feat(core): Chatting with n8n agent workflows (no-changelog) (#21088)

packages/@n8n/api-types/src/chat-hub.ts

@@ -4,14 +4,16 @@ import { Z } from 'zod-class';
 /**
  * Supported AI model providers
  */
-export const chatHubProviderSchema = z.enum(['openai', 'anthropic', 'google']);
+export const chatHubLLMProviderSchema = z.enum(['openai', 'anthropic', 'google']);
+export type ChatHubLLMProvider = z.infer<typeof chatHubLLMProviderSchema>;
 
+export const chatHubProviderSchema = z.enum([...chatHubLLMProviderSchema.options, 'n8n'] as const);
 export type ChatHubProvider = z.infer<typeof chatHubProviderSchema>;
 
 /**
  * Map of providers to their credential types
  */
-export const PROVIDER_CREDENTIAL_TYPE_MAP: Record<ChatHubProvider, string> = {
+export const PROVIDER_CREDENTIAL_TYPE_MAP: Record<ChatHubLLMProvider, string> = {
 	openai: 'openAiApi',
 	anthropic: 'anthropicApi',
 	google: 'googlePalmApi',
@@ -20,12 +22,37 @@ export const PROVIDER_CREDENTIAL_TYPE_MAP: Record<ChatHubProvider, string> = {
 /**
  * Chat Hub conversation model configuration
  */
-export const chatHubConversationModelSchema = z.object({
-	provider: chatHubProviderSchema,
+const openAIModelSchema = z.object({
+	provider: z.literal('openai'),
+	name: z.string(),
 	model: z.string(),
-	workflowId: z.string().nullable().default(null),
 });
 
+const anthropicModelSchema = z.object({
+	provider: z.literal('anthropic'),
+	name: z.string(),
+	model: z.string(),
+});
+
+const googleModelSchema = z.object({
+	provider: z.literal('google'),
+	name: z.string(),
+	model: z.string(),
+});
+
+const n8nModelSchema = z.object({
+	provider: z.literal('n8n'),
+	name: z.string(),
+	workflowId: z.string(),
+});
+
+export const chatHubConversationModelSchema = z.discriminatedUnion('provider', [
+	openAIModelSchema,
+	anthropicModelSchema,
+	googleModelSchema,
+	n8nModelSchema,
+]);
+
 export type ChatHubConversationModel = z.infer<typeof chatHubConversationModelSchema>;
 
 /**
@@ -43,7 +70,10 @@ export type ChatModelsRequest = z.infer<typeof chatModelsRequestSchema>;
  */
 export type ChatModelsResponse = Record<
 	ChatHubProvider,
-	{ models: Array<{ name: string }>; error?: string }
+	{
+		models: ChatHubConversationModel[];
+		error?: string;
+	}
 >;
 
 export class ChatHubSendMessageRequest extends Z.class({

packages/@n8n/api-types/src/index.ts

@@ -11,6 +11,7 @@ export {
 	type ChatHubConversationModel,
 	chatHubProviderSchema,
 	type ChatHubProvider,
+	type ChatHubLLMProvider,
 	type ChatHubMessageType,
 	type ChatHubMessageStatus,
 	PROVIDER_CREDENTIAL_TYPE_MAP,

packages/cli/src/modules/chat-hub/chat-hub.controller.ts

@@ -18,14 +18,7 @@ import { strict as assert } from 'node:assert';
 import { ChatHubService } from './chat-hub.service';
 import { ChatModelsRequestDto } from './dto/chat-models-request.dto';
 
-/* eslint-disable @typescript-eslint/naming-convention */
-const JSONL_STREAM_HEADERS = {
-	'Content-Type': 'application/json-lines; charset=utf-8',
-	'Transfer-Encoding': 'chunked',
-	'Cache-Control': 'no-cache',
-	Connection: 'keep-alive',
-};
-/* eslint-enable @typescript-eslint/naming-convention */
+import { ResponseError } from '@/errors/response-errors/abstract/response.error';
 
 @RestController('/chat')
 export class ChatHubController {
@@ -70,9 +63,6 @@ export class ChatHubController {
 		res: Response,
 		@Body payload: ChatHubSendMessageRequest,
 	) {
-		res.writeHead(200, JSONL_STREAM_HEADERS);
-		res.flushHeaders();
-
 		this.logger.debug(`Chat send request received: ${JSON.stringify(payload)}`);
 
 		try {
@@ -80,21 +70,25 @@ export class ChatHubController {
 				...payload,
 				userId: req.user.id,
 			});
-		} catch (executionError: unknown) {
-			assert(executionError instanceof Error);
+		} catch (error: unknown) {
+			assert(error instanceof Error);
 
-			this.logger.error(`Error in chat send endpoint: ${executionError}`);
+			this.logger.error(`Error in chat send endpoint: ${error}`);
 
 			if (!res.headersSent) {
+				if (error instanceof ResponseError) {
+					throw error;
+				}
+
 				res.status(500).json({
 					code: 500,
-					message: executionError.message,
+					message: error.message,
 				});
 			} else if (!res.writableEnded) {
 				res.write(
 					JSON.stringify({
 						type: 'error',
-						content: executionError.message,
+						content: error.message,
 					}) + '\n',
 				);
 				res.flush();
@@ -113,9 +107,6 @@ export class ChatHubController {
 		@Param('messageId') editId: ChatMessageId,
 		@Body payload: ChatHubEditMessageRequest,
 	) {
-		res.writeHead(200, JSONL_STREAM_HEADERS);
-		res.flushHeaders();
-
 		this.logger.debug(`Chat edit request received: ${JSON.stringify(payload)}`);
 
 		try {
@@ -125,22 +116,25 @@ export class ChatHubController {
 				editId,
 				userId: req.user.id,
 			});
-		} catch (executionError: unknown) {
-			assert(executionError instanceof Error);
+		} catch (error: unknown) {
+			assert(error instanceof Error);
 
