FlowiseAI/Flowise · error · Error

Error in Agent node: ${error instanceof Error ? error.messag

Error message

Error in Agent node: ${error instanceof Error ? error.message : String(error)}

What it means

This is the top-level catch in Agent.run. It reports analytics via onLLMError when handlers and llmIds are present, re-throws Abort errors verbatim, and wraps every other error as 'Error in Agent node: <message>'. Seeing this means a downstream operation inside run() failed and was re-packaged; the inner cause is in the interpolated message.

Source

Thrown at packages/components/nodes/agentflow/Agent/Agent.ts:1546

                            (usedTools && usedTools.length > 0)) && {
                            additional_kwargs: {
                                ...(artifacts && artifacts.length > 0 && { artifacts }),
                                ...(fileAnnotations && fileAnnotations.length > 0 && { fileAnnotations }),
                                ...(usedTools && usedTools.length > 0 && { usedTools })
                            }
                        })
                    }
                ]
            }
        } catch (error) {
            if (options.analyticHandlers && llmIds) {
                await options.analyticHandlers.onLLMError(llmIds, error instanceof Error ? error.message : String(error))
            }

            if (error instanceof Error && error.message === 'Aborted') {
                throw error
            }
            throw new Error(`Error in Agent node: ${error instanceof Error ? error.message : String(error)}`)
        }
    }

    /**
     * Extracts built-in used tools from response metadata and processes image generation results
     */
    private async extractBuiltInUsedTools(response: AIMessageChunk, builtInUsedTools: IUsedTool[] = []): Promise<IUsedTool[]> {
        if (!response.response_metadata) {
            return builtInUsedTools
        }

        const { output, tools, groundingMetadata, urlContextMetadata } = response.response_metadata as {
            output?: any[]
            tools?: any[]
            groundingMetadata?: { webSearchQueries?: string[] }
            urlContextMetadata?: { urlMetadata?: any[] }
        }

View on GitHub (pinned to abe4a8601a)

Solutions

Read the interpolated message to find the root cause (e.g. '401 Unauthorized', 'context_length_exceeded').
For auth errors, refresh/rotate the model credential attached to the agent's model node.
For rate-limit/context errors, reduce input size, lower history, or switch to a higher-limit model.
For transient failures, retry the request; for Abort, check the upstream cancellation source.
If analytics are configured, inspect onLLMError events for structured diagnostics.

Defensive patterns

Strategy: try-catch

Try / catch

try {
  await agent.run(nodeData, input, options)
} catch (e) {
  const msg = e instanceof Error ? e.message : String(e)
  if (msg === 'Aborted') { /* user cancellation — expected */ return }
  if (msg.startsWith('Error in Agent node: ')) {
    const inner = msg.slice('Error in Agent node: '.length)
    // classify inner: auth / rate-limit / context-length / tool failure
  }
  throw e
}

Prevention

Keep model credentials valid and rotate before expiry.
Cap chat history length to avoid context overflow.
Wire onLLMError analytics to an alerting channel.
Distinguish Abort (expected) from real failures in your handler.

When it happens

Trigger: Any uncaught throw inside Agent.run: model invocation failure (auth, rate limit, network), tool init failure, message serialization error, structured-output schema mismatch, or one of the explicit guards above. AbortController.abort() is the one case re-thrown without wrapping.

Common situations: Expired/invalid LLM API key; provider rate limit or 5xx; token-context overflow; a tool node throwing during init; malformed BaseMessage objects from custom memory; transient network blip to the model provider.

Related errors

AI-assisted analysis of FlowiseAI/Flowise@abe4a8601a (2026-08-12). Data as JSON: /api/errors/d7c92ee44e7632f1. Report an issue: GitHub.