This workflow follows the Agent → OpenAI Chat recipe pattern — see all workflows that pair these two integrations.
The workflow JSON
Copy or download the full n8n JSON below. Paste it into a new n8n workflow, add your credentials, activate. Full import guide →
{
"name": "Multi-Agent ReAct Orchestrator",
"nodes": [
{
"parameters": {
"httpMethod": "POST",
"path": "multi-agent-chat",
"authentication": "headerAuth",
"responseMode": "responseNode",
"options": {
"responseHeaders": {
"entries": [
{
"name": "Access-Control-Allow-Origin",
"value": "={{ $env.ALLOWED_ORIGINS || '*' }}"
},
{
"name": "Content-Type",
"value": "application/json"
},
{
"name": "X-Request-ID",
"value": "={{ $execution.id }}"
}
]
}
}
},
"id": "webhook-main",
"name": "Main Webhook",
"type": "n8n-nodes-base.webhook",
"typeVersion": 2,
"position": [
200,
400
],
"credentials": {
"httpHeaderAuth": {
"name": "<your credential>"
}
}
},
{
"parameters": {
"conditions": {
"options": {
"caseSensitive": true,
"leftValue": "",
"typeValidation": "strict"
},
"conditions": [
{
"id": "validate-input",
"leftValue": "={{ $json.chatInput && $json.chatInput.length > 0 }}",
"rightValue": true,
"operator": {
"type": "boolean",
"operation": "equals"
}
}
],
"combinator": "and"
},
"options": {}
},
"id": "validate-input",
"name": "Validate Input",
"type": "n8n-nodes-base.if",
"typeVersion": 2,
"position": [
350,
400
]
},
{
"parameters": {
"mode": "runOnceForAllItems",
"jsCode": "// Initialize state with comprehensive error handling\nconst input = $input.first().json;\n\n// Validate required fields\nif (!input.chatInput || typeof input.chatInput !== 'string') {\n throw new Error('Invalid input: chatInput is required and must be a string');\n}\n\nconst state = {\n sessionId: input.sessionId || $execution.id,\n userQuery: input.chatInput.trim(),\n history: Array.isArray(input.history) ? input.history : [],\n phase: 'planning',\n plan: null,\n currentStep: 0,\n stepResults: [],\n iterations: 0,\n maxIterations: parseInt($env.MAX_ITERATIONS) || 5,\n maxStepsPerPlan: parseInt($env.MAX_STEPS_PER_PLAN) || 10,\n events: [],\n errors: [],\n startTime: Date.now(),\n executionId: $execution.id,\n metadata: {\n inputLength: input.chatInput.length,\n historyLength: (input.history || []).length,\n timestamp: new Date().toISOString()\n }\n};\n\n// Emit RUN_STARTED event\nstate.events.push({\n type: 'RUN_STARTED',\n runId: $execution.id,\n threadId: state.sessionId,\n timestamp: new Date().toISOString()\n});\n\nreturn { json: state };"
},
"id": "init-state",
"name": "Initialize State",
"type": "n8n-nodes-base.code",
"typeVersion": 2,
"position": [
500,
400
],
"onError": "continueRegularOutput"
},
{
"parameters": {
"respondWith": "json",
"responseBody": "={{ JSON.stringify({ error: 'Invalid request: chatInput is required', code: 'INVALID_INPUT' }) }}",
"options": {
"responseCode": 400
}
},
"id": "error-invalid-input",
"name": "Error: Invalid Input",
"type": "n8n-nodes-base.respondToWebhook",
"typeVersion": 1.1,
"position": [
500,
600
]
},
{
"parameters": {
"promptType": "define",
"text": "={{ $json.userQuery }}",
"hasOutputParser": true,
"options": {
"systemMessage": "You are the PLANNING AGENT in a multi-agent ReAct orchestration system.\n\nYour ONLY job is to analyze the user's request and create a detailed, executable plan.\n\nIMPORTANT CONSTRAINTS:\n- Maximum {{ $json.maxStepsPerPlan }} steps per plan\n- Each step must be atomic and independently executable\n- Always include error recovery considerations\n\nAvailable tools (use exact names):\n- browser_agent: Web automation with Playwright (navigate, click, fill, scrape, screenshot)\n- code_interpreter_agent: Python execution (calculations, data processing, file generation)\n- web_search_agent: Quick web searches (SerpAPI, Wikipedia)\n- deep_research_agent: Comprehensive multi-source research\n- rag_search_agent: Internal knowledge base semantic search\n- none: Pure reasoning/synthesis without tools\n\nPlan Structure Requirements:\n1. goal: Clear, measurable objective\n2. complexity: low (1-2 steps), medium (3-5 steps), high (6+ steps)\n3. steps: Array of executable steps\n4. successCriteria: How to verify completion\n5. fallbackStrategy: What to do if primary approach fails\n\nOutput ONLY valid JSON matching the required schema."