-			this.logger.error(`Error in chat edit endpoint: ${executionError}`);
+			this.logger.error(`Error in chat edit endpoint: ${error}`);
 
 			if (!res.headersSent) {
+				if (error instanceof ResponseError) {
+					throw error;
+				}
+
 				res.status(500).json({
 					code: 500,
-					message: executionError.message,
+					message: error.message,
 				});
 			} else if (!res.writableEnded) {
 				res.write(
 					JSON.stringify({
 						type: 'error',
-						content: executionError.message,
-						id: payload.replyId,
+						content: error.message,
 					}) + '\n',
 				);
 				res.flush();
@@ -159,9 +153,6 @@ export class ChatHubController {
 		@Param('messageId') retryId: ChatMessageId,
 		@Body payload: ChatHubRegenerateMessageRequest,
 	) {
-		res.writeHead(200, JSONL_STREAM_HEADERS);
-		res.flushHeaders();
-
 		this.logger.debug(`Chat retry request received: ${JSON.stringify(payload)}`);
 
 		try {
@@ -171,22 +162,25 @@ export class ChatHubController {
 				retryId,
 				userId: req.user.id,
 			});
-		} catch (executionError: unknown) {
-			assert(executionError instanceof Error);
+		} catch (error: unknown) {
+			assert(error instanceof Error);
 
-			this.logger.error(`Error in chat retry endpoint: ${executionError}`);
+			this.logger.error(`Error in chat retry endpoint: ${error}`);
 
 			if (!res.headersSent) {
+				if (error instanceof ResponseError) {
+					throw error;
+				}
+
 				res.status(500).json({
 					code: 500,
-					message: executionError.message,
+					message: error.message,
 				});
 			} else if (!res.writableEnded) {
 				res.write(
 					JSON.stringify({
 						type: 'error',
-						content: executionError.message,
-						id: payload.replyId,
+						content: error.message,
 					}) + '\n',
 				);
 				res.flush();

packages/cli/src/modules/chat-hub/chat-hub.service.ts

@@ -1,10 +1,10 @@
 import {
 	PROVIDER_CREDENTIAL_TYPE_MAP,
 	type ChatHubProvider,
+	type ChatHubLLMProvider,
 	type ChatModelsResponse,
 	type ChatHubConversationsResponse,
 	type ChatHubConversationResponse,
-	chatHubProviderSchema,
 	ChatHubMessageDto,
 	type ChatMessageId,
 	type ChatSessionId,
@@ -41,6 +41,7 @@ import {
 	type IRun,
 	jsonParse,
 	StructuredChunk,
+	RESPOND_TO_CHAT_NODE_TYPE,
 } from 'n8n-workflow';
 import { v4 as uuidv4 } from 'uuid';
 
@@ -53,6 +54,8 @@ import { ExecutionService } from '@/executions/execution.service';
 import { DynamicNodeParametersService } from '@/services/dynamic-node-parameters.service';
 import { getBase } from '@/workflow-execute-additional-data';
 import { WorkflowExecutionService } from '@/workflows/workflow-execution.service';
+import { WorkflowFinderService } from '@/workflows/workflow-finder.service';
+import { WorkflowService } from '@/workflows/workflow.service';
 
 import type { ChatHubMessage } from './chat-hub-message.entity';
 import { CONVERSATION_TITLE_GENERATION_PROMPT } from './chat-hub.constants';
@@ -68,7 +71,7 @@ import { ChatHubSessionRepository } from './chat-session.repository';
 import { getMaxContextWindowTokens } from './context-limits';
 import { captureResponseWrites } from './stream-capturer';
 
-const providerNodeTypeMapping: Record<ChatHubProvider, INodeTypeNameVersion> = {
+const providerNodeTypeMapping: Record<ChatHubLLMProvider, INodeTypeNameVersion> = {
 	openai: {
 		name: '@n8n/n8n-nodes-langchain.lmChatOpenAi',
 		version: 1.2,
@@ -93,6 +96,15 @@ const NODE_NAMES = {
 	CLEAR_CHAT_MEMORY: 'Clear Chat Memory',
 } as const;
 
+/* eslint-disable @typescript-eslint/naming-convention */
+const JSONL_STREAM_HEADERS = {
+	'Content-Type': 'application/json-lines; charset=utf-8',
+	'Transfer-Encoding': 'chunked',
+	'Cache-Control': 'no-cache',
+	Connection: 'keep-alive',
+};
+/* eslint-enable @typescript-eslint/naming-convention */
+
 @Service()
 export class ChatHubService {
 	constructor(
@@ -101,6 +113,8 @@ export class ChatHubService {
 		private readonly nodeParametersService: DynamicNodeParametersService,
 		private readonly executionRepository: ExecutionRepository,
 		private readonly workflowExecutionService: WorkflowExecutionService,
+		private readonly workflowService: WorkflowService,
+		private readonly workflowFinderService: WorkflowFinderService,
 		private readonly workflowRepository: WorkflowRepository,
 		private readonly sharedWorkflowRepository: SharedWorkflowRepository,
 		private readonly activeExecutions: ActiveExecutions,
@@ -111,48 +125,52 @@ export class ChatHubService {
 
 	async getModels(
 		user: User,
-		credentialIds: Record<ChatHubProvider, string | null>,
+		credentialIds: Record<ChatHubLLMProvider, string | null>,
 	): Promise<ChatModelsResponse> {
 		const additionalData = await getBase({ userId: user.id });
+		// The n8n provider is disabled for now.
+		// const providers = chatHubProviderSchema.options;
+		const providers = ['openai', 'anthropic', 'google'] as ChatHubLLMProvider[];
 
 		const allCredentials = await this.credentialsFinderService.findCredentialsForUser(user, [
 			'credential:read',
 		]);
 
 		const responses = await Promise.all(
-			chatHubProviderSchema.options.map<
-				Promise<[ChatHubProvider, ChatModelsResponse[ChatHubProvider]]>
-			>(async (provider) => {
-				const credentialId = credentialIds[provider];
+			providers.map<Promise<[ChatHubProvider, ChatModelsResponse[ChatHubProvider]]>>(
+				async (provider: ChatHubProvider) => {
+					const credentials: INodeCredentials = {};
 
-				if (!credentialId) {
-					return [provider, { models: [] }];
-				}
+					if (provider !== 'n8n') {
+						const credentialId = credentialIds[provider];
+						if (!credentialId) {
+							return [provider, { models: [] }];
+						}
 
-				// Ensure the user has the permission to read the credential
-				if (!allCredentials.some((credential) => credential.id === credentialId)) {
-					return [
-						provider,
-						{ models: [], error: 'Could not retrieve models. Verify credentials.' },
-					];
-				}
+						// Ensure the user has the permission to read the credential
+						if (!allCredentials.some((credential) => credential.id === credentialId)) {
+							return [
+								provider,
+								{ models: [], error: 'Could not retrieve models. Verify credentials.' },
+							];
+						}
 
