OpenAI Integration
Connect directly to OpenAI's API for advanced use cases beyond what TinyGPT provides: custom model parameters, DALL-E image generation, Whisper transcription, and fine-tuned models.
When to use OpenAI vs TinyGPT
| Feature | TinyGPT | OpenAI Integration |
|---|---|---|
| Setup | No API key needed | Requires OpenAI API key |
| Models | TinyGPT default | GPT-4o, GPT-4, GPT-3.5, DALL-E, Whisper |
| Configuration | Simplified (System Prompt + Query + Schema) | Full API parameters |
| Best for | Standard classification, extraction, writing | Advanced prompting, image generation, audio |
| Credits | TinyCommand credits | OpenAI API billing (separate) |
Setup
- Get an API key from platform.openai.com
- App Authorizations → OpenAI → paste your API key
Available actions
| Action | What it does |
|---|---|
| Chat Completion | Send a message to GPT-4/3.5 and get a response |
| Image Generation | Generate images with DALL-E |
| Audio Transcription | Transcribe audio with Whisper |
| Embeddings | Generate text embeddings |
Common patterns
Custom AI with full control
Trigger → HTTP Request (GET context data) → OpenAI Chat Completion (custom system prompt + context) →
Process response → Send Email
Image generation pipeline
Trigger (product description) → OpenAI DALL-E (generate product image) →
Upload to storage → Update product record
Tip
Use TinyGPT for most AI tasks; it's simpler and uses TinyCommand credits. Use the direct OpenAI integration only when you need specific models, image generation, audio processing, or fine-tuned models.
Error handling
If this node fails during execution:
- The execution stops at this node (unless Continue on error is enabled)
- The error appears in the Execution History with the full error message
- Downstream nodes don't execute
To handle errors gracefully:
- Enable Continue on error in the node settings
- Add an If-Else node after to check if the operation succeeded
- Route failures to a notification (Send Email, Send Slack)
Debugging
- Check the Execution History for the failed run
- Click the failed node to see its input and error
- Verify the input data is correct (check previous node outputs)
- Test with sample data using the Test tab