OpenAI
GPT models — an alternative LLM backend for the AI nodes.
OpenAI has no typed nodes of its own: it connects as a stored credential that 9 generic nodes use — the node picks OpenAI from the connection you select.
Before you connect OpenAI
- Authentication
- API key
- Credential
- {"api_key"}
- Connection test
- Available: the stored credential can be checked against the provider when you add it. This is not a record that a test was run or passed.
- Prerequisites
- The provider adapter is chosen from the connection; `model` must belong to it.
- Anthropic and DeepSeek connections are refused: neither ships an embeddings API, and substituting another vendor's model would produce vectors that do not match the rest of the store.
- Only prompt_injection, topics and language call a model. A guard that runs pii / secrets / length needs no credential.
- Only op=summarize calls a model; get / append / clear need no credential at all. The provider adapter is chosen from the connection and `model` must belong to it.
Prerequisites are taken from the connection notes of the nodes listed below that accept your OpenAI credential.
What you can do with OpenAI
Operations of the generic nodes your OpenAI connection drives, grouped by what they do to your data.
Other operations 9
- AI Agent
- AI Classify
- AI Embed
- AI Extract
- AI Generate
- AI Guard
- and 3 more
Nodes that use your OpenAI connection
These generic nodes accept your OpenAI credential. Each chooses the provider from the connection you select, so the same node also works with the other providers shown.
- AI Agentai.agentTool-using LLM agent: model + memory + tools, with a step-by-step trace.
- AI Classifyai.classifyPick one label from a fixed list, with a confidence score.
- AI Embedai.embedTurn text into an embedding vector with an OpenAI or Google model.Also works withGoogle Gemini
- AI Extractai.extractPull structured fields out of text against a JSON Schema, with optional human review.
- AI Generateai.generateFill a {{var}} prompt template from the input and generate free text.
- AI Guardai.guardCheck text for PII, secrets, prompt injection, topics, length or language.
- LLM Callai.llmFree-form prompt to any configured LLM provider, optionally JSON-schema constrained.
- AI Memoryai.memoryRead, extend, clear or summarise an ai.agent conversation history.
- AI Transformai.transformSummarize, translate, paraphrase or clean text with a curated prompt.
A sample OpenAI workflow
- 1A core triggerStart the run from a chat, form, manual, schedule, webhook, workflow trigger — no OpenAI trigger is needed.
- 2AI Agentai.agentTool-using LLM agent: model + memory + tools, with a step-by-step trace.
An outline built from this provider’s node contracts, not a recorded run. Each step’s output fields are available to the next in the field picker.
OpenAI questions
- How does OCNORA authenticate with OpenAI?
- With a stored API key credential, shaped {"api_key"}. A connection test can check it when you add it.
- What can I automate with OpenAI?
- OpenAI has no typed nodes of its own. Your OpenAI connection is used by AI Agent (ai.agent), AI Classify (ai.classify), AI Embed (ai.embed), AI Extract (ai.extract), AI Generate (ai.generate), AI Guard (ai.guard) and 3 more.
- What if the OpenAI operation I need is not listed?
- Where OpenAI has an HTTP API, call it from the HTTP node with the same credential, or describe the step to the AI builder, which asks our team for a typed node when one is missing.
- Are there templates that use OpenAI?
- Not yet. Start from a blank workflow, or describe the workflow to Build with AI.
- How are OpenAI calls controlled?
- Every call and every run step is audited, runs count against the tenant’s token and USD budgets, and Workflow Health pre-flights the workflow when you save it.
More in AI & models
See your first OpenAI workflow run.
Book a technical demo. We add the connection, pick a template or describe the workflow to Build with AI, and Workflow Health checks it before it runs.