Function●Since v1.5

toolErrorMiddleware

Converts selected tool execution errors into error ToolMessages.

Handling is opt-in: ToolErrorMiddlewareConfig.onError must return content for an error to be sent to the model. Returning nothing propagates the original error unchanged. LangGraph control-flow signals always propagate and are never passed to onError.

Prefer returning content that identifies the error type without including the raw error message, which may contain sensitive or internal details.

This middleware does not retry. To retry before handling an error, place toolRetryMiddleware({ onFailure: "error" }) after this middleware.

toolErrorMiddleware(
  config: ToolErrorMiddlewareConfig
): AgentMiddleware<undefined, undefined, unknown, readonly ClientTool | ServerTool[], readonly []>

Parameters

Name	Type	Description
`config`*	`ToolErrorMiddlewareConfig`

Example

import { createAgent, toolErrorMiddleware } from "langchain";

const agent = createAgent({
  model,
  tools: [searchTool],
  middleware: [
    toolErrorMiddleware({
      onError: (error, request) => {
        if (error instanceof TypeError) {
          return `Tool '${request.toolCall.name}' failed with TypeError.`;
        }
      },
    }),
  ],
});

View source on GitHub

toolErrorMiddleware

Parameters

Example

LangChain Assistant

Menu

toolErrorMiddleware

Parameters

Example