-				try {
-					const credentials = {
-						[PROVIDER_CREDENTIAL_TYPE_MAP[provider]]: { name: '', id: credentialId },
-					};
+						credentials[PROVIDER_CREDENTIAL_TYPE_MAP[provider]] = { name: '', id: credentialId };
+					}
 
-					return [
-						provider,
-						await this.fetchModelsForProvider(provider, credentials, additionalData),
-					];
-				} catch {
-					return [
-						provider,
-						{ models: [], error: 'Could not retrieve models. Verify credentials.' },
-					];
-				}
-			}),
+					try {
+						return [
+							provider,
+							await this.fetchModelsForProvider(user, provider, credentials, additionalData),
+						];
+					} catch {
+						return [
+							provider,
+							{ models: [], error: 'Could not retrieve models. Verify credentials.' },
+						];
+					}
+				},
+			),
 		);
 
 		return responses.reduce<ChatModelsResponse>(
@@ -164,11 +182,13 @@ export class ChatHubService {
 				openai: { models: [] },
 				anthropic: { models: [] },
 				google: { models: [] },
+				n8n: { models: [] },
 			},
 		);
 	}
 
 	private async fetchModelsForProvider(
+		user: User,
 		provider: ChatHubProvider,
 		credentials: INodeCredentials,
 		additionalData: IWorkflowExecuteAdditionalData,
@@ -180,6 +200,8 @@ export class ChatHubService {
 				return await this.fetchAnthropicModels(credentials, additionalData);
 			case 'google':
 				return await this.fetchGoogleModels(credentials, additionalData);
+			case 'n8n':
+				return await this.fetchCustomAgentWorkflows(user);
 		}
 	}
 
@@ -197,7 +219,11 @@ export class ChatHubService {
 		);
 
 		return {
-			models: resourceLocatorResults.results.map((result) => ({ name: String(result.value) })),
+			models: resourceLocatorResults.results.map((result) => ({
+				provider: 'openai',
+				name: String(result.value),
+				model: String(result.value),
+			})),
 		};
 	}
 
@@ -215,7 +241,11 @@ export class ChatHubService {
 		);
 
 		return {
-			models: resourceLocatorResults.results.map((result) => ({ name: String(result.value) })),
+			models: resourceLocatorResults.results.map((result) => ({
+				provider: 'anthropic',
+				name: String(result.value),
+				model: String(result.value),
+			})),
 		};
 	}
 
@@ -271,7 +301,28 @@ export class ChatHubService {
 		);
 
 		return {
-			models: results.map((result) => ({ name: String(result.value) })),
+			models: results.map((result) => ({
+				provider: 'google',
+				name: String(result.value),
+				model: String(result.value),
+			})),
+		};
+	}
+
+	private async fetchCustomAgentWorkflows(
+		user: User,
+	): Promise<ChatModelsResponse[ChatHubProvider]> {
+		const nodeTypes = [CHAT_TRIGGER_NODE_TYPE];
+		const workflows = await this.workflowService.getWorkflowsWithNodesIncluded(user, nodeTypes);
+
+		return {
+			models: workflows
+				.filter((workflow) => workflow.active)
+				.map((workflow) => ({
+					provider: 'n8n',
+					name: workflow.name ?? 'Unnamed workflow',
+					workflowId: workflow.id,
+				})),
 		};
 	}
 
@@ -389,39 +440,26 @@ export class ChatHubService {
 		provider: ChatHubProvider,
 		credentials: INodeCredentials,
 	): string | null {
+		if (provider === 'n8n') {
+			return null;
+		}
+
 		return credentials[PROVIDER_CREDENTIAL_TYPE_MAP[provider]]?.id ?? null;
 	}
 
 	async sendHumanMessage(res: Response, user: User, payload: HumanMessagePayload) {
 		const { sessionId, messageId, replyId, message } = payload;
+		const provider = payload.model.provider;
 
 		const selectedModel: ModelWithCredentials = {
 			...payload.model,
-			credentialId: this.pickCredentialId(payload.model.provider, payload.credentials),
+			credentialId:
+				provider !== 'n8n' ? this.pickCredentialId(provider, payload.credentials) : null,
 		};
 
 		const workflow = await this.messageRepository.manager.transaction(async (trx) => {
-			const credential = await this.ensureCredentials(
-				user,
-				payload.model,
-				payload.credentials,
-				trx,
-			);
 			const session = await this.getChatSession(user, sessionId, selectedModel, true, trx);
-
-			// Ensure that the previous message exists in the session
-			if (payload.previousMessageId) {
-				const previousMessage = await this.messageRepository.getOneById(
-					payload.previousMessageId,
-					sessionId,
-					[],
-					trx,
-				);
-				if (!previousMessage) {
-					throw new BadRequestError('The previous message does not exist in the session');
-				}
-			}
-
+			await this.ensurePreviousMessage(payload.previousMessageId, sessionId, trx);
 			const messages = Object.fromEntries((session.messages ?? []).map((m) => [m.id, m]));
 			const history = this.buildMessageHistory(messages, payload.previousMessageId);
 
@@ -434,15 +472,15 @@ export class ChatHubService {
 				trx,
 			);
 
-			return await this.createChatWorkflow(
-				session.id,
-				credential.projectId,
-				history,
+			if (provider !== 'n8n') {
+				return await this.prepareBaseChatWorkflow(user, payload, sessionId, history, message, trx);
+			}
+
+			return await this.prepareCustomAgentWorkflow(
+				user,
+				sessionId,
+				payload.model.workflowId,
 				message,
-				payload.credentials,
-				payload.model,
-				payload.previousMessageId === null, // generate title on receiving the first human message only
-				trx,
 			);
 		});
 
