Skip to main content

Research & Insights

AI Without Screens: Deploying Large Language Models Over SMS Gateways

AI is expanding rapidly, but the hardware interface required to access it is narrowing — unless you rethink the channel

By Levi Dombrovsky••5 min read

The current delivery mechanism for AI is centralized around high-speed internet, browsers, and resource-heavy apps. The convergence of cutting-edge LLM backends with the world’s most resilient low-bandwidth protocol — SMS — can deliver a world-class research assistant to a $30 flip phone.

We are living in the golden age of Artificial Intelligence. Large Language Models (LLMs) are transforming how we write, code, research, and manage information. However, the current delivery mechanism for AI is heavily centralized around high-speed internet, modern web browsers, and resource-heavy smartphone applications.

This creates a massive technological paradox: AI is expanding rapidly, but the hardware interface required to access it is narrowing. For users who choose to live offline or use basic feature phones, the benefits of the AI revolution are completely out of reach.

At Connect2Kehilla, we are exploring a different frontier: the convergence of cutting-edge AI backend infrastructure with the world's most resilient, low-bandwidth protocol — SMS.

The Architecture of Text-Based AI

Deploying an LLM via an SMS gateway requires a complete rethink of AI interaction design:

  • Prompt engineering for 160 characters: Unlike a web interface where an AI can generate a 1,000-word essay, an SMS-based AI must be strictly programmed for extreme density and utility. It must summarize complex information, translate phrases, or answer technical queries within strict character limits.
  • Contextual filtering: For traditional communities, an SMS AI gateway acts as a safe, text-only proxy to the world's information. By implementing rigid, backend safety filters, developers can provide access to raw, educational data (e.g. medical information, historical facts, business math) while completely blocking access to inappropriate visual or text content.
  • Conversational state management: Because SMS is an asynchronous channel, the backend middleware must efficiently map phone numbers to specific vector-database sessions, allowing users to text a question, wait, and follow up hours later without losing context.

These are the same design constraints we explore in Screenless Systems: Building UX/UI for SMS Interfaces — taken one step further, with an LLM behind the gateway.

The Democratization of Intelligence

The future of technology is not about adding more pixels or screen real estate; it is about extending accessibility. By leveraging SMS as an interface for AI, we can deliver the power of a world-class research assistant to a $30 flip phone, bridging the digital divide without forcing users to sacrifice their mental health or community standards to a smartphone — closing the very gap described in Why the App-Only World Is Broken.

Connect2Kehilla is a free community SMS information accessibility program operated by Education on the Go Corp, a 501(c)(3) public charity (EIN 92-1172505). To collaborate or support the program, see /partners.