Meta introduces Edge LLMs for Quest to enable local AI companions

AI · 5 min read

Meta’s new Edge LLM family is tailored to run efficiently on Quest-class hardware with aggressive quantization and pruning, enabling features like voice-based scene descriptions, conversational NPCs in VR, and on-device summarization of user activity. These models are designed to balance latency, battery, and privacy by keeping inference local unless cloud fallback is necessary.

The company also released an SDK to integrate Edge LLMs with Unity and Meta’s in-house Spatial Authoring tools, allowing developers to bind local conversational agents to environment state and user intent signals. Meta emphasizes opt-in privacy controls and the ability to export minimal telemetry for model improvement.

Game and social app developers see this as an opportunity to create richer single-player or small-group experiences that don’t rely on continuous network connectivity. Meta’s documentation provides sample interactions and performance budgets to guide studios building for constrained hardware.