@@ -457,7 +495,125 @@ export class ChatHubService {
 				selectedModel,
 			);
 		} finally {
-			await this.deleteChatWorkflow(workflow.workflowData.id);
+			if (provider !== 'n8n') {
+				await this.deleteChatWorkflow(workflow.workflowData.id);
+			}
+		}
+	}
+
+	private async prepareBaseChatWorkflow(
+		user: User,
+		payload: HumanMessagePayload,
+		sessionId: ChatSessionId,
+		history: ChatHubMessage[],
+		message: string,
+		trx: EntityManager,
+	) {
+		const credential = await this.ensureCredentials(user, payload.model, payload.credentials, trx);
+
+		return await this.createChatWorkflow(
+			sessionId,
+			credential.projectId,
+			history,
+			message,
+			payload.credentials,
+			payload.model,
+			payload.previousMessageId === null, // generate title on receiving the first human message only
+			trx,
+		);
+	}
+
+	private async prepareCustomAgentWorkflow(
+		user: User,
+		sessionId: ChatSessionId,
+		workflowId: string,
+		message: string,
+	) {
+		const workflowEntity = await this.workflowFinderService.findWorkflowForUser(
+			workflowId,
+			user,
+			['workflow:read'],
+			{ includeTags: false, includeParentFolder: false },
+		);
+
+		if (!workflowEntity) {
+			throw new BadRequestError('Workflow not found');
+		}
+
+		const chatTriggers = workflowEntity.nodes.filter(
+			(node) => node.type === CHAT_TRIGGER_NODE_TYPE,
+		);
+
+		if (chatTriggers.length !== 1) {
+			throw new BadRequestError('Workflow must have exactly one chat trigger');
+		}
+
+		const chatResponseNodes = workflowEntity.nodes.filter(
+			(node) => node.type === RESPOND_TO_CHAT_NODE_TYPE,
+		);
+
+		if (chatResponseNodes.length > 0) {
+			throw new BadRequestError(
+				'Respond to Chat nodes are not supported in custom agent workflows',
+			);
+		}
+
+		const agents = workflowEntity.nodes.filter((node) => node.type === AGENT_LANGCHAIN_NODE_TYPE);
+		if (agents.length !== 1) {
+			throw new BadRequestError('Workflow must have exactly one AI Agent node');
+		}
+
+		return {
+			workflowData: {
+				...workflowEntity,
+				// Since this mechanism executes workflows as manual one-off executions
+				// we need to clear any pinData the WF might have.
+				// TODO: Implement a separate execution mode for chats to avoid such workarounds.
+				pinData: {},
+			},
+			triggerToStartFrom: {
+				name: chatTriggers[0].name,
+				data: {
+					startTime: Date.now(),
+					executionTime: 0,
+					executionIndex: 0,
+					executionStatus: 'success',
+					data: {
+						main: [
+							[
+								{
+									json: {
+										sessionId,
+										action: 'sendMessage',
+										chatInput: message,
+									},
+								},
+							],
+						],
+					},
+					source: [null],
+				} satisfies ITaskData,
+			},
+		};
+	}
+
+	private async ensurePreviousMessage(
+		previousMessageId: ChatMessageId | null,
+		sessionId: string,
+		trx?: EntityManager,
+	) {
+		if (!previousMessageId) {
+			return;
+		}
+
+		const previousMessage = await this.messageRepository.getOneById(
+			previousMessageId,
+			sessionId,
+			[],
+			trx,
+		);
+		if (!previousMessage) {
+			throw new BadRequestError('The previous message does not exist in the session');
 		}
 	}
 
@@ -465,7 +621,10 @@ export class ChatHubService {
 		const { sessionId, editId, messageId, replyId } = payload;
 		const selectedModel: ModelWithCredentials = {
 			...payload.model,
-			credentialId: this.pickCredentialId(payload.model.provider, payload.credentials),
+			credentialId:
+				payload.model.provider !== 'n8n'
+					? this.pickCredentialId(payload.model.provider, payload.credentials)
+					: null,
 		};
 
 		const workflow = await this.messageRepository.manager.transaction(async (trx) => {
@@ -542,10 +701,12 @@ export class ChatHubService {
 
 	async regenerateAIMessage(res: Response, user: User, payload: RegenerateMessagePayload) {
 		const { sessionId, retryId, replyId } = payload;
-
 		const selectedModel: ModelWithCredentials = {
 			...payload.model,
-			credentialId: this.pickCredentialId(payload.model.provider, payload.credentials),
+			credentialId:
+				payload.model.provider !== 'n8n'
+					? this.pickCredentialId(payload.model.provider, payload.credentials)
+					: null,
 		};
 
 		const { workflow, retryOfMessageId, previousMessageId } =
@@ -659,15 +820,17 @@ export class ChatHubService {
 
 		// Capture the streaming response as it's being generated to save
 		// partial messages in the database when generation gets cancelled.
-		let message = '';
+		let partialMessage = '';
 		const onChunk = (chunk: string) => {
 			const data = jsonParse<StructuredChunk>(chunk);
 			if (data && data.type === 'item' && typeof data.content === 'string') {
-				message += data.content;
+				partialMessage += data.content;
 			}
 		};
 
 		const stream = captureResponseWrites(res, onChunk);
+		stream.writeHead(200, JSONL_STREAM_HEADERS);
+		stream.flushHeaders();
 
 		const { executionId } = await this.workflowExecutionService.executeManually(
 			{
@@ -688,7 +851,7 @@ export class ChatHubService {
 			sessionId,
 			executionId,
 			previousMessageId,
-			message,
+			message: partialMessage,
 			selectedModel,
 			retryOfMessageId,
 			status: 'running',
@@ -712,7 +875,7 @@ export class ChatHubService {
 
 					if (execution.status === 'canceled') {
 						await this.messageRepository.updateChatMessage(replyId, {
-							content: message || 'Generation cancelled.',
+							content: partialMessage || 'Generation cancelled.',
 							status: 'cancelled',
 						});
 						return;
@@ -737,13 +900,13 @@ export class ChatHubService {
 			// TODO: We should consider can we just save the output from the captured stream always instead
 			// of parsing it from execution data, which seems error prone, especially with custom workflows.
 			// That could make handling multiple agents, multiple runes, tool executions etc easier...?
-			const output = this.getAIOutput(execution, NODE_NAMES.REPLY_AGENT);
-			if (!output) {
-				throw new OperationalError('No response generated');
-			}
+			// const output = this.getAIOutput(execution, NODE_NAMES.REPLY_AGENT);
+			// if (!output) {
+			// 	throw new OperationalError('No response generated');
+			// }
 
 			await this.messageRepository.updateChatMessage(replyId, {
-				content: output,
+				content: partialMessage,
 				status: 'success',
 			});
 