}
},
"id": "planning-agent",
"name": "Planning Agent",
"type": "@n8n/n8n-nodes-langchain.agent",
"typeVersion": 1.7,
"position": [
700,
400
],
"onError": "continueRegularOutput"
},
{
"parameters": {
"modelId": {
"__rl": true,
"value": "gpt-4o",
"mode": "list"
},
"options": {
"temperature": 0.2,
"maxTokens": 2000,
"timeout": 60000
}
},
"id": "planner-llm",
"name": "Planner LLM",
"type": "@n8n/n8n-nodes-langchain.lmChatOpenAi",
"typeVersion": 1,
"position": [
700,
600
]
},
{
"parameters": {
"schemaType": "fromJson",
"jsonSchemaExample": "{\n \"goal\": \"string\",\n \"complexity\": \"low|medium|high\",\n \"steps\": [\n {\n \"id\": \"step_1\",\n \"description\": \"string\",\n \"tool\": \"browser_agent|code_interpreter_agent|web_search_agent|deep_research_agent|rag_search_agent|none\",\n \"toolInput\": \"string\",\n \"expectedOutput\": \"string\",\n \"dependsOn\": [],\n \"timeout\": 30000,\n \"retryCount\": 1\n }\n ],\n \"successCriteria\": \"string\",\n \"fallbackStrategy\": \"string\"\n}"
},
"id": "plan-output-parser",
"name": "Plan Output Parser",
"type": "@n8n/n8n-nodes-langchain.outputParserStructured",
"typeVersion": 1.2,
"position": [
700,
750
]
},
{
"parameters": {
"sessionIdType": "customKey",
"sessionKey": "=planning_{{ $json.sessionId }}",
"contextWindowLength": 10
},
"id": "planner-memory",
"name": "Planner Memory",
"type": "@n8n/n8n-nodes-langchain.memoryBufferWindow",
"typeVersion": 1.2,
"position": [
850,
600
]
},
{
"parameters": {
"mode": "runOnceForAllItems",
"jsCode": "// Parse and validate planning agent output with comprehensive error handling\nconst state = $('Initialize State').first().json;\nconst agentOutput = $input.first().json;\n\ntry {\n let plan;\n \n // Handle both structured output and raw text\n if (agentOutput.output && typeof agentOutput.output === 'object') {\n plan = agentOutput.output;\n } else if (agentOutput.output && typeof agentOutput.output === 'string') {\n const jsonMatch = agentOutput.output.match(/\\{[\\s\\S]*\\}/);\n if (!jsonMatch) throw new Error('No valid JSON found in planner output');\n plan = JSON.parse(jsonMatch[0]);\n } else {\n throw new Error('Invalid planner output format');\n }\n \n // Validate plan structure\n if (!plan.goal || !Array.isArray(plan.steps) || plan.steps.length === 0) {\n throw new Error('Invalid plan: missing goal or steps');\n }\n \n // Enforce step limits\n if (plan.steps.length > state.maxStepsPerPlan) {\n plan.steps = plan.steps.slice(0, state.maxStepsPerPlan);\n plan.truncated = true;\n }\n \n // Add default values to steps\n plan.steps = plan.steps.map((step, idx) => ({\n id: step.id || `step_${idx + 1}`,\n description: step.description || 'Unnamed step',\n tool: step.tool || 'none',\n toolInput: step.toolInput || '',\n expectedOutput: step.expectedOutput || 'Output not specified',\n dependsOn: step.dependsOn || [],\n timeout: step.timeout || 60000,\n retryCount: step.retryCount || 1,\n status: 'pending'\n }));\n \n state.plan = plan;\n state.phase = 'executing';\n \n // Emit events\n state.events.push(\n {\n type: 'STEP_STARTED',\n stepId: 'planning',\n stepName: 'Creating execution plan',\n stepType: 'planning',\n timestamp: new Date().toISOString()\n },\n {\n type: 'STATE_DELTA',\n delta: [\n { op: 'replace', path: '/phase', value: 'executing' },\n { op: 'add', path: '/plan', value: plan }\n ],\n timestamp: new Date().toISOString()\n },\n {\n type: 'STEP_FINISHED',\n stepId: 'planning',\n status: 'complete',\n result: { stepsCount: plan.steps.length, complexity: plan.complexity },\n timestamp: new Date().toISOString()\n }\n );\n \n} catch (error) {\n state.errors.push({\n phase: 'planning',\n message: error.message,\n timestamp: new Date().toISOString()\n });\n state.phase = 'error';\n state.errorMessage = `Planning failed: ${error.message}`;\n}\n\nreturn { json: state };"
},
"id": "parse-plan",
"name": "Parse & Validate Plan",
