FlowiseAI/Flowise · error · Error

Error in LLM node: ${error instanceof Error ? error.message

Error message

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

What it means

The catch-all at the end of LLM_Agentflow.run(). Any exception not matching the special-case 'Aborted' (user-initiated abort, which is re-thrown verbatim) is wrapped as `Error in LLM node: <message>`. The analytics onLLMError handler is invoked first with the raw message. This wrapper preserves the original message text but replaces the error type and stack origin.

Source

Thrown at packages/components/nodes/agentflow/LLM/LLM.ts:694

                        name: nodeData?.label ? nodeData?.label.toLowerCase().replace(/\s/g, '_').trim() : nodeData?.id,
                        ...(((artifacts && artifacts.length > 0) || (fileAnnotations && fileAnnotations.length > 0)) && {
                            additional_kwargs: {
                                ...(artifacts && artifacts.length > 0 && { artifacts }),
                                ...(fileAnnotations && fileAnnotations.length > 0 && { fileAnnotations })
                            }
                        })
                    }
                ]
            }
        } 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 LLM node: ${error instanceof Error ? error.message : String(error)}`)
        }
    }

    /**
     * Handles memory management based on the specified memory type
     */
    private async handleMemory({
        messages,
        memoryType,
        pastChatHistory,
        runtimeChatHistory,
        llmNodeInstance,
        nodeData,
        userMessage,
        input,
        abortController,
        options,
        modelConfig

View on GitHub (pinned to abe4a8601a)

Solutions

Read the analytics dashboard (onLLMError fires) or server logs for the inner message after the colon.
If the message mentions auth/key — fix the model credential.
If it mentions tokens/context — trim chat history or reduce input size.
If it mentions model not found — correct the model name in the model component.
For transient provider errors, add retry at the provider/flow level.
Patch the wrapper to use `{ cause: error }` so the original stack survives.

Example fix

// before
            throw new Error(`Error in LLM node: ${error instanceof Error ? error.message : String(error)}`)
// after
            throw new Error(`Error in LLM node: ${error instanceof Error ? error.message : String(error)}`, { cause: error })

Defensive patterns

Strategy: try-catch

Try / catch

try {
  await llmNode.run(nodeData, input, options)
} catch (e) {
  const msg = (e as Error).message
  if (msg === 'Aborted') return // user-initiated, swallow
  if (msg.startsWith('Error in LLM node: ')) {
    const inner = msg.slice('Error in LLM node: '.length)
    // branch on inner: auth, context length, model not found
  }
  throw e
}

Prevention

Pre-validate provider credentials and model name before the LLM node.
Bound chat history length to avoid context-overflow errors.
Subscribe to onLLMError analytics to capture the raw inner message.
Pass a stable AbortController so aborts are intentional.

When it happens

Trigger: The chat model throws (auth failure, rate limit, context-length exceeded, invalid model name), memory/vector-store retrieval fails, a structured-output schema is invalid, or the provider SDK raises any non-Abort error.

Common situations: Wrong/missing API key for the provider; model name typo or deprecated model; prompt+history exceeds token limit; tool/structured-output JSON schema mismatch; transient provider 5xx.

Related errors

AI-assisted analysis of FlowiseAI/Flowise@abe4a8601a (2026-08-12). Data as JSON: /api/errors/fd7a6ddb6faf4172. Report an issue: GitHub.