@@ -795,7 +958,10 @@ export class ChatHubService {
 					text: "={{ $('When chat message received').item.json.chatInput }}",
 					options: {
 						enableStreaming: true,
-						maxTokensFromMemory: getMaxContextWindowTokens(model.provider, model.model),
+						maxTokensFromMemory:
+							model.provider !== 'n8n'
+								? getMaxContextWindowTokens(model.provider, model.model)
+								: undefined,
 					},
 				},
 				type: AGENT_LANGCHAIN_NODE_TYPE,
@@ -1050,8 +1216,13 @@ export class ChatHubService {
 
 	private createModelNode(
 		credentials: INodeCredentials,
-		{ provider, model }: ChatHubConversationModel,
+		conversationModel: ChatHubConversationModel,
 	): INode {
+		if (conversationModel.provider === 'n8n') {
+			throw new OperationalError('Custom agent workflows do not require a model node');
+		}
+
+		const { provider, model } = conversationModel;
 		const common = {
 			position: [600, 500] as [number, number],
 			id: uuidv4(),

packages/cli/src/modules/chat-hub/chat-hub.types.ts

@@ -1,7 +1,15 @@
-import type { ChatHubConversationModel, ChatMessageId, ChatSessionId } from '@n8n/api-types';
+import type {
+	ChatHubConversationModel,
+	ChatHubProvider,
+	ChatMessageId,
+	ChatSessionId,
+} from '@n8n/api-types';
 import type { INodeCredentials } from 'n8n-workflow';
 
-export interface ModelWithCredentials extends ChatHubConversationModel {
+export interface ModelWithCredentials {
+	provider: ChatHubProvider;
+	model?: string;
+	workflowId?: string;
 	credentialId: string | null;
 }
 

packages/cli/src/modules/chat-hub/context-limits.ts

@@ -1,4 +1,4 @@
-import type { ChatHubProvider } from '@n8n/api-types';
+import type { ChatHubLLMProvider } from '@n8n/api-types';
 
 /* eslint-disable @typescript-eslint/naming-convention */
 
@@ -8,7 +8,7 @@ import type { ChatHubProvider } from '@n8n/api-types';
 // If the limit is set to 0, it means either the model has no defined limit or the information
 // is not availabl and no context window trimming is applied. Similarly, if the model used is
 // not listed, no limit is applied.
-export const maxContextWindowTokens: Record<ChatHubProvider, Record<string, number>> = {
+export const maxContextWindowTokens: Record<ChatHubLLMProvider, Record<string, number>> = {
 	openai: {
 		'chatgpt-4o-latest': 128000,
 		'codex-mini-latest': 0,
@@ -140,7 +140,7 @@ export const maxContextWindowTokens: Record<ChatHubProvider, Record<string, numb
 };
 
 export const getMaxContextWindowTokens = (
-	provider: ChatHubProvider,
+	provider: ChatHubLLMProvider,
 	model: string,
 ): number | undefined => {
 	const limit = maxContextWindowTokens[provider]?.[model] ?? 0;

packages/frontend/editor-ui/src/features/ai/chatHub/ChatView.vue

@@ -19,6 +19,7 @@ import { useUsersStore } from '@/features/settings/users/users.store';
 import {
 	chatHubConversationModelSchema,
 	type ChatHubProvider,
+	type ChatHubLLMProvider,
 	chatHubProviderSchema,
 	PROVIDER_CREDENTIAL_TYPE_MAP,
 	type ChatHubConversationModel,
@@ -104,6 +105,10 @@ const selectedCredentials = useLocalStorage<CredentialsMap>(
 const autoSelectCredentials = computed<CredentialsMap>(() =>
 	Object.fromEntries(
 		chatHubProviderSchema.options.map((provider) => {
+			if (provider === 'n8n') {
+				return [provider, null];
+			}
+
 			const lastCreatedCredential =
 				credentialsStore
 					.getCredentialsByType(PROVIDER_CREDENTIAL_TYPE_MAP[provider])
@@ -125,6 +130,22 @@ const credentialsId = computed(() =>
 	selectedModel.value ? mergedCredentials.value[selectedModel.value.provider] : undefined,
 );
 
+const modelRequiresCredentials = computed(() => {
+	if (!selectedModel.value) return false;
+
+	return selectedModel.value?.provider !== 'n8n';
+});
+
+const isMissingSelectedCredential = computed(() => {
+	if (!selectedModel.value) return false;
+
+	if (!modelRequiresCredentials.value) {
+		return false;
+	}
+
+	return !credentialsId.value;
+});
+
 const editingMessageId = ref<string>();
 const didSubmitInCurrentSession = ref(false);
 const initialization = ref({ credentialsFetched: false, modelsFetched: false });
@@ -224,18 +245,28 @@ onMounted(async () => {
 });
 
 function onSubmit(message: string) {
-	if (!message.trim() || chatStore.isResponding || !selectedModel.value || !credentialsId.value) {
+	if (
+		!message.trim() ||
+		chatStore.isResponding ||
+		!selectedModel.value ||
+		isMissingSelectedCredential.value
+	) {
 		return;
 	}
 
 	didSubmitInCurrentSession.value = true;
 
-	chatStore.sendMessage(sessionId.value, message, selectedModel.value, {
-		[PROVIDER_CREDENTIAL_TYPE_MAP[selectedModel.value.provider]]: {
-			id: credentialsId.value,
-			name: '',
-		},
-	});
+	const credentials = {};
+	if (selectedModel.value.provider !== 'n8n' && credentialsId.value) {
+		Object.assign(credentials, {
+			[PROVIDER_CREDENTIAL_TYPE_MAP[selectedModel.value.provider]]: {
+				id: credentialsId.value,
+				name: '',
+			},
+		});
+	}
+
+	chatStore.sendMessage(sessionId.value, message, selectedModel.value, credentials);
 
 	inputRef.value?.setText('');
 
@@ -269,12 +300,20 @@ function handleEditMessage(message: ChatHubMessageDto) {
 
 	const messageToEdit = message.revisionOfMessageId ?? message.id;
 
-	chatStore.editMessage(sessionId.value, messageToEdit, message.content, selectedModel.value, {
-		[PROVIDER_CREDENTIAL_TYPE_MAP[selectedModel.value.provider]]: {
-			id: credentialsId.value,
-			name: '',
-		},
-	});
+	chatStore.editMessage(
+		sessionId.value,
+		messageToEdit,
+		message.content,
+		selectedModel.value,
+		selectedModel.value.provider === 'n8n'
+			? {}
+			: {
+					[PROVIDER_CREDENTIAL_TYPE_MAP[selectedModel.value.provider]]: {
+						id: credentialsId.value,
+						name: '',
+					},
+				},
+	);
 	editingMessageId.value = undefined;
 }
 