"type": "n8n-nodes-base.code",
"typeVersion": 2,
"position": [
900,
400
],
"onError": "continueRegularOutput"
},
{
"parameters": {
"conditions": {
"options": {
"caseSensitive": true,
"leftValue": "",
"typeValidation": "strict"
},
"conditions": [
{
"id": "check-plan-error",
"leftValue": "={{ $json.phase !== 'error' }}",
"rightValue": true,
"operator": {
"type": "boolean",
"operation": "equals"
}
}
],
"combinator": "and"
},
"options": {}
},
"id": "check-plan-success",
"name": "Plan Valid?",
"type": "n8n-nodes-base.if",
"typeVersion": 2,
"position": [
1050,
400
]
},
{
"parameters": {
"conditions": {
"options": {
"caseSensitive": true,
"leftValue": "",
"typeValidation": "strict"
},
"conditions": [
{
"id": "check-more-steps",
"leftValue": "={{ $json.currentStep < $json.plan.steps.length && $json.iterations < $json.maxIterations && $json.phase !== 'error' }}",
"rightValue": true,
"operator": {
"type": "boolean",
"operation": "equals"
}
}
],
"combinator": "and"
},
"options": {}
},
"id": "check-steps",
"name": "More Steps?",
"type": "n8n-nodes-base.if",
"typeVersion": 2,
"position": [
1200,
400
]
},
{
"parameters": {
"promptType": "define",
"text": "Execute this step:\n\nStep: {{ $json.plan.steps[$json.currentStep].description }}\nTool: {{ $json.plan.steps[$json.currentStep].tool }}\nInput: {{ $json.plan.steps[$json.currentStep].toolInput }}\nExpected Output: {{ $json.plan.steps[$json.currentStep].expectedOutput }}\n\nPrevious Results Context:\n{{ JSON.stringify($json.stepResults.slice(-3), null, 2) }}\n\nExecute the step using the appropriate tool and provide results.",
"options": {
"systemMessage": "You are the MANAGER AGENT responsible for executing plan steps.\n\nYou orchestrate specialized tool agents:\n- browser_agent: Web automation (Playwright MCP)\n- code_interpreter_agent: Python code execution\n- web_search_agent: Web searches\n- deep_research_agent: Comprehensive research\n- rag_search_agent: Knowledge base search\n\nExecution Guidelines:\n1. Call the specified tool with the exact input provided\n2. If tool fails, attempt recovery using fallback strategies\n3. Summarize results clearly and concisely\n4. Report any errors or unexpected outcomes\n5. Stay focused on the current step only\n\nALWAYS call the tool before providing a response."
}
},
"id": "manager-agent",
"name": "Manager Agent",
"type": "@n8n/n8n-nodes-langchain.agent",
"typeVersion": 1.7,
"position": [
1400,
300
],
"onError": "continueRegularOutput"
},
{
"parameters": {
"modelId": {
"__rl": true,
"value": "gpt-4o",
"mode": "list"
},
"options": {
"temperature": 0.4,
"maxTokens": 4000,
"timeout": 120000
}
},
"id": "manager-llm",
"name": "Manager LLM",
"type": "@n8n/n8n-nodes-langchain.lmChatOpenAi",
"typeVersion": 1,
"position": [
1400,
500
]
},
{
"parameters": {
"name": "browser_agent",
"description": "Web automation with Playwright MCP. Capabilities: navigate URLs, click elements, fill forms, extract data, take screenshots, handle dynamic content. Input: natural language description of browser task.",
"workflowId": {
"__rl": true,
"value": "={{ $env.BROWSER_AGENT_WORKFLOW_ID }}",
"mode": "id"
},
"specifyInputSchema": true,
"fields": {
"values": [
{
"name": "task",
"type": "string",
"description": "Browser automation task description",
"required": true
}
]
}
},
"id": "tool-browser",
"name": "Browser Agent Tool",
"type": "@n8n/n8n-nodes-langchain.toolWorkflow",
"typeVersion": 1.2,
"position": [
1250,
700
]
},
{
"parameters": {
"name": "code_interpreter_agent",
"description": "Python code execution for calculations, data processing, analysis, file generation. Input: Python code or description of computation needed.",
"workflowId": {
"__rl": true,
"value": "={{ $env.CODE_AGENT_WORKFLOW_ID }}",
"mode": "id"
},
"specifyInputSchema": true,
"fields": {
"values": [
{
"name": "task",
"type": "string",
"description": "Python code or computation description",
