The OpenAI Chat Model sub-node (LmChatOpenAi) is the primary interface for integrating GPT-4o, GPT-4o-mini, and all OpenAI chat models into n8n workflows. It connects to the AI Agent node as a Language Model provider, authenticates via API key (stored with AES‑256‑CBC encryption), and dynamically loads models available to your account. The node exposes Temperature (0–2, default 1) to control randomness, Maximum Number of Tokens to cap response length (up to 32,768 tokens for GPT-4o), Response Format (Text, JSON Schema, or JSON Object), Frequency Penalty (−2.0 to 2.0), Presence Penalty (−2.0 to 2.0), system messages to define assistant behavior, and a configurable Timeout with Max Retries for production resilience [1] [2].

How do you configure OpenAI credentials and select the optimal GPT-4o model in n8n?

Navigate to Settings → Credentials → New Credential, select OpenAI API, and paste your API key from platform.openai.com/api-keys. n8n encrypts the key with AES‑256‑CBC and stores it as an encrypted password field. For OAuth2 users, a community draft PR adds access-token support alongside existing API key authentication, enabling usage with ChatGPT Plus subscriptions. Once saved, add an AI Agent or Basic LLM Chain root node, attach an OpenAI Chat Model sub-node, and select gpt-4o from the dynamically-loaded Model dropdown. [1] [4]

For high-volume content generation pipelines, begin prototyping with gpt-4o-mini ($0.15 per 1 M input tokens) — its lower cost and faster response times allow rapid iteration. Switch to full gpt-4o only when output quality from the mini variant is unacceptable. The Model dropdown displays only models available to your specific API account; if a model is missing, verify your key has the model:read scope and check platform.openai.com/usage for any account-level restrictions. For production credential management, use a dedicated OpenAI API key scoped to the specific models your workflows require, and rotate keys every 90 days. For the complete credential security reference, see the n8n Credential Nodes guide.

How do you craft effective system messages, user prompts, and assistant roles in the n8n OpenAI node?

The OpenAI Chat Model node supports three message roles that form a conversation structure. The System message defines persistent behavioral instructions — the agent’s personality, rules, output format constraints — and cannot be overridden by subsequent user inputs. The User message carries the actual query, task, or data to process. The Assistant message provides example responses, establishes tone, or supplies context from prior turns. Messages are constructed as a collection of role‑content pairs. [6] [3]

For system prompts defining the agent’s operational constraints, include the agent persona description, the task’s scope (e.g., “You classify support emails into billing, technical, account, or general categories”), output format instructions (e.g., “Respond ONLY in valid JSON with keys for category, urgency, and summary”), and explicit guardrails (e.g., “If uncertain, set urgency to ‘low’ and state uncertainty in the summary”). In n8n’s AI Agent, access the system message by clicking “Add Option” at the bottom of the agent configuration panel and selecting “System Message” from the dropdown. The system message is processed before every user interaction and remains fixed throughout the conversation, while the user message changes with each input. For the complete prompt engineering framework with six real‑world examples ranging from basic chat to multi‑step orchestration, see the n8n OpenAI Prompt Chain Tutorial.

How do you parse GPT-4o outputs into structured JSON and validate the response schema?

The OpenAI Chat Model node supports three response formats: Text (free-form natural language), JSON Object (guarantees valid JSON output from the model), and JSON Schema (enforces a user-defined JSON schema so the model output conforms exactly to the specified structure). For production automation, always use JSON Schema or JSON Object — never rely on parsing free-text responses, which can vary in format and break downstream nodes. [1] [7]

For Basic LLM Chain users, add a Structured Output Parser sub-node — it handles schema enforcement automatically without touching the Options panel. For Code‑node–based parsing, extract the response content via $json.choices[0].message.content, parse with JSON.parse() inside a try‑catch block, and return the structured object. The community n8n-nodes-openai-structured-outputs node provides an alternative Extract JSON operation that accepts an arbitrary JSON schema and unstructured text, then returns structured JSON validated against that schema. The key prescription: instruct the model in the system message to respond “ONLY in valid JSON” with explicit field names and types, then validate the output schema with a Code or Output Parser node before downstream consumption. For production, add a fallback Code node that catches JSON parsing failures and retries the OpenAI call with a clarified prompt. For the complete output parsing reference, see the n8n Code Node Transformation guide.

