We've upgraded the LLM Actions you already use, generate_text_action and generate_structured_value_action, with a boost in performance and a batch of powerful new capabilities. Best of all, it's a drop-in upgrade. Everything you've already built keeps working exactly as before, with nothing you need to change.
⚡ Better performance, no work required
Your existing Compound Actions and Conversational Processes now benefit from improved model quality and efficiency automatically, especially on reasoning-heavy tasks. Same inputs, same outputs, better results.
✨ What's new
- 📄 File input
- Pass documents straight into the model, no preprocessing required. Supports PDFs, spreadsheets (Excel/CSV/TSV), rich documents (Word/RTF/ODT), presentations (PowerPoint/Keynote), and a wide range of text and code formats.
- 🎞️ Images and files together
- Combine images and files in a single request, and pass multiple of either as a list, from a single List[File] slot or mapped from separate File slots:
- file: [data.contract, data.addendum]
- Combine images and files in a single request, and pass multiple of either as a list, from a single List[File] slot or mapped from separate File slots:
- 🌐 Web search
- Turn on enable_web_search to let the model pull live information from the web into its context instead of relying only on its training data.
- Want to keep it focused?
- Optionally scope results with web_search_allowed_domains or web_search_blocked_domains (up to 100 domains each). Available on all models except gpt-5 with minimal reasoning.
💡 Tip
For image and file inputs, we recommend using gpt-5 or newer for the best results.
🛠️ What you can build
- OCR and document Q&A
- Comparing multiple files (for example, contract vs. addendum)
- Structured extraction from receipts, invoices, and forms
- Answering questions that need up-to-date, real-world information
🛡️ A note on safe use
Web search and file input let the model read content from outside sources. When designing flows that mix sensitive data, external content, and outbound actions, follow the security guidance in the docs to stay protected against prompt-injection risks.
📖 Full details, examples, and the model reference: https://docs.moveworks.com/agent-studio/actions/llm-actions