"required": true
}
]
}
},
"id": "tool-code",
"name": "Code Interpreter Tool",
"type": "@n8n/n8n-nodes-langchain.toolWorkflow",
"typeVersion": 1.2,
"position": [
1350,
700
]
},
{
"parameters": {
"name": "web_search_agent",
"description": "Quick web searches using SerpAPI and Wikipedia. Input: search query string.",
"workflowId": {
"__rl": true,
"value": "={{ $env.SEARCH_AGENT_WORKFLOW_ID }}",
"mode": "id"
},
"specifyInputSchema": true,
"fields": {
"values": [
{
"name": "query",
"type": "string",
"description": "Search query",
"required": true
}
]
}
},
"id": "tool-search",
"name": "Web Search Tool",
"type": "@n8n/n8n-nodes-langchain.toolWorkflow",
"typeVersion": 1.2,
"position": [
1450,
700
]
},
{
"parameters": {
"name": "deep_research_agent",
"description": "Comprehensive multi-source research on complex topics. Synthesizes from multiple sources. Input: research topic or question.",
"workflowId": {
"__rl": true,
"value": "={{ $env.RESEARCH_AGENT_WORKFLOW_ID }}",
"mode": "id"
},
"specifyInputSchema": true,
"fields": {
"values": [
{
"name": "topic",
"type": "string",
"description": "Research topic or question",
"required": true
}
]
}
},
"id": "tool-research",
"name": "Deep Research Tool",
"type": "@n8n/n8n-nodes-langchain.toolWorkflow",
"typeVersion": 1.2,
"position": [
1550,
700
]
},
{
"parameters": {
"name": "rag_search_agent",
"description": "Semantic search over internal knowledge base using vector similarity. Input: natural language query.",
"workflowId": {
"__rl": true,
"value": "={{ $env.RAG_AGENT_WORKFLOW_ID }}",
"mode": "id"
},
"specifyInputSchema": true,
"fields": {
"values": [
{
"name": "query",
"type": "string",
"description": "Semantic search query",
"required": true
}
]
}
},
"id": "tool-rag",
"name": "RAG Search Tool",
"type": "@n8n/n8n-nodes-langchain.toolWorkflow",
"typeVersion": 1.2,
"position": [
1650,
700
]
},
{
"parameters": {
"mode": "runOnceForAllItems",
"jsCode": "// Process manager agent result with comprehensive error handling\nconst prevState = $('More Steps?').first().json;\nconst managerOutput = $input.first().json;\n\nconst state = JSON.parse(JSON.stringify(prevState));\n\ntry {\n const currentStep = state.plan.steps[state.currentStep];\n const hasError = managerOutput.error || !managerOutput.output;\n \n // Record step result\n const stepResult = {\n stepId: currentStep.id,\n description: currentStep.description,\n tool: currentStep.tool,\n result: hasError ? { error: managerOutput.error || 'No output' } : managerOutput.output,\n success: !hasError,\n toolCalls: managerOutput.intermediateSteps || [],\n startedAt: currentStep.startedAt || new Date().toISOString(),\n completedAt: new Date().toISOString(),\n duration: Date.now() - (currentStep.startedAtMs || Date.now())\n };\n \n state.stepResults.push(stepResult);\n \n // Emit TOOL_CALL events\n if (managerOutput.intermediateSteps) {\n for (const step of managerOutput.intermediateSteps) {\n if (step.action && step.action.tool) {\n state.events.push({\n type: 'TOOL_CALL_START',\n toolCallId: step.action.toolCallId || `tool_${Date.now()}_${Math.random().toString(36).substr(2, 9)}`,\n toolCallName: step.action.tool,\n timestamp: new Date().toISOString()\n });\n \n state.events.push({\n type: 'TOOL_CALL_RESULT',\n toolCallId: step.action.toolCallId || `tool_${Date.now()}_${Math.random().toString(36).substr(2, 9)}`,\n result: typeof step.observation === 'string' ? step.observation.slice(0, 1000) : JSON.stringify(step.observation).slice(0, 1000),\n timestamp: new Date().toISOString()\n });\n }\n }\n }\n \n // Update step status\n state.plan.steps[state.currentStep].status = hasError ? 'failed' : 'complete';\n \n // Emit STEP_FINISHED\n state.events.push({\n type: 'STEP_FINISHED',\n stepId: currentStep.id,\n status: hasError ? 'failed' : 'complete',\n result: { summary: stepResult.result?.toString?.