FlowiseAI/Flowise · error · Error

Could not find JSON block in the output.

Error message

Could not find JSON block in the output.

What it means

ConditionAgent.parseJsonMarkdown throws 'Could not find JSON block in the output.' when it cannot find any recognized start/end delimiters (```json, ```, backticks, or { } ) with a valid ordering in the LLM response. This means the model returned plain prose or a format the extractor does not recognize — no JSON region to even attempt parsing.

Source

Thrown at packages/components/nodes/agentflow/ConditionAgent/ConditionAgent.ts:246

                if (endIndex !== -1) {
                    if (jsonString[endIndex] === '}') {
                        endIndex += 1
                    }
                    break
                }
            }
        }

        if (startIndex !== -1 && endIndex !== -1 && startIndex < endIndex) {
            const extractedContent = jsonString.slice(startIndex, endIndex).trim()
            try {
                return JSON.parse(extractedContent)
            } catch (error) {
                throw new Error(`Invalid JSON object. Error: ${error}`)
            }
        }

        throw new Error('Could not find JSON block in the output.')
    }

    async run(nodeData: INodeData, question: string, options: ICommonObject): Promise<any> {
        let llmIds: ICommonObject | undefined
        let analyticHandlers = options.analyticHandlers as AnalyticHandler

        try {
            const abortController = options.abortController as AbortController

            // Extract input parameters
            const model = nodeData.inputs?.conditionAgentModel as string
            const modelConfig = nodeData.inputs?.conditionAgentModelConfig as ICommonObject
            if (!model) {
                throw new Error('Model is required')
            }
            const modelName = modelConfig?.model ?? modelConfig?.modelName

            const conditionAgentInput = nodeData.inputs?.conditionAgentInput as string

View on GitHub (pinned to abe4a8601a)

Solutions

1. Reinforce JSON output in the system prompt and keep the few-shot example intact.
2. Switch to a model that reliably follows formatting instructions.
3. Reduce injected chat history / input size so the instruction is not crowded out.
4. If overriding the system prompt, ensure it still mandates a fenced JSON object with an 'output' key.

function hasJsonRegion(s) {
  return /\{[\s\S]*\}|```json/i.test(s ?? '')
}
if (!hasJsonRegion(responseContent)) {
  // retry the LLM call with a stronger JSON instruction instead of throwing
}

function containsJsonBlock(s) {
  return /```json|```|`|\{/.test(s ?? '')
}

try { parsed = parseJsonMarkdown(responseContent) }
catch (e) {
  if (/Could not find JSON block/.test(e.message)) {
    // re-prompt with explicit 'respond ONLY with fenced JSON'
  } else throw e
}

When it happens

Trigger: LLM returns a pure natural-language answer with no JSON; model returns only the scenario name as text without braces/fences; empty response; response using a delimiter the parser doesn't look for (e.g. indented YAML, XML tags).

Common situations: Model ignores the structured-output instruction entirely; system prompt overridden in a way that drops the JSON requirement; very small model that can't follow formatting; context window dominated by history leaving no room for the instruction.

Related errors

• Invalid JSON object. Error: ${error}
• LLM response is missing the "output" key or it is not a stri
• Failed to parse a valid scenario from the LLM's response. Pl
• Model is required
• Scenarios are required

AI-assisted analysis of FlowiseAI/Flowise@abe4a8601a (2026-08-12). Data as JSON: /api/errors/ae8a8d549f473cc6. Report an issue: GitHub.