AI workflow assistant (beta)

The AI assistant generates node graphs from text descriptions. Describe what you want to do and it builds the pipeline for you.

Open it from View > AI Assistant in the menu bar.

Setup

1. Select a Provider from the dropdown
2. Paste your API Key (not needed for local providers such as Ollama)
3. Click the refresh button to load available models
4. Pick a model

Providers

Provider	API Key	Notes
Ollama	not needed	Local. Install from ollama.com, then pull a model (ollama pull gemma3:12b)
Ollama Cloud	needed	Create an account at ollama.com, then get a key at ollama.com/settings/keys
OpenAI	needed	Get a key at platform.openai.com/api-keys
Claude	needed	Get a key at console.anthropic.com
Groq	needed	Free tier available at console.groq.com
Gemini	needed	Free tier available at aistudio.google.com
RunPod	needed	Serverless vLLM endpoint. Enter your Endpoint ID separately

API keys are stored locally and never sent anywhere except the selected provider.

Usage

1. Type a description of the workflow you want, e.g.:

   Load a CSV, filter out data with area > 100, then make a bar plot

2. Include current workflow as context (optional): check this to let the AI see and modify your existing pipeline instead of building from scratch

3. Verbose node descriptions (optional): sends full docstrings to the model for better context. Larger prompt but more accurate results.

4. Click Generate Workflow. This takes 10-60 seconds depending on the model.

5. Review the generated JSON in the preview panel

6. Click Load into Canvas to append the nodes, or Replace Canvas to start fresh

Tips

• Simpler prompts work better. Break complex pipelines into steps.
• If the model gets a port name wrong (you'll see warnings), try regenerating or fix the connection manually.