()?.slice(0, 200) || 'Completed' },\n timestamp: new Date().toISOString()\n });\n \n // Move to next step\n state.currentStep++;\n state.iterations++;\n \n // Emit STATE_DELTA\n state.events.push({\n type: 'STATE_DELTA',\n delta: [\n { op: 'replace', path: '/plan/currentStepIndex', value: state.currentStep },\n { op: 'add', path: '/stepResults/-', value: stepResult }\n ],\n timestamp: new Date().toISOString()\n });\n \n // Emit next STEP_STARTED if more steps\n if (state.currentStep < state.plan.steps.length) {\n const nextStep = state.plan.steps[state.currentStep];\n nextStep.startedAt = new Date().toISOString();\n nextStep.startedAtMs = Date.now();\n nextStep.status = 'active';\n \n state.events.push({\n type: 'STEP_STARTED',\n stepId: nextStep.id,\n stepName: nextStep.description,\n stepType: nextStep.tool,\n timestamp: new Date().toISOString()\n });\n }\n \n} catch (error) {\n state.errors.push({\n phase: 'execution',\n step: state.currentStep,\n message: error.message,\n timestamp: new Date().toISOString()\n });\n}\n\nreturn { json: state };"
},
"id": "process-step-result",
"name": "Process Step Result",
"type": "n8n-nodes-base.code",
"typeVersion": 2,
"position": [
1600,
300
],
"onError": "continueRegularOutput"
},
{
"parameters": {
"promptType": "define",
"text": "Review the execution results:\n\nOriginal Goal: {{ $json.plan.goal }}\nSuccess Criteria: {{ $json.plan.successCriteria }}\nTotal Steps: {{ $json.plan.steps.length }}\nCompleted Steps: {{ $json.stepResults.length }}\n\nStep Results:\n{{ JSON.stringify($json.stepResults, null, 2) }}\n\nErrors Encountered: {{ $json.errors.length > 0 ? JSON.stringify($json.errors) : 'None' }}\n\nProvide quality assessment.",
"hasOutputParser": true,
"options": {
"systemMessage": "You are the QUALITY CONTROL AGENT.\n\nYour job is to:\n1. Review all execution results against success criteria\n2. Assess overall quality and completeness\n3. Identify any issues or gaps\n4. Determine if revision is needed\n5. Provide a comprehensive final summary\n\nScoring Guidelines:\n- 9-10: Exceptional, exceeds requirements\n- 7-8: Good, meets all requirements\n- 5-6: Acceptable, minor issues\n- 3-4: Poor, significant issues\n- 1-2: Failed, does not meet requirements\n\nBe thorough but fair. Only require revision for critical issues.\nAlways explain your reasoning."
}
},
"id": "qc-agent",
"name": "Quality Control Agent",
"type": "@n8n/n8n-nodes-langchain.agent",
"typeVersion": 1.7,
"position": [
1400,
500
],
"onError": "continueRegularOutput"
},
{
"parameters": {
"modelId": {
"__rl": true,
"value": "gpt-4o",
"mode": "list"
},
"options": {
"temperature": 0.1,
"maxTokens": 2000,
"timeout": 60000
}
},
"id": "qc-llm",
"name": "QC LLM",
"type": "@n8n/n8n-nodes-langchain.lmChatOpenAi",
"typeVersion": 1,
"position": [
1400,
700
]
},
{
"parameters": {
"schemaType": "fromJson",
"jsonSchemaExample": "{\n \"approved\": true,\n \"qualityScore\": 8,\n \"issues\": [],\n \"suggestions\": [],\n \"requiresRevision\": false,\n \"revisionInstructions\": \"\",\n \"finalSummary\": \"Comprehensive summary of completed work\",\n \"reasoning\": \"Explanation of quality assessment\"\n}"
},
"id": "qc-output-parser",
"name": "QC Output Parser",
"type": "@n8n/n8n-nodes-langchain.outputParserStructured",
"typeVersion": 1.2,
"position": [
1400,
850
]
},
{
"parameters": {
"mode": "runOnceForAllItems",
