AI Agent Workflows includes direct chat actions for OpenAI and Google Gemini, plus OpenAI-compatible integrations for DeepSeek, Groq, and OpenRouter. The workflow editor can load available text-generation models after the provider has a valid API key. AI Agent nodes can also use a connected chat model, optional memory, tools, a fallback model, and a structured output parser.
Screenshot to add: AI provider actions and AI Agent attachments on the workflow canvas
Choose a provider
- OpenAI: use the OpenAI chat action or attach an OpenAI chat model to an AI Agent.
- Google Gemini: use the Gemini generate-content action or attach a Gemini chat model.
- DeepSeek, Groq, and OpenRouter: use their supported chat actions or attach them as chat models.
Configure a provider in the workflow editor
- Open a workflow and add the required AI action or AI Agent node.
- Select the node and provide the provider API key in the node configuration when prompted.
- After the key is accepted, choose a text-generation model from the available model list.
- Configure the prompt and map only the workflow data the model needs.
- Test the node and inspect its output before connecting it to actions that change or send data.
Add capabilities to an AI Agent
Select the AI Agent node to open its Parameters and Settings tabs. Connect a Chat Model to run the agent. Depending on the task, you can also attach memory, tools, a fallback chat model, and a structured output parser. These attachments are not all required; the editor validates the connections that are required for the selected configuration.
Before you test
- Use an API key issued by the provider you selected.
- Model choices are retrieved from the provider, so the available list can change and may be empty if the key or provider request fails.
- Do not treat a saved key as proof that a complete workflow has been tested. Run a node or workflow test and review the result.
- Only send external services the trigger data needed for the task.