fix(AIC-3506): strip tools from LangChain model constructor kwargs

packages/langchain-agents/src/__tests__/handler.test.ts

@@ -1016,7 +1016,6 @@ describe('model source', () => {
     expect(MockChatOpenAI).toHaveBeenCalledWith({
       temperature: 0.2,
       max_tokens: 512,
-      tools: ['openai-tool'],
       model: 'gpt-4o',
     });
     expect(mockCreateAgent).toHaveBeenCalledWith(expect.objectContaining({ model: constructed }));

packages/langchain-agents/src/__tests__/native-graph.test.ts

@@ -180,7 +180,11 @@ describe('toLangGraph', () => {
 
   it('spreads model.parameters into the default ChatOpenAI constructor', async () => {
     const root = makeNode('root', '', []);
-    root.config.model.parameters = { temperature: 0.2, max_tokens: 512 };
+    root.config.model.parameters = {
+      temperature: 0.2,
+      max_tokens: 512,
+      tools: [{ name: 'openai-tool' }],
+    };
     const def = makeGraphDef([root], {}, 'root');
     MockChatOpenAI.mockClear();
     await toLangGraph(Promise.resolve(def)).invoke('hi');

packages/langchain-agents/src/handler.ts

@@ -263,7 +263,8 @@ function modelConstructorArgs(config: AiConfigRep, fallbackName: string): Record
   const parameters = {
     ...(config.model?.parameters && typeof config.model.parameters === 'object' ? config.model.parameters : {}),
   };
-  if ((config.provider?.name ?? '').toLowerCase() === 'bedrock') delete parameters.tools;
+  // Tools are bound from config.tools. A tools key here is forwarded raw and rejected.
+  delete parameters.tools;
   return { ...parameters, model: resolvedModelName(config, fallbackName) };
 }
 

packages/langchain-agents/src/native-graph.ts

@@ -142,13 +142,19 @@ export const toLangGraph = (
     const toolHandlers = opts?.toolHandlers ?? {};
     const modelFactory =
       opts?.modelFactory ??
-      ((node) =>
-        new ChatOpenAI({
+      ((node) => {
+        const parameters = {
           ...(node.config.model.parameters && typeof node.config.model.parameters === 'object'
             ? node.config.model.parameters
             : {}),
+        };
+        // Tools are bound from the node config. A tools key here is forwarded raw and rejected.
+        delete parameters.tools;
+        return new ChatOpenAI({
+          ...parameters,
           model: node.config.model.name,
-        }));
+        });
+      });
     const ldContext = opts?.context;
 
     return trace.getTracer('@launchdarkly/ai-langchain-agents').startActiveSpan('launchdarkly.graph', async (span) => {

packages/langchain-messages/src/__tests__/handler.test.ts

@@ -1291,7 +1291,6 @@ describe('model source', () => {
     expect(ChatOpenAI).toHaveBeenCalledWith({
       temperature: 0.2,
       max_tokens: 512,
-      tools: ['openai-tool'],
       model: 'gpt-4o',
     });
   });

packages/langchain-messages/src/handler.ts

@@ -173,7 +173,8 @@ function modelConstructorArgs(config: AiConfigRep, fallbackName: string): Record
   const parameters = {
     ...(config.model?.parameters && typeof config.model.parameters === 'object' ? config.model.parameters : {}),
   };
-  if ((config.provider?.name ?? '').toLowerCase() === 'bedrock') delete parameters.tools;
+  // Tools are bound from config.tools. A tools key here is forwarded raw and rejected.
+  delete parameters.tools;
   // Name from the config always wins over a colliding `model` key in the parameter bag.
   return { ...parameters, model: resolvedModelName(config, fallbackName) };
 }
@@ -184,7 +185,7 @@ function modelConstructorArgs(config: AiConfigRep, fallbackName: string): Record
  * Otherwise, the provider and model name from the AI config are used to
  * instantiate the appropriate model via a dynamic import, so that neither
  * @langchain/openai, @langchain/anthropic, nor @langchain/aws is a hard
- * dependency. Parameters are passed through unchanged.
+ * dependency. Parameters are passed through, except `tools`, which is bound separately.
  */
 async function resolveBaseModel(config: AiConfigRep, llm?: LangChainModelSource): Promise<BaseChatModel> {
   const invocation = configForModelCall(config);