"jsCode": "// Parse QC result with error handling\nconst state = $('More Steps?').first().json;\nconst qcOutput = $input.first().json;\n\ntry {\n let qcResult;\n \n // Handle structured or raw output\n if (qcOutput.output && typeof qcOutput.output === 'object') {\n qcResult = qcOutput.output;\n } else if (qcOutput.output && typeof qcOutput.output === 'string') {\n const jsonMatch = qcOutput.output.match(/\\{[\\s\\S]*\\}/);\n if (!jsonMatch) throw new Error('No JSON in QC output');\n qcResult = JSON.parse(jsonMatch[0]);\n } else {\n throw new Error('Invalid QC output format');\n }\n \n // Ensure required fields\n qcResult = {\n approved: qcResult.approved ?? true,\n qualityScore: Math.min(10, Math.max(1, qcResult.qualityScore || 5)),\n issues: qcResult.issues || [],\n suggestions: qcResult.suggestions || [],\n requiresRevision: qcResult.requiresRevision ?? false,\n revisionInstructions: qcResult.revisionInstructions || '',\n finalSummary: qcResult.finalSummary || 'Task completed.',\n reasoning: qcResult.reasoning || ''\n };\n \n state.qcResult = qcResult;\n state.phase = qcResult.approved ? 'complete' : 'revision_needed';\n \n // Emit QC events\n state.events.push(\n {\n type: 'STEP_STARTED',\n stepId: 'quality_control',\n stepName: 'Quality Control Review',\n stepType: 'review',\n timestamp: new Date().toISOString()\n },\n {\n type: 'STEP_FINISHED',\n stepId: 'quality_control',\n status: qcResult.approved ? 'complete' : 'failed',\n result: { qualityScore: qcResult.qualityScore, approved: qcResult.approved },\n timestamp: new Date().toISOString()\n },\n {\n type: 'STATE_DELTA',\n delta: [\n { op: 'replace', path: '/phase', value: state.phase },\n { op: 'add', path: '/qualityScore', value: qcResult.qualityScore }\n ],\n timestamp: new Date().toISOString()\n }\n );\n \n} catch (error) {\n // Default to approved on QC parse error to avoid blocking\n state.qcResult = {\n approved: true,\n qualityScore: 5,\n issues: [`QC parsing error: ${error.message}`],\n suggestions: [],\n requiresRevision: false,\n revisionInstructions: '',\n finalSummary: state.stepResults.map(r => r.result).join('\\n\\n') || 'Task completed with QC warnings.',\n reasoning: 'QC parsing failed, defaulting to approved'\n };\n state.phase = 'complete';\n state.errors.push({\n phase: 'qc',\n message: error.message,\n timestamp: new Date().toISOString()\n });\n}\n\nreturn { json: state };"
},
"id": "parse-qc",
"name": "Parse QC Result",
"type": "n8n-nodes-base.code",
"typeVersion": 2,
"position": [
1600,
500
],
"onError": "continueRegularOutput"
},
{
"parameters": {
"conditions": {
"options": {
"caseSensitive": true,
"leftValue": "",
"typeValidation": "strict"
},
"conditions": [
{
"id": "check-revision",
"leftValue": "={{ $json.qcResult && $json.qcResult.requiresRevision && $json.iterations < $json.maxIterations }}",
"rightValue": true,
"operator": {
"type": "boolean",
"operation": "equals"
}
}
],
"combinator": "and"
},
"options": {}
},
"id": "check-revision",
"name": "Needs Revision?",
"type": "n8n-nodes-base.if",
"typeVersion": 2,
"position": [
1800,
500
]
},
{
"parameters": {
"mode": "runOnceForAllItems",
"jsCode": "// Generate final response\nconst state = $json;\n\n// Emit RUN_FINISHED\nstate.events.push({\n type: 'RUN_FINISHED',\n runId: state.executionId,\n threadId: state.sessionId,\n status: state.phase === 'complete' ? 'success' : (state.phase === 'error' ? 'error' : 'partial'),\n timestamp: new Date().toISOString()\n});\n\n// Build final output\nconst finalResponse = {\n success: state.phase === 'complete' || state.phase === 'revision_needed',\n output: state.qcResult?.finalSummary || state.errorMessage || 'Task processing completed.',\n plan: state.plan,\n stepResults: state.stepResults,\n qualityScore: state.qcResult?.qualityScore || null,\n iterations: state.iterations,\n events: state.events,\n errors: state.errors,\n metadata: {\n executionId: state.executionId,\n sessionId: state.sessionId,\n executionTime: Date.now() - state.startTime,\n totalSteps: state.plan?.steps?.length || 0,\n completedSteps: state.stepResults.length,\n phase: state.phase\n }\n};\n\nreturn { json: finalResponse };"
},
"id": "generate-response",
"name": "Generate Response",
"type": "n8n-nodes-base.code",
"typeVersion": 2,
"position": [
2000,
400
]
},
{
"parameters": {
"mode": "runOnceForAllItems",
