Yelp has rolled out a hybrid user interface architecture for its AI Assistant that combines native mobile controls with server-driven conversation rendering, allowing the company to push interface updates across platforms without requiring new app releases. The design, detailed in a company blog post published this month, uses Yelp's CHAOS framework—Content Hosting Architecture with Optimization Strategies—to handle conversation content while navigation and message input remain native. The Assistant first appeared on iOS in 2024 to help users generate project requests for local service professionals, and has since grown to field questions about businesses and enable reservations, appointments, and food orders across web, iOS, and Android.
The implementation splits view configuration from conversation data, according to Yelp's technical documentation. When a chat session begins, the client retrieves the CHAOS configuration and sets up a dataset for message events; later events refresh the existing view without pulling the configuration again. Events cover persistent user and bot messages, temporary typing indicators and quick replies, and links that open destinations. The backend can dispatch actions to manage the interface—for instance, once a project request finishes, it returns a message, a link to view matches, and an action that locks chat input. The client shows submitted messages and a typing indicator, disables input, and fires requests to the backend; on response, it displays the reply and quick replies and unlocks input. Recoverable errors kick off retries, while non-recoverable failures disable input and tell users to try again later.
Daniel Andrade Groppe, Services Tech Lead at Yelp, noted that "our existing CHAOS framework lacked some of the capabilities the Assistant required, including text input fields and network requests," and explained that extending the framework would have postponed delivery, prompting the team to adopt a hybrid approach with native navigation and message input paired with CHAOS rendering for the conversation area. Yelp engineers acknowledge the architecture introduces coordination challenges—engineers needed time to learn both the Assistant and CHAOS, and shared code can still behave differently across platforms. The separation of view configuration from conversation data enables independent feature work, the report states, and the architecture allowed the team to reuse the Android Assistant to launch the Biz Ask Anything pilot.
Server-driven UI architectures let backend systems define what clients display and how they respond to interactions, so teams can change interfaces without shipping new mobile app versions, according to the report. Airbnb has documented a similar strategy through its Ghost Platform, which uses server-defined UI across web and mobile clients; both approaches let teams update supported presentation without requiring every change to ship in a new client release. Yelp's version applies this model to a conversational interface while keeping native controls and client-side handling for message state and interactions. The event-driven design decouples rendering from conversation logic, meaning the backend can evolve independently of the view layer—when the server wants to disable input after a request completes, it simply sends an action, and the client responds without needing to understand the business rule behind it.
The report states this work laid the groundwork for Yelp's next-generation Assistant, which now connects business information with actions users can take. The hybrid model supports faster iteration on conversational features while preserving platform-specific controls, positioning the Assistant to expand its scope across Yelp's service marketplace. Companies building conversational interfaces face a classic trade-off between control and velocity—pure native development offers polish but slows rollout, while full server rendering ships fast but can feel generic. Yelp's hybrid bet suggests the future may belong to teams willing to split the difference, keeping the seams invisible to users while gaining the flexibility to experiment behind them.

