Salesforce · Filed Feb 14, 2025 · Published Aug 6, 2026 · verified — real USPTO data

Salesforce Patents a Way to Run AI Inside App Screens Without an Internet Connection

Most AI features in business software require a live connection to a distant server. Salesforce is filing a patent that would let AI run directly inside individual panels of its apps, no internet required.

Salesforce Patent: AI Models Running Offline in UI Components — figure from US 2026/0228022 A1
Figure from the official USPTO publication.
See all 5 drawings from this filing ↓
Publication number US 2026/0228022 A1
Applicant Salesforce, Inc.
Filing date Feb 14, 2025
Publication date Aug 6, 2026
Inventors Siva Aditya Yedida, Santosh Kumar Chikoti, Subhashini Puram, Venkat Ramana Reddy Parine
CPC classification 706/11
Grant likelihood Medium
Examiner CENTRAL, DOCKET (Art Unit OPAP)
Status Docketed New Case - Ready for Examination (Mar 27, 2025)
Document 20 claims

How Salesforce's offline AI inside app panels works

Imagine you're a sales rep filling out a customer record on your laptop, but you're on a plane with no Wi-Fi. Normally, any AI-powered suggestion (like auto-filling a field or flagging a risk) would just be unavailable. Salesforce's patent describes a way to bake small AI models directly into the parts of the app you're looking at, so they can still work even when your device is offline.

Think of the interface as made up of building blocks, like panels, forms, and widgets. This patent describes a system where a lightweight AI model gets downloaded ahead of time, stored on your device, and attached to one of those building blocks. When you open that panel, the AI runs locally on your machine using data already stored there.

The key phrase is lightweight AI model. These aren't the giant models that power ChatGPT. They're smaller, task-specific versions designed to run comfortably on a laptop or phone without needing a data center behind them.

How the AI model gets embedded in a UI component

The patent describes a four-step process:

  • Fetch: The app pulls one or more pre-built lightweight AI models from a central repository (a library of ready-made models).
  • Store locally: Those models get saved in the device's local memory, tied to the specific application.
  • Embed in UI components: Using a scripting library (a set of code tools that lets developers attach functionality to interface elements), the models are wired directly into configurable UI components, the individual panels, cards, or forms that make up the app's screen.
  • Execute on-device: When needed, the model runs right there on the device, using local data, and produces an output without ever phoning home to a server.

The patent specifically calls out an offline mode, meaning the entire AI inference loop (input data in, prediction out) can complete without any network connectivity.

The "configurable" part is important: Salesforce's platform (like Salesforce CRM or its low-code tools) lets customers and developers build their own app layouts. This patent would let those custom layouts include AI capabilities that travel with the component itself, not with the server.

We find one patent like this every day. Get the best of each week in your inbox, free →

What this means for Salesforce users on the go

For enterprise software, connectivity gaps are a real business problem. Field workers, traveling sales teams, and anyone in a low-signal environment often lose access to AI features exactly when they need them most. A patent like this points toward AI assistance that works wherever you are, not just where your IT team has configured a good connection.

For Salesforce specifically, this fits into a broader push to embed AI (the company calls its AI layer "Agentforce") more deeply into everyday workflows. Making those AI features device-resident rather than cloud-dependent would be a meaningful reliability upgrade for customers who pay for the platform and expect consistent behavior.

Editorial take

This is a sensible, practical patent rather than a flashy one. The engineering challenge it addresses (keeping AI features functional when connectivity drops) is a genuine pain point for enterprise field software. It won't generate headlines, but if it ships in Salesforce's low-code builder tools, a lot of field sales and service teams will benefit.

There are more where this came from

We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.

The drawings

5 drawing sheets from US 2026/0228022 A1 · click any drawing to enlarge

Patent filing page

Source. Full patent text and figures from the official USPTO publication PDF.

Editorial commentary on a publicly published patent application. Not legal advice.