"jsCode": "// Prepare for revision loop\nconst state = JSON.parse(JSON.stringify($json));\n\nstate.currentStep = 0;\nstate.stepResults = [];\nstate.phase = 'executing';\nstate.iterations++;\n\n// Clear step statuses\nstate.plan.steps.forEach(step => {\n step.status = 'pending';\n delete step.startedAt;\n delete step.startedAtMs;\n});\n\n// Emit revision events\nstate.events.push(\n {\n type: 'STEP_STARTED',\n stepId: 'revision',\n stepName: `Revision ${state.iterations}: ${state.qcResult?.revisionInstructions?.slice(0, 100) || 'Re-executing plan'}`,\n stepType: 'revision',\n timestamp: new Date().toISOString()\n },\n {\n type: 'STATE_DELTA',\n delta: [\n { op: 'replace', path: '/phase', value: 'executing' },\n { op: 'replace', path: '/iterations', value: state.iterations }\n ],\n timestamp: new Date().toISOString()\n }\n);\n\nreturn { json: state };"
},
"id": "prepare-revision",
"name": "Prepare Revision",
"type": "n8n-nodes-base.code",
"typeVersion": 2,
"position": [
2000,
600
]
},
{
"parameters": {
"respondWith": "json",
"responseBody": "={{ JSON.stringify($json) }}",
"options": {
"responseCode": "={{ $json.success ? 200 : ($json.phase === 'error' ? 500 : 207) }}"
}
},
"id": "respond-webhook",
"name": "Respond to Webhook",
"type": "n8n-nodes-base.respondToWebhook",
"typeVersion": 1.1,
"position": [
2200,
400
]
},
{
"parameters": {
"sessionIdType": "customKey",
"sessionKey": "=session_{{ $json.sessionId }}",
"contextWindowLength": 20
},
"id": "short-term-memory",
"name": "Short-Term Memory",
"type": "@n8n/n8n-nodes-langchain.memoryBufferWindow",
"typeVersion": 1.2,
"position": [
550,
600
]
},
{
"parameters": {
"mode": "runOnceForAllItems",
"jsCode": "// Handle planning error - generate error response\nconst state = $json;\n\nstate.events.push({\n type: 'RUN_FINISHED',\n runId: state.executionId,\n threadId: state.sessionId,\n status: 'error',\n timestamp: new Date().toISOString()\n});\n\nconst errorResponse = {\n success: false,\n error: state.errorMessage || 'Planning failed',\n output: `I encountered an error while planning: ${state.errorMessage}. Please try rephrasing your request.`,\n events: state.events,\n errors: state.errors,\n metadata: {\n executionId: state.executionId,\n sessionId: state.sessionId,\n executionTime: Date.now() - state.startTime,\n phase: 'error'\n }\n};\n\nreturn { json: errorResponse };"
},
"id": "handle-plan-error",
"name": "Handle Plan Error",
"type": "n8n-nodes-base.code",
"typeVersion": 2,
"position": [
1200,
550
]
}
],
"connections": {
"Main Webhook": {
"main": [
[
{
"node": "Validate Input",
"type": "main",
"index": 0
}
]
]
},
"Validate Input": {
"main": [
[
{
"node": "Initialize State",
"type": "main",
"index": 0
}
],
[
{
"node": "Error: Invalid Input",
"type": "main",
"index": 0
}
]
]
},
"Initialize State": {
"main": [
[
{
"node": "Planning Agent",
"type": "main",
"index": 0
}
]
]
},
"Planning Agent": {
"main": [
[
{
"node": "Parse & Validate Plan",
"type": "main",
"index": 0
}
]
]
},
"Planner LLM": {
"ai_languageModel": [
[
{
"node": "Planning Agent",
"type": "ai_languageModel",
"index": 0
}
]
]
},
"Plan Output Parser": {
"ai_outputParser": [
[
{
"node": "Planning Agent",
"type": "ai_outputParser",
"index": 0
}
]
]
},
"Planner Memory": {
"ai_memory": [
[
{
"node": "Planning Agent",
"type": "ai_memory",
"index": 0
}
]
]
},
"Short-Term Memory": {
"ai_memory": [
[
{
"node": "Manager Agent",
"type": "ai_memory",
"index": 0
}
]
]
},
"Parse & Validate Plan": {
"main": [
[
{
"node": "Plan Valid?",
"type": "main",
"index": 0
}
]
]
},
"Plan Valid?": {
"main": [
[
{
"node": "More Steps?",
"type": "main",
"index": 0
}
],
[
{
"node": "Handle Plan Error",
"type": "main",
"index": 0
}
]
]
},
"Handle Plan Error": {
"main": [
[
{
"node": "Respond to Webhook",
"type": "main",
"index": 0
}
]
]
},
"More Steps?": {
"main": [
[
{
"node": "Manager Agent",
"type": "main",
"index": 0
}
],
[
{
"node": "Quality Control Agent",
"type": "main",
"index": 0
}