@@ -290,27 +329,34 @@ function handleRegenerateMessage(message: ChatHubMessageDto) {
 
 	const messageToRetry = message.retryOfMessageId ?? message.id;
 
-	chatStore.regenerateMessage(sessionId.value, messageToRetry, selectedModel.value, {
-		[PROVIDER_CREDENTIAL_TYPE_MAP[selectedModel.value.provider]]: {
-			id: credentialsId.value,
-			name: '',
-		},
-	});
+	chatStore.regenerateMessage(
+		sessionId.value,
+		messageToRetry,
+		selectedModel.value,
+		selectedModel.value.provider === 'n8n'
+			? {}
+			: {
+					[PROVIDER_CREDENTIAL_TYPE_MAP[selectedModel.value.provider]]: {
+						id: credentialsId.value,
+						name: '',
+					},
+				},
+	);
 }
 
 function handleSelectModel(selection: ChatHubConversationModel) {
 	selectedModel.value = selection;
 }
 
-function handleSelectCredentials(provider: ChatHubProvider, credentialsId: string) {
-	selectedCredentials.value = { ...selectedCredentials.value, [provider]: credentialsId };
+function handleSelectCredentials(provider: ChatHubProvider, id: string) {
+	selectedCredentials.value = { ...selectedCredentials.value, [provider]: id };
 }
 
 function handleSwitchAlternative(messageId: string) {
 	chatStore.switchAlternative(sessionId.value, messageId);
 }
 
-function handleConfigureCredentials(provider: ChatHubProvider) {
+function handleConfigureCredentials(provider: ChatHubLLMProvider) {
 	const credentialType = PROVIDER_CREDENTIAL_TYPE_MAP[provider];
 	const existingCredentials = credentialsStore.getCredentialsByType(credentialType);
 
@@ -327,7 +373,7 @@ function handleConfigureModel() {
 	headerRef.value?.openModelSelector();
 }
 
-function handleCreateNewCredential(provider: ChatHubProvider) {
+function handleCreateNewCredential(provider: ChatHubLLMProvider) {
 	uiStore.openNewCredential(PROVIDER_CREDENTIAL_TYPE_MAP[provider]);
 }
 </script>
@@ -352,7 +398,7 @@ function handleCreateNewCredential(provider: ChatHubProvider) {
 		/>
 
 		<CredentialSelectorModal
-			v-if="credentialSelectorProvider"
+			v-if="credentialSelectorProvider && credentialSelectorProvider !== 'n8n'"
 			:key="credentialSelectorProvider"
 			:provider="credentialSelectorProvider"
 			:initial-value="mergedCredentials[credentialSelectorProvider] ?? null"
@@ -411,7 +457,7 @@ function handleCreateNewCredential(provider: ChatHubProvider) {
 						:class="$style.prompt"
 						:is-responding="chatStore.isResponding"
 						:selected-model="selectedModel"
-						:is-credentials-selected="!!credentialsId"
+						:is-missing-credentials="isMissingSelectedCredential"
 						@submit="onSubmit"
 						@stop="onStop"
 						@select-model="handleConfigureModel"

packages/frontend/editor-ui/src/features/ai/chatHub/chat.store.ts

@@ -366,7 +366,7 @@ export const useChatStore = defineStore(CHAT_STORE, () => {
 			name: 'User',
 			content: message,
 			provider: null,
-			model: model?.model ?? null,
+			model: model.provider === 'n8n' ? null : model.model,
 			workflowId: null,
 			executionId: null,
 			status: 'success',

packages/frontend/editor-ui/src/features/ai/chatHub/chat.utils.ts

@@ -11,7 +11,7 @@ export function findOneFromModelsResponse(
 ): ChatHubConversationModel | undefined {
 	for (const provider of chatHubProviderSchema.options) {
 		if (response[provider].models.length > 0) {
-			return { model: response[provider].models[0].name, provider, workflowId: null };
+			return response[provider].models[0];
 		}
 	}
 

packages/frontend/editor-ui/src/features/ai/chatHub/components/ChatConversationHeader.vue

@@ -5,11 +5,7 @@ import ModelSelector from '@/features/ai/chatHub/components/ModelSelector.vue';
 import { useChatHubSidebarState } from '@/features/ai/chatHub/composables/useChatHubSidebarState';
 import { CHAT_VIEW } from '@/features/ai/chatHub/constants';
 import { useCredentialsStore } from '@/features/credentials/credentials.store';
-import {
-	type ChatHubConversationModel,
-	type ChatHubProvider,
-	type ChatSessionId,
-} from '@n8n/api-types';
+import type { ChatHubConversationModel, ChatHubLLMProvider, ChatSessionId } from '@n8n/api-types';
 import { N8nIconButton } from '@n8n/design-system';
 import { computed, useTemplateRef } from 'vue';
 import { useRouter } from 'vue-router';
@@ -21,7 +17,7 @@ const { selectedModel, credentials } = defineProps<{
 
 const emit = defineEmits<{
 	selectModel: [ChatHubConversationModel];
-	setCredentials: [provider: ChatHubProvider];
+	setCredentials: [provider: ChatHubLLMProvider];
 	renameConversation: [id: ChatSessionId, title: string];
 }>();
 

packages/frontend/editor-ui/src/features/ai/chatHub/components/ChatMessage.vue

@@ -48,7 +48,7 @@ const speech = useSpeechSynthesis(messageContent, {
 });
 
 const credentialTypeName = computed(() => {
-	if (message.type !== 'ai' || !message.provider) {
+	if (message.type !== 'ai' || !message.provider || message.provider === 'n8n') {
 		return null;
 	}
 	return PROVIDER_CREDENTIAL_TYPE_MAP[message.provider] ?? null;

packages/frontend/editor-ui/src/features/ai/chatHub/components/ChatPrompt.vue

@@ -1,26 +1,25 @@
 <script setup lang="ts">
 import { useToast } from '@/composables/useToast';
 import { providerDisplayNames } from '@/features/ai/chatHub/constants';
-import type { ChatHubConversationModel, ChatHubProvider } from '@n8n/api-types';
+import type { ChatHubConversationModel, ChatHubLLMProvider } from '@n8n/api-types';
 import { N8nIconButton, N8nInput } from '@n8n/design-system';
 import { useSpeechRecognition } from '@vueuse/core';
-import { computed } from 'vue';
-import { ref, useTemplateRef, watch } from 'vue';
+import { computed, ref, useTemplateRef, watch } from 'vue';
 