How do you build a multi-step prompt chain with sequential GPT-4o calls in n8n?

A prompt chain connects multiple AI Agent or Basic LLM Chain nodes in sequence, where each node’s output feeds the next node’s input. The canonical three‑stage chain for content generation: Stage 1 – Outline: an OpenAI node generates a 5‑section article outline; Stage 2 – Write: a Set node extracts the outline text and feeds it into a second OpenAI node that writes the full article; Stage 3 – Evaluate: a third OpenAI node scores the draft against SEO and readability criteria. [8] [9]

For branching chains, insert an IF node after the first AI call to check output quality — for example, verifying that a generated outline contains at least five sections. If the check fails, route back to the same AI Agent node with a revised prompt specifying the deficiency; if it passes, continue to the writing stage. This self‑correcting loop improves output consistency for production content pipelines. The chained-request pattern executes sequential AI model calls with intermediate processing, reducing API costs by 30–50% compared to monolithic single-prompt approaches by enabling selective model invocation — smaller, cheaper models handle classification and extraction stages while the full model is reserved for generation. For the complete three‑stage pipeline with evaluation, see the n8n OpenAI Prompt Chain Tutorial.

How do you add retry logic and handle rate limits for production OpenAI workflows in n8n?

OpenAI returns HTTP 429 Too Many Requests when you exceed rate limits. Production workflows must handle this gracefully with two complementary strategies. Built‑in retry: in the OpenAI Chat Model node’s Settings panel, enable Retry On Fail, set Max Retries to 3, and Wait Between Retries to 1 second. n8n retries with linear backoff for transient errors. Deliberate throttling: place a SplitInBatches node before the OpenAI node to process 5–10 items per batch, with a Wait node (1–5 seconds) between batches. [5] [10]

For high‑volume pipelines processing 500+ items, combine batching with an Error Trigger workflow: when a 429 or 5xx error occurs, the error workflow catches the failure, waits with exponential backoff (1 s, 2 s, 4 s, up to 60 s), then re‑enqueues the item via the Execute Workflow node. Track costs by logging token usage from the OpenAI node’s output metadata — the response includes completion_tokens, prompt_tokens, and total_tokens — multiply by your model’s per‑token price for exact cost per execution. Set a daily budget alert using a Schedule Trigger that sums logs and notifies Slack if the total exceeds a threshold. For the complete error handling architecture with exponential‑backoff retry loops, see the n8n Error Handling Nodes guide.

How do you generate and store DALL‑E images from text prompts using the OpenAI Image node?

The OpenAI node supports image generation via the DALL‑E family of models. Configure the Resource → Image → Generate Image operation, then select the model (dall‑e‑3 recommended for production), enter a text prompt (up to 4,000 characters for dall‑e‑3, 1,000 for dall‑e‑2), and configure output parameters: Quality (Standard or HD for dall‑e‑3), Resolution (1024×1024, 1792×1024, or 1024×1792 for dall‑e‑3; 1024×1024 only for dall‑e‑2), and Style (Natural or Vivid for dall‑e‑3). [11] [12]

For multi‑modal pipelines, combine GPT‑4o with DALL‑E 3 through the AI Agent’s tool connector: attach an HTTP Request Tool sub-node that calls the OpenAI Images API (https://api.openai.com/v1/images/generations), describe the tool as “Generate an image from a text prompt using DALL‑E 3”, and the AI Agent decides when to invoke it based on user requests. The community n8n-nodes-openai-image-generator node provides a dedicated interface for DALL‑E with parameters for size, quality, and style, plus support for image variations. Store generated images in Google Drive, Azure Blob Storage, or AWS S3 using the respective n8n nodes. For the complete agent‑driven image generation pipeline with Telegram delivery, see the n8n OpenAI Prompt Chain Tutorial.