n8n has two kinds of built-in OpenAI nodes: the OpenAI app node, which you use as a normal workflow step for text, images, audio, files, video and conversations, and the OpenAI Chat Model sub-node, which gives an AI Agent its model. Both use one OpenAI credential. Community nodes add things the built-in nodes don’t cover, mostly other gateways and providers that speak the OpenAI API, Azure OpenAI sign-in options, and observability. This guide covers the built-in nodes and their common errors (from n8n’s documentation) and the most used OpenAI community packages (from our own npm data).
Which OpenAI node do you need?
| You want to | Use |
|---|---|
| Generate text, analyze or create images, transcribe audio, manage files, make a video, as a workflow step | OpenAI node (built in) |
| Give an AI Agent an OpenAI model, with web search, file search or code interpreter | OpenAI Chat Model (built in, sub-node) |
| Use OpenAI through LiteLLM, Langfuse, Azure with Microsoft sign-in, mTLS or another gateway | A community node (see below) |
Credentials
One OpenAI credential works for the OpenAI node, the OpenAI Chat Model, Embeddings OpenAI and LM OpenAI. You need:
- An API Key: in your OpenAI account, open the API keys page, select Create new secret key and copy it into n8n.
- An Organization ID, only if you belong to several OpenAI organizations. You find it on your Organization Settings page. Requests made with it count toward that organization’s quota.
On n8n Cloud, you can skip the credential: select Use Gateway credits in the credential field of nodes that support it, and the node runs without an OpenAI account.
The OpenAI node
Operations:
- Text: Generate a Chat Completion, Generate a Model Response, Classify Text for Violations
- Image: Analyze Image, Generate an Image, Edit an Image
- Audio: Generate Audio, Transcribe a Recording, Translate a Recording
- File: Delete a File, List Files, Upload a File
- Video: Generate a Video
- Conversation: Create, Get, Update and Remove a Conversation
Version notes from n8n’s docs: OpenAI node V2 is available from n8n 1.117.0, supports OpenAI’s Responses API, and drops the Assistants API, which OpenAI plans to retire. Since n8n 1.29.0 the OpenAI node replaces the old OpenAI Assistant node.
The two Text operations, Generate a Chat Completion and Generate a Model Response, accept tools. When you connect a tool, the OpenAI node becomes a root node that forms a cluster with the tool sub-nodes.
The OpenAI Chat Model sub-node
Use it as the model for an AI Agent or another AI root node. n8n loads the Model list from OpenAI, so you only see the models your account can use.
- Use Responses API: off by default, which uses the Chat Completions API, where you manage conversation state yourself, for example with a Simple Memory sub-node. Turned on, it uses the Responses API, which can call several built-in tools in one request and keep a conversation with a
conversation_id. n8n’s docs note that OpenAI recommends the Responses API for new projects. - Built-in tools (Responses API only): Web Search, File Search over files you uploaded, and Code Interpreter for Python in a sandbox. They only work when the chat model is connected to an AI Agent node, not to a Basic LLM Chain, for example.
- Options: Frequency Penalty, Maximum Number of Tokens, Presence Penalty, Sampling Temperature, Timeout, Max Retries, Top P. With the Responses API also Conversation ID, Prompt Cache Key, Safety Identifier, Service Tier (Auto, Flex, Default, Priority), Metadata (up to 16 pairs), Top Logprobs (0 to 20), Output Format (Text, JSON Schema or JSON Object) and Prompt (a prompt saved in the OpenAI dashboard).
One thing to know: the chat model is a sub-node, and in sub-nodes an expression such as {{ $json.name }} always resolves to the first input item, not to each item in turn.
Common errors and fixes
“The service is receiving too many requests from you”
You hit OpenAI’s rate limits. n8n suggests two workarounds: split the data with a Loop Over Items node and add a Wait node at the end of each batch, or call the OpenAI API with the HTTP Request node and its built-in batch limit option instead of the OpenAI node.
“Insufficient quota”
Your OpenAI account doesn’t have enough credit or capacity: the trial ended, the balance is empty, or a usage limit was reached. In your OpenAI settings, select the right organization and project, then check the billing overview, the organization’s usage limits and the project’s limits, and OpenAI’s status page. A new balance can take a while to show. In n8n, make sure the credential uses a valid key from the account you topped up and that the node uses that credential. n8n also notes that quota errors sometimes persist right after a top-up, and recommends creating a new API key once the account has credit.
“Bad request – please check your parameters”
n8n couldn’t interpret OpenAI’s error message. Run the same request with the HTTP Request node to see the detailed error.
OpenAI community nodes
Our OpenAI hub lists 24 packages that pass our quality rule, with 8,319 npm downloads a week between them. The most downloaded, with the description each author gives on npm (weekly npm downloads as of 27 October 2026):
| Package | Weekly downloads | Author’s description |
|---|---|---|
| n8n-nodes-openai-advanced | 4,580 | OpenAI with LiteLLM prompt caching support |
| n8n-nodes-openai-langfuse | 1,279 | Langfuse + OpenAI-compatible LLM provider |
| n8n-nodes-chatgpt-plan | 1,241 | ChatGPT Codex chat model with device-code account connection, for self-hosted n8n |
| n8n-nodes-azure-openai-ms-oauth2 | 415 | Azure OpenAI Chat Model and Embeddings with Microsoft OAuth2 |
| n8n-nodes-openai-gpt5 | 166 | GPT-5 with PDF processing, web search and timeout settings via the Responses API |
| @rlquilez/n8n-nodes-openai-litellm | 130 | OpenAI-compatible LLM provider with structured JSON metadata injection |
| n8n-nodes-azure-openai-delegate | 88 | Azure OpenAI chat model through an APIM gateway with delegated Microsoft Entra OAuth2 |
| n8n-nodes-azure-openai-image | 44 | Azure OpenAI image generation (DALL-E) |
| n8n-nodes-mtls-openai | 37 | OpenAI-compatible Chat Model and Embeddings with mTLS client certificates |
As the descriptions show, most of these are about how you reach the model, not about new OpenAI features: through LiteLLM or Langfuse, through Azure with Microsoft sign-in or an API gateway, or behind mTLS. Read each README for setup, and check its node page here for the latest release and, where it has one, its GitHub repository for open issues.
How to choose
- Calling OpenAI directly with an API key, or with Gateway credits on n8n Cloud: the built-in nodes.
- Going through LiteLLM, Langfuse, Azure with Microsoft sign-in, or an mTLS proxy: the matching community node.
To install a package on a self-hosted instance, an owner or admin enters its npm name in Settings > Community Nodes; full steps are in How to Install n8n Community Nodes. On n8n Cloud only verified community nodes can be installed, see Verified vs Unverified n8n Community Nodes. Related: DeepSeek in n8n and MCP in n8n.
FAQ
Can I use OpenAI in n8n without an API key?
On n8n Cloud, yes: select Use Gateway credits in the credential field. Self-hosted n8n needs your own OpenAI API key.
What is the difference between the OpenAI node and the OpenAI Chat Model?
The OpenAI node is a normal workflow step for text, image, audio, file, video and conversation operations. The OpenAI Chat Model is a sub-node that provides the model to an AI Agent or other AI root node.
How do I give an n8n agent web search with OpenAI?
In the OpenAI Chat Model, turn on Use Responses API and enable the Web Search built-in tool, and connect the model to an AI Agent node.
Why does n8n say “insufficient quota” after I added credit?
The new balance can take a while to show, and n8n’s docs recommend issuing a new API key once the account has credit. Also check that the credential’s key belongs to the organization and project you topped up.