-const { selectedModel } = defineProps<{
+const { selectedModel, isMissingCredentials } = defineProps<{
 	isResponding: boolean;
 	selectedModel: ChatHubConversationModel | null;
-	isCredentialsSelected: boolean;
+	isMissingCredentials: boolean;
 }>();
 
 const emit = defineEmits<{
 	submit: [string];
 	stop: [];
 	selectModel: [];
-	setCredentials: [ChatHubProvider];
+	setCredentials: [ChatHubLLMProvider];
 }>();
 
-const inputRef = useTemplateRef('inputRef');
+const inputRef = useTemplateRef<HTMLElement>('inputRef');
 const message = ref('');
 
 const toast = useToast();
@@ -36,9 +35,7 @@ const placeholder = computed(() => {
 		return 'Select a model';
 	}
 
-	const modelName = selectedModel.model;
-
-	return `Message ${modelName}`;
+	return `Message ${selectedModel.name}`;
 });
 
 function onMic() {
@@ -110,9 +107,12 @@ defineExpose({
 				Please <a href="" @click.prevent="emit('selectModel')">select a model</a> to start a
 				conversation
 			</div>
-			<div v-else-if="!isCredentialsSelected" :class="$style.callout">
+			<div v-else-if="isMissingCredentials" :class="$style.callout">
 				Please
-				<a href="" @click.prevent="emit('setCredentials', selectedModel.provider)">
+				<a
+					href=""
+					@click.prevent="emit('setCredentials', selectedModel.provider as ChatHubLLMProvider)"
+				>
 					set credentials
 				</a>
 				for {{ providerDisplayNames[selectedModel.provider] }} to start a conversation
@@ -126,7 +126,7 @@ defineExpose({
 				autocomplete="off"
 				:autosize="{ minRows: 1, maxRows: 6 }"
 				autofocus
-				:disabled="!isCredentialsSelected || !selectedModel"
+				:disabled="isMissingCredentials || !selectedModel"
 				@keydown="handleKeydownTextarea"
 			/>
 
@@ -136,7 +136,7 @@ defineExpose({
 					native-type="button"
 					type="secondary"
 					title="Attach"
-					:disabled="!isCredentialsSelected || !selectedModel || isResponding"
+					:disabled="isMissingCredentials || !selectedModel || isResponding"
 					icon="paperclip"
 					icon-size="large"
 					text
@@ -147,7 +147,7 @@ defineExpose({
 					native-type="button"
 					:title="speechInput.isListening.value ? 'Stop recording' : 'Voice input'"
 					type="secondary"
-					:disabled="!isCredentialsSelected || !selectedModel || isResponding"
+					:disabled="isMissingCredentials || !selectedModel || isResponding"
 					:icon="speechInput.isListening.value ? 'square' : 'mic'"
 					:class="{ [$style.recording]: speechInput.isListening.value }"
 					icon-size="large"
@@ -156,7 +156,7 @@ defineExpose({
 				<N8nIconButton
 					v-if="!isResponding"
 					native-type="submit"
-					:disabled="!isCredentialsSelected || !selectedModel || !message.trim()"
+					:disabled="isMissingCredentials || !selectedModel || !message.trim()"
 					title="Send"
 					icon="arrow-up"
 					icon-size="large"

packages/frontend/editor-ui/src/features/ai/chatHub/components/ChatSessionMenuItem.vue

@@ -102,7 +102,11 @@ watch(
 			/>
 		</template>
 		<template #icon>
-			<N8nIcon v-if="session.provider === null" size="medium" icon="message-circle" />
+			<N8nIcon
+				v-if="session.provider === null || session.provider === 'n8n'"
+				size="medium"
+				icon="message-circle"
+			/>
 			<CredentialIcon
 				v-else
 				:credential-type-name="PROVIDER_CREDENTIAL_TYPE_MAP[session.provider]"

packages/frontend/editor-ui/src/features/ai/chatHub/components/CredentialSelectorModal.vue

@@ -5,18 +5,18 @@ import Modal from '@/components/Modal.vue';
 import { useCredentialsStore } from '@/features/credentials/credentials.store';
 import type { ICredentialsResponse } from '@/features/credentials/credentials.types';
 import { createEventBus } from '@n8n/utils/event-bus';
-import { PROVIDER_CREDENTIAL_TYPE_MAP, type ChatHubProvider } from '@n8n/api-types';
+import { type ChatHubLLMProvider, PROVIDER_CREDENTIAL_TYPE_MAP } from '@n8n/api-types';
 import { providerDisplayNames } from '@/features/ai/chatHub/constants';
 import CredentialIcon from '@/features/credentials/components/CredentialIcon.vue';
 
 const props = defineProps<{
-	provider: ChatHubProvider;
+	provider: ChatHubLLMProvider;
 	initialValue: string | null;
 }>();
 
 const emit = defineEmits<{
-	select: [provider: ChatHubProvider, credentialId: string];
-	createNew: [provider: ChatHubProvider];
+	select: [provider: ChatHubLLMProvider, credentialId: string];
+	createNew: [provider: ChatHubLLMProvider];
 }>();
 
 const credentialsStore = useCredentialsStore();

packages/frontend/editor-ui/src/features/ai/chatHub/components/ModelSelector.vue

@@ -3,11 +3,12 @@ import { computed, useTemplateRef } from 'vue';
 import { N8nNavigationDropdown, N8nIcon, N8nButton, N8nText } from '@n8n/design-system';
 import { type ComponentProps } from 'vue-component-type-helpers';
 import {
-	type ChatHubConversationModel,
-	type ChatHubProvider,
 	chatHubProviderSchema,
-	type ChatModelsResponse,
 	PROVIDER_CREDENTIAL_TYPE_MAP,
+	type ChatHubConversationModel,
+	type ChatModelsResponse,
+	type ChatHubLLMProvider,
+	type ChatHubProvider,
 } from '@n8n/api-types';
 import { providerDisplayNames } from '@/features/ai/chatHub/constants';
 import CredentialIcon from '@/features/credentials/components/CredentialIcon.vue';
@@ -21,63 +22,87 @@ const props = defineProps<{
 
 const emit = defineEmits<{
 	change: [ChatHubConversationModel];
-	configure: [ChatHubProvider];
+	configure: [ChatHubLLMProvider];
 }>();
 
 const dropdownRef = useTemplateRef('dropdownRef');
 
 const menu = computed(() =>
-	chatHubProviderSchema.options.map((provider) => {
-		const models = props.models?.[provider].models ?? [];
-		const error = props.models?.[provider].error;
+	chatHubProviderSchema.options
+		.filter((provider) => provider !== 'n8n') // Hide n8n provider for now
+		.map((provider: ChatHubProvider) => {
+			const models = props.models?.[provider].models ?? [];
+			const error = props.models?.[provider].error;
 