]
]
},
"Manager Agent": {
"main": [
[
{
"node": "Process Step Result",
"type": "main",
"index": 0
}
]
]
},
"Manager LLM": {
"ai_languageModel": [
[
{
"node": "Manager Agent",
"type": "ai_languageModel",
"index": 0
}
]
]
},
"Browser Agent Tool": {
"ai_tool": [
[
{
"node": "Manager Agent",
"type": "ai_tool",
"index": 0
}
]
]
},
"Code Interpreter Tool": {
"ai_tool": [
[
{
"node": "Manager Agent",
"type": "ai_tool",
"index": 0
}
]
]
},
"Web Search Tool": {
"ai_tool": [
[
{
"node": "Manager Agent",
"type": "ai_tool",
"index": 0
}
]
]
},
"Deep Research Tool": {
"ai_tool": [
[
{
"node": "Manager Agent",
"type": "ai_tool",
"index": 0
}
]
]
},
"RAG Search Tool": {
"ai_tool": [
[
{
"node": "Manager Agent",
"type": "ai_tool",
"index": 0
}
]
]
},
"Process Step Result": {
"main": [
[
{
"node": "More Steps?",
"type": "main",
"index": 0
}
]
]
},
"Quality Control Agent": {
"main": [
[
{
"node": "Parse QC Result",
"type": "main",
"index": 0
}
]
]
},
"QC LLM": {
"ai_languageModel": [
[
{
"node": "Quality Control Agent",
"type": "ai_languageModel",
"index": 0
}
]
]
},
"QC Output Parser": {
"ai_outputParser": [
[
{
"node": "Quality Control Agent",
"type": "ai_outputParser",
"index": 0
}
]
]
},
"Parse QC Result": {
"main": [
[
{
"node": "Needs Revision?",
"type": "main",
"index": 0
}
]
]
},
"Needs Revision?": {
"main": [
[
{
"node": "Generate Response",
"type": "main",
"index": 0
}
],
[
{
"node": "Prepare Revision",
"type": "main",
"index": 0
}
]
]
},
"Generate Response": {
"main": [
[
{
"node": "Respond to Webhook",
"type": "main",
"index": 0
}
]
]
},
"Prepare Revision": {
"main": [
[
{
"node": "More Steps?",
"type": "main",
"index": 0
}
]
]
}
},
"settings": {
"executionOrder": "v1",
"saveManualExecutions": true,
"callerPolicy": "workflowsFromSameOwner",
"errorWorkflow": "={{ $env.ERROR_WORKFLOW_ID }}"
},
"staticData": null,
"tags": [
"multi-agent",
"react",
"orchestration",
"production"
],
"triggerCount": 0
}
Credentials you'll need
Each integration node will prompt for credentials when you import. We strip credential IDs before publishing — you'll add your own.
httpHeaderAuth
Pro
For the full experience including quality scoring and batch install features for each workflow upgrade to Pro
About this workflow
Multi-Agent ReAct Orchestrator. Uses agent, lmChatOpenAi, outputParserStructured, memoryBufferWindow. Webhook trigger; 29 nodes.
Source: https://github.com/NarsistPanda/Claude1-n8n-avatar/blob/232d49109540ee19b25950df82fafc5f2a646f63/n8n-workflows/multi-agent-react.json — original creator credit. Request a take-down →
Related workflows
Workflows that share integrations, category, or trigger type with this one. All free to copy and import.
🧠 Gwen – The AI Voice Marketing Agent Gwen is your intelligent voice-powered marketing assistant built in n8n. She combines the power of OpenAI, ElevenLabs, and automation workflows to handle content
Tool Workflow, Memory Buffer Window, Agent +10
This workflow is an AI-powered Dental Appointment Assistant that automates appointment booking, rescheduling, and cancellations through Telegram or a Webhook. It uses intelligent agents to understand
Memory Buffer Window, Output Parser Structured, Mcp Client Tool +12
Are you drowning in daily operational chaos, desperately trying to juggle sales, projects, content, and client communication? Imagine an AI brain that handles it all, freeing you to lead your business
Telegram Trigger, Telegram, OpenAI +13
Enhance your support, onboarding, and internal knowledge workflows with an intelligent RAG-powered chatbot that responds using live data stored in Google Sheets. 🤖📚 Built for teams that rely on struct
Chat Trigger, Output Parser Structured, Memory Buffer Window +6
Template Carnaval - time instagram. Uses toolWorkflow, lmChatOpenAi, memoryBufferWindow, agent. Event-driven trigger; 56 nodes.
Tool Workflow, OpenAI Chat, Memory Buffer Window +10