feat: support langchain vision mode

app/api/langchain/tool/agent/agentapi.ts

@@ -37,10 +37,11 @@ import {
   HumanMessage,
   AIMessage,
 } from "@langchain/core/messages";
+import { MultimodalContent } from "@/app/client/api";
 
 export interface RequestMessage {
   role: string;
-  content: string;
+  content: string | MultimodalContent[];
 }
 
 export interface RequestBody {
@@ -324,11 +325,18 @@ export class AgentApi {
       reqBody.messages
         .slice(0, reqBody.messages.length - 1)
         .forEach((message) => {
-          if (message.role === "system")
+          if (message.role === "system" && typeof message.content === "string")
             pastMessages.push(new SystemMessage(message.content));
           if (message.role === "user")
-            pastMessages.push(new HumanMessage(message.content));
-          if (message.role === "assistant")
+            typeof message.content === "string"
+              ? pastMessages.push(new HumanMessage(message.content))
+              : pastMessages.push(
+                  new HumanMessage({ content: message.content }),
+                );
+          if (
+            message.role === "assistant" &&
+            typeof message.content === "string"
+          )
             pastMessages.push(new AIMessage(message.content));
         });
 
@@ -370,7 +378,7 @@ export class AgentApi {
       const MEMORY_KEY = "chat_history";
       const prompt = ChatPromptTemplate.fromMessages([
         new MessagesPlaceholder(MEMORY_KEY),
-        ["user", "{input}"],
+        new MessagesPlaceholder("input"),
         new MessagesPlaceholder("agent_scratchpad"),
       ]);
       const modelWithTools = llm.bind({
@@ -378,9 +386,7 @@ export class AgentApi {
       });
       const runnableAgent = RunnableSequence.from([
         {
-          input: (i: { input: string; steps: ToolsAgentStep[] }) => {
-            return i.input;
-          },
+          input: (i) => i.input,
           agent_scratchpad: (i: { input: string; steps: ToolsAgentStep[] }) => {
             return formatToOpenAIToolMessages(i.steps);
           },
@@ -401,11 +407,15 @@ export class AgentApi {
         agent: runnableAgent,
         tools,
       });
-
+      const lastMessageContent = reqBody.messages.slice(-1)[0].content;
+      const lastHumanMessage =
+        typeof lastMessageContent === "string"
+          ? new HumanMessage(lastMessageContent)
+          : new HumanMessage({ content: lastMessageContent });
       executor
         .invoke(
           {
-            input: reqBody.messages.slice(-1)[0].content,
+            input: [lastHumanMessage],
             signal: this.controller.signal,
           },
           {

app/client/platforms/openai.ts

@@ -399,9 +399,10 @@ export class ChatGPTApi implements LLMApi {
   }
 
   async toolAgentChat(options: AgentChatOptions) {
+    const visionModel = isVisionModel(options.config.model);
     const messages = options.messages.map((v) => ({
       role: v.role,
-      content: getMessageTextContent(v),
+      content: visionModel ? v.content : getMessageTextContent(v),
     }));
 
     const modelConfig = {