-		const modelOptions =
-			models.length > 0
-				? models.map<ComponentProps<typeof N8nNavigationDropdown>['menu'][number]>((model) => ({
-						id: `${provider}::${model.name}`,
-						title: model.name,
-						disabled: false,
-					}))
-				: error
-					? [{ id: `${provider}::error`, disabled: true, title: error }]
-					: [];
+			const modelOptions =
+				models.length > 0
+					? models.map<ComponentProps<typeof N8nNavigationDropdown>['menu'][number]>((model) => {
+							const identifier = model.provider === 'n8n' ? model.workflowId : model.model;
 
-		return {
-			id: provider,
-			title: providerDisplayNames[provider],
-			submenu: modelOptions.concat([
-				...(modelOptions.length > 0 ? [{ isDivider: true as const, id: 'divider' }] : []),
-				{
+							return {
+								id: `${provider}::${identifier}`,
+								title: model.name,
+								disabled: false,
+							};
+						})
+					: error
+						? [{ id: `${provider}::error`, value: null, disabled: true, title: error }]
+						: [];
+
+			const submenu = modelOptions.concat([
+				...(provider !== 'n8n' && modelOptions.length > 0
+					? [{ isDivider: true as const, id: 'divider' }]
+					: []),
+			]);
+
+			if (provider !== 'n8n') {
+				submenu.push({
 					id: `${provider}::configure`,
 					icon: 'settings',
 					title: 'Configure credentials...',
 					disabled: false,
-				},
-			]),
-		};
-	}),
+				});
+			}
+
+			return {
+				id: provider,
+				hidden: true,
+				title: providerDisplayNames[provider],
+				submenu,
+			};
+		}),
 );
 
 const selectedLabel = computed(() => {
 	if (!props.selectedModel) return 'Select model';
-	return props.selectedModel.model;
+	return props.selectedModel.name;
 });
 
 function onSelect(id: string) {
-	// Format is "provider::model"
-	const [provider, model] = id.split('::');
+	// Format is "provider::identifier", where identifier is either "configure", model name, or workflow ID for n8n
+	const [provider, identifier] = id.split('::');
 	const parsedProvider = chatHubProviderSchema.safeParse(provider).data;
 
 	if (!parsedProvider) {
 		return;
 	}
 
-	if (model === 'configure') {
+	if (identifier === 'configure' && parsedProvider !== 'n8n') {
 		emit('configure', parsedProvider);
 		return;
 	}
 
-	emit('change', { provider: parsedProvider, model, workflowId: null });
+	const model = parsedProvider === 'n8n' ? null : identifier;
+	const workflowId = parsedProvider === 'n8n' ? identifier : null;
+	const selected = props.models?.[parsedProvider].models.find((m) =>
+		m.provider === 'n8n' ? m.workflowId === workflowId : m.model === model,
+	);
+
+	if (!selected) {
+		return;
+	}
+
+	emit('change', selected);
 }
 
 onClickOutside(
@@ -95,7 +120,7 @@ defineExpose({
 		<template #item-icon="{ item }">
 			<CredentialIcon
 				v-if="item.id in PROVIDER_CREDENTIAL_TYPE_MAP"
-				:credential-type-name="PROVIDER_CREDENTIAL_TYPE_MAP[item.id as ChatHubProvider]"
+				:credential-type-name="PROVIDER_CREDENTIAL_TYPE_MAP[item.id as ChatHubLLMProvider]"
 				:size="16"
 				:class="$style.menuIcon"
 			/>
@@ -103,8 +128,10 @@ defineExpose({
 
 		<N8nButton :class="$style.dropdownButton" type="secondary" text>
 			<CredentialIcon
-				v-if="selectedModel"
-				:credential-type-name="PROVIDER_CREDENTIAL_TYPE_MAP[selectedModel.provider]"
+				v-if="selectedModel && selectedModel.provider in PROVIDER_CREDENTIAL_TYPE_MAP"
+				:credential-type-name="
+					PROVIDER_CREDENTIAL_TYPE_MAP[selectedModel.provider as ChatHubLLMProvider]
+				"
 				:size="credentialsName ? 20 : 16"
 				:class="$style.icon"
 			/>

packages/frontend/editor-ui/src/features/ai/chatHub/constants.ts

@@ -10,6 +10,7 @@ export const providerDisplayNames: Record<ChatHubProvider, string> = {
 	openai: 'OpenAI',
 	anthropic: 'Anthropic',
 	google: 'Google',
+	n8n: 'n8n',
 };
 
 export const MOBILE_MEDIA_QUERY = '(max-width: 768px)';

packages/workflow/src/constants.ts

@@ -41,6 +41,7 @@ export const AI_TRANSFORM_NODE_TYPE = 'n8n-nodes-base.aiTransform';
 export const FORM_NODE_TYPE = 'n8n-nodes-base.form';
 export const FORM_TRIGGER_NODE_TYPE = 'n8n-nodes-base.formTrigger';
 export const CHAT_TRIGGER_NODE_TYPE = '@n8n/n8n-nodes-langchain.chatTrigger';
+export const RESPOND_TO_CHAT_NODE_TYPE = '@n8n/n8n-nodes-langchain.chat';
 export const WAIT_NODE_TYPE = 'n8n-nodes-base.wait';
 export const RESPOND_TO_WEBHOOK_NODE_TYPE = 'n8n-nodes-base.respondToWebhook';
 export const HTML_NODE_TYPE = 'n8n-nodes-base.html';