erupt-ai/src/main/java/xyz/erupt/ai/tool/AiToolboxManager.java

+            if (StringUtils.isNotBlank(request.arguments())) {
+                JsonObject jsonObject = GsonFactory.getGson().fromJson(request.arguments(), JsonObject.class);
+                for (int i = 0; i < method.getParameters().length; i++) {
+                    String paramName = method.getParameters()[i].getName();
+                    if (jsonObject.has(paramName)) {
+                        args[i] = GsonFactory.getGson().fromJson(jsonObject.get(paramName), method.getGenericParameterTypes()[i]);
+                    }
+                }


AiToolboxManager.invoke() extracts argument values using reflection parameter names (method.getParameters()[i].getName()). LangChain4j tool specs typically use @P values as the JSON argument keys, so parameters annotated with @P (e.g. "Erupt Name") will not be found and args will stay null. Use the @P value as the lookup key (fallback to reflection name) and keep the same key naming strategy used when building the ToolSpecification.

Copilot · 2026-03-04T11:29:38Z

erupt-ai/src/main/java/xyz/erupt/ai/tool/AiToolboxManager.java

+            for (Method method : clazz.getDeclaredMethods()) {
+                if (method.isAnnotationPresent(Tool.class)) {
+                    aiMethodMap.put(method.getName(), method);
+                }


aiMethodMap uses method.getName() as the key. If two @AiToolbox classes expose @Tool methods with the same name (or overloaded methods), later entries will overwrite earlier ones and tool execution will call the wrong method. Consider using a unique key (e.g., fully qualified "ClassName#methodName" or LangChain4j ToolSpecification name) and returning that as the tool name to the model/MCP.

Copilot · 2026-03-04T11:29:39Z

erupt-ai/src/main/java/xyz/erupt/ai/core/LlmCore.java

+        chatMessages.add(0, SystemMessage.from(EruptSpringUtil.getBean(AiProp.class).getSystemPrompt()));
+        List<ToolSpecification> specs = new ArrayList<>();
+        for (Method method : AiToolboxManager.getAiMethodMap().values()) {
+            specs.add(ToolSpecifications.toolSpecificationFrom(method));
+        }
+        ChatRequest request = ChatRequest.builder().messages(chatMessages).toolSpecifications(specs).build();


LlmCore.chatSse() always attaches toolSpecifications, even when erupt.ai.enableFunctionCall is false. This still advertises tools to the model and can cause tool execution requests that the rest of the pipeline won’t execute. Gate toolSpecifications creation on AiProp#isEnableFunctionCall (or pass a flag in) so disabling function calls fully disables tool usage.

Suggested change

chatMessages.add(0, SystemMessage.from(EruptSpringUtil.getBean(AiProp.class).getSystemPrompt()));

List<ToolSpecification> specs = new ArrayList<>();

for (Method method : AiToolboxManager.getAiMethodMap().values()) {

specs.add(ToolSpecifications.toolSpecificationFrom(method));

}

ChatRequest request = ChatRequest.builder().messages(chatMessages).toolSpecifications(specs).build();

AiProp aiProp = EruptSpringUtil.getBean(AiProp.class);

chatMessages.add(0, SystemMessage.from(aiProp.getSystemPrompt()));

ChatRequest.ChatRequestBuilder requestBuilder = ChatRequest.builder().messages(chatMessages);

if (aiProp.isEnableFunctionCall()) {

List<ToolSpecification> specs = new ArrayList<>();

for (Method method : AiToolboxManager.getAiMethodMap().values()) {

specs.add(ToolSpecifications.toolSpecificationFrom(method));

}

requestBuilder.toolSpecifications(specs);

}

ChatRequest request = requestBuilder.build();

Copilot · 2026-03-04T11:29:39Z

erupt-ai/src/main/java/xyz/erupt/ai/service/LLMService.java

+                            for (ToolExecutionRequest request : it.getAiMessage().toolExecutionRequests()) {
+                                try {
+                                    Object rtn = AiToolboxManager.invoke(request);
+                                    if (null != rtn) {
+                                        functionCallRtn.add(rtn.toString());
+                                    }
+                                } catch (Exception e) {
+                                    log.error("Function call error", e);
+                                    this.stopSse(emitter, chatMessage, llmModal, e.toString());
+                                }
+                            }
+                            if (functionCallRtn.isEmpty()) {
+                                message = "Completed !";
+                            } else {
+                                for (String s : functionCallRtn) {
+                                    chatMessages.add(AiMessage.from(s));
+                                }
+                                message = llm.chat(llmRequest, chatMessages);
+                                this.sendSseMessage(emitter, message);
+                            }


In the tool-execution error path, stopSse() completes the emitter but execution continues (no return/break) and the code later merges/persists messages and completes the emitter again. After calling stopSse(), the callback should immediately abort further processing to avoid duplicate DB writes and SSE lifecycle errors.

