Yelp’s AI Assistant Combines Native Controls With Server-Driven Chat

Originally reported on November 15, 2026.

Yelp’s AI assistant uses a hybrid interface: native mobile controls handle navigation and message input, while Yelp’s server-driven CHAOS framework renders the conversation area. The approach lets Yelp change parts of the chat interface without releasing a new mobile app version.

The assistant launched on iOS in 2024 to help users submit requests to local service professionals. Yelp later expanded it to answer business-related questions and support bookings, appointments and food orders.

CHAOS—short for Content Hosting Architecture with Optimization Strategies—supports web, iOS and Android. Yelp service technology lead Daniel Andrade Groppe said the framework lacked capabilities the assistant needed, including text input fields and network requests. Rather than delay delivery to extend CHAOS, the team kept navigation and message input native.

Yelp’s AI Assistant Combines Native Controls With Server-Driven Chat

The implementation separates interface configuration from conversation data. When a chat starts, the client fetches a CHAOS configuration and initializes a message-event dataset. Later events update the existing view without fetching the configuration again. Events can represent persistent user and assistant messages, temporary typing indicators and quick replies, or links that open another page. The backend can also send interface actions, such as disabling chat input after a service request is completed.

On the client, submitted messages and typing indicators appear while input is disabled and a request is sent. After a response arrives, the client displays the reply and quick replies, then re-enables input. Recoverable errors trigger an automatic retry; unrecoverable failures disable input and ask users to try again later.

Yelp engineers also noted trade-offs: developers need to understand both the assistant and CHAOS, and shared logic can still behave differently across platforms. Yelp said the architecture helped it reuse the Android assistant for a Biz Ask Anything pilot and laid groundwork for a newer assistant that connects business information with actions users can take.

Source: infoq.cn (originally published 2026-11-15 15:00 UTC+8)