Get started with n8n: first agent, AI API keys and DeepSeek

n8n is installed and the admin account is created — what’s next? Almost everything people install n8n for in 2026 is related to AI: agents, chatbots, and text processing. In this article, we’ll look at the first n8n screen, choosing between an agent and a workflow, language model API keys, and connecting DeepSeek.

This article is written for the full n8n installation — with a sandbox, Python, and web search, as described in the “Full n8n installation” section. If you have a single n8n container from the quick installation, some features will be unavailable — see the last section.

This article was checked in November 2026 on n8n 2.46.

What to do immediately after installation

• Get a free Community Edition key. It unlocks folders for workflows, editor debugging, and user execution data. The process is described in the “Free key” section of the installation guide.
• Enable two-factor authentication if the n8n panel is accessible from the internet: Settings → Personal.
• Check the time zone. Scheduled workflows depend on it: in the full installation, it is set by the GENERIC_TIMEZONE and TZ lines in ~/n8n/.env.

First screen: Build an agent or Build a workflow

After signing in, n8n offers two ways to get started — Let's build your first automation:

• Build an agent — create an Agent in the Agent Builder. An Agent is a separate object in a project, not part of a workflow: it has a model, instructions, tools, memory, a knowledge base, schedules, and channels (Telegram, Slack, Linear, Microsoft Teams). You can test it in the Preview window and publish it with the Publish button.
• Build a workflow — open a blank canvas and build a workflow from nodes: a trigger, data processing, and integrations. AI is added to a workflow with the AI Agent node and child nodes for the model, memory, and tools.

The n8n documentation explicitly warns that an Agent from Agent Builder and the AI Agent node are different things. An Agent exists independently, while the AI Agent node works only inside a workflow. They can be connected: the Message an Agent node sends a message to an Agent from any workflow.

What to choose first:

• A chatbot or assistant that replies in a messenger, searches the web, and uses your documents — Build an agent. It is faster because you do not need to assemble a chain of nodes.
• Automation with clearly defined steps (website inquiry → CRM → notification) or AI used as one step among others — Build a workflow.

Where to get a language model API key

Both Agents and AI nodes in workflows need a language model API key. The key is saved once as Credentials and can then be connected to any agents and nodes from that provider. The main options are:

DeepSeek in an Agent (Build an agent)

1. Click Build an agent (or open the Agents tab in a project → Create Agent) and enter the Agent’s name.
2. On the Agent tab, open the Model field and select the DeepSeek provider, then choose a model. If you have not created DeepSeek credentials yet, n8n will prompt you to do so — paste in the key (see the DeepSeek in the assistant section for where to get a key and which models are available).
3. In the Instructions field, describe what the Agent should do: its role, response style, and limitations.
4. If necessary, add tools (Tools → Add tool), enable web search (Capabilities → Web search), and upload documents on the Knowledge tab.
5. Test the Agent in the Preview window and click Publish.

Web search in the full installation runs through a self-hosted SearXNG instance, while the knowledge base and code execution run through the sandbox. The sandbox connection is available under Settings → Agents → Shared sandbox: it is used by both Agents and n8n Assistant.

DeepSeek in a workflow (Build a workflow)

1. In n8n, open Overview → Credentials → Create credential, select DeepSeek, and paste in the key (if you already created one for the Agent, this step is not necessary).
2. In the workflow, connect a DeepSeek Chat Model node to the AI Agent node (or Basic LLM Chain) and select these credentials.
3. Select a model in the Model field — n8n will load the models available to your account.

If you do not have a DeepSeek account or want to keep all models on a single balance, connect DeepSeek through OpenRouter: create OpenRouter credentials, connect an OpenRouter Chat Model node to the Agent, and select a DeepSeek model from the list. The same key will work for other models as well — you can compare DeepSeek with Claude or GPT by simply switching the model in the node.

n8n Assistant: an assistant in the editor

n8n Assistant is a built-in AI assistant. It can:

• build and edit workflows based on a text description — you write in the chat what you want to automate, and the assistant adds and configures the nodes;
• suggest instructions, tools, and skills for Agents;
• analyze failed executions and suggest fixes;
• answer questions about n8n while taking into account what is currently open in the editor.

The full installation already includes the sandbox, so the assistant only needs a key. It is configured under Settings → Assistant, with four providers available: Anthropic, OpenAI, OpenRouter, and Self-hosted or OpenAI-compatible endpoint.

DeepSeek in the assistant (n8n Assistant)

DeepSeek is one of the most affordable and capable options for both the assistant and Agents: its models support tool calling. The same key is used across n8n:

1. On platform.deepseek.com, open API keys and click Create new secret key.
2. Copy the key: it will not be shown again. Requests are charged against your account balance in the Top up section.

The DeepSeek API currently has two models: deepseek-flash (fast and inexpensive, suitable for most tasks) and deepseek-v4-pro (more capable for complex tasks and more expensive).

Base URL is the API address to which n8n sends requests. In the DeepSeek credentials (for both the Agent and the DeepSeek Chat Model node), leave the default value https://api.deepseek.com: this is DeepSeek’s OpenAI-compatible endpoint. Use the same address for the assistant.

DeepSeek does not have a separate entry in the n8n Assistant provider list, but its API is compatible with OpenAI, so there are two options.

Using your own DeepSeek key. Select the Self-hosted or OpenAI-compatible endpoint provider, enter https://api.deepseek.com as the endpoint, and provide the DeepSeek key. With environment variables, these are N8N_INSTANCE_AI_MODEL_URL and N8N_INSTANCE_AI_MODEL_API_KEY. Note that the n8n community has reported that requests may still be sent to OpenAI with this configuration. If the assistant does not respond, check this first.

Through OpenRouter — the option documented by n8n. Add the following lines to ~/n8n/.env and restart the containers (docker compose up -d):

N8N_INSTANCE_AI_MODEL=openrouter/deepseek/deepseek-v4-pro
N8N_INSTANCE_AI_MODEL_API_KEY=your_OpenRouter_key

If you have a minimal installation

The quick installation from the n8n installation guide runs a single n8n container — without a sandbox, Python runners, or SearXNG. This creates the following limitations:

• n8n Assistant does not work, even if a key is configured: it executes code in the sandbox and cannot start without it.
• Agents lose some capabilities: the knowledge base and code execution also require the sandbox, while web search requires SearXNG or the paid Brave Search service.
• The Code node supports only JavaScript; Python runs in the separate runners container.

Workflows (Build a workflow) and the AI Agent node work without restrictions: they only need a model API key. DeepSeek is connected in the same way, with the DeepSeek Chat Model node or through OpenRouter, as described above. The model runs on the provider’s side, so you do not need a powerful server.

To get all features, you have two options:

1. Switch to the full installation. It requires a VPS with 2 vCPU and 4 GB RAM.
2. Connect the cloud-based Daytona sandbox (a separate account): n8n recommends it for production servers if you do not want to run a privileged container on your own server.

What’s next

The key is connected — you can start building your first Agent. Step-by-step examples, from an Agent with a calculator to a Telegram bot with memory, are available in the AI Agents in n8n article. If n8n turns out not to be right for you, check out n8n alternatives.