Copilot · 2026-03-04T11:29:39Z

erupt-ai/src/main/java/xyz/erupt/ai/service/LLMService.java

+                    String message = it.getAiMessage().text();
+                    if (aiProp.isEnableFunctionCall()) {
+                        if (it.getAiMessage().hasToolExecutionRequests()) {
+                            List<String> functionCallRtn = new ArrayList<>();
+                            for (ToolExecutionRequest request : it.getAiMessage().toolExecutionRequests()) {
+                                try {
+                                    Object rtn = AiToolboxManager.invoke(request);
+                                    if (null != rtn) {
+                                        functionCallRtn.add(rtn.toString());
+                                    }
+                                } catch (Exception e) {
+                                    log.error("Function call error", e);
+                                    this.stopSse(emitter, chatMessage, llmModal, e.toString());
+                                }
+                            }
+                            if (functionCallRtn.isEmpty()) {
+                                message = "Completed !";
+                            } else {
+                                for (String s : functionCallRtn) {
+                                    chatMessages.add(AiMessage.from(s));
+                                }
+                                message = llm.chat(llmRequest, chatMessages);
+                                this.sendSseMessage(emitter, message);


Tool execution results are appended back into the conversation as AiMessage (chatMessages.add(AiMessage.from(s))) and then sent to the model via llm.chat(...). LangChain4j tool calling expects tool results to be returned as tool-result messages associated with the original ToolExecutionRequest (often including request.id), otherwise the model may ignore the results or repeatedly request the same tool. Use the appropriate LangChain4j message type for tool execution results and include the tool name/request id when adding them to chatMessages.

Copilot · 2026-03-04T11:29:39Z

erupt-ai/src/main/java/xyz/erupt/ai/controller/McpController.java

+                        if (null != p) {
+                            description = p.value();
+                            required.add(parameter.getName());
                        }
+                        McpTool.SchemaProperties schema = new McpTool.SchemaProperties();
+                        schema.setType(McpUtil.toMcp(parameter.getType()));
+                        schema.setDescription(description);
+                        mcpTool.getInputSchema().getProperties().put(parameter.getName(), schema);


MCP tool schemas are built using reflection parameter names (parameter.getName()) for both the JSON property keys and the required list, while @P is used as a description. If @P is intended to define the tool argument name (as in LangChain4j), MCP clients will send the wrong keys and AiToolboxManager.invoke() won’t be able to bind arguments. Use @P value as the schema property key/required entry (fallback to parameter.getName() when @P is absent), and keep it consistent with tool invocation argument parsing.

Suggested change

if (null != p) {

description = p.value();

required.add(parameter.getName());

}

McpTool.SchemaProperties schema = new McpTool.SchemaProperties();

schema.setType(McpUtil.toMcp(parameter.getType()));

schema.setDescription(description);

mcpTool.getInputSchema().getProperties().put(parameter.getName(), schema);

String argName = parameter.getName();

if (null != p) {

description = p.value();

if (description != null && !description.isEmpty()) {

argName = description;

}

required.add(argName);

}

McpTool.SchemaProperties schema = new McpTool.SchemaProperties();

schema.setType(McpUtil.toMcp(parameter.getType()));

schema.setDescription(description);

mcpTool.getInputSchema().getProperties().put(argName, schema);

erupts added 11 commits February 27, 2026 18:37

add erupt-print config API

586f3ee

langchain4j add system prompt to chat

adb7603

langchain4j add AI context

f4bf72f

resupport claude and Gemini

3bd2c2b

Fix timezone bug in ISO_8601

2735b0f

erupt-ai support doubao

6def8f0

erupt-ai support more llm models

4b072fa

erupt-ai support new Function Call

e6e212d

Refactor erupt-ai

d777010

Refactor erupt-ai

77dfde8

Refactor erupt-ai

318d1bc

erupts merged commit 673a0f8 into develop Mar 4, 2026
1 check failed

erupts deleted the erupt-ai-langchain4j branch March 4, 2026 11:21

erupts requested a review from Copilot March 4, 2026 11:22

Copilot started reviewing on behalf of erupts March 4, 2026 11:23 View session

Copilot AI reviewed Mar 4, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Erupt-AI use langchain4j#342

Erupt-AI use langchain4j#342
erupts merged 11 commits intodevelopfrom
erupt-ai-langchain4j

erupts commented Mar 4, 2026

Uh oh!

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Mar 4, 2026

Uh oh!

Copilot AI Mar 4, 2026

Uh oh!

Copilot AI Mar 4, 2026

Uh oh!

Copilot AI Mar 4, 2026

Uh oh!

Copilot AI Mar 4, 2026

Uh oh!

Copilot AI Mar 4, 2026

Uh oh!

Copilot AI Mar 4, 2026

Uh oh!

Copilot AI Mar 4, 2026

Uh oh!

Copilot AI Mar 4, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

	return (String) AiToolboxManager.invoke(request);
	return String.valueOf(AiToolboxManager.invoke(request));

Uh oh!

Conversation

erupts commented Mar 4, 2026

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Mar 4, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 4, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 4, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 4, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 4, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 4, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 4, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 4, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 4, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants