Meta Launches TikTok-Style App Pocket for AI Mini Games, Unveils Coding Assistant Muse Code

Meta has officially launched a new social application called Pocket in the US market. The app adopts a TikTok-style interaction format, allowing users to create, share, play, and remix AI-generated mini games and micro-apps through natural language prompts (vibe-code), which Meta calls "gizmos." Pocket's underlying technology is primarily based on a standalone app called Gizmo. In March 2026, Meta purchased a non-exclusive technology license from developer Atma Sciences and absorbed its team; with Pocket's launch, the original Gizmo app has begun to be phased out. On the account front, users must have a Meta account to create a Pocket profile, and the two are deeply linked, meaning users cannot delete their Pocket profile without deactivating their entire Meta account.

The barrier to creation on Pocket is extremely low. Users simply open the app, tap the create button at the bottom of the screen, and enter a description of what they want to build; the more specific the prompt, the better the generated result. After the AI completes game generation, creators can choose whether to allow other users to play or remix their creations. According to Meta's official description, players can experience these mini apps by tapping, swiping, or dragging on the screen, or by tilting or shaking their smart devices. The app also supports custom music and sound effects, real-time camera capture, and voice input through the microphone. Meta has explicitly stated that it will collect user interaction data with AI and mini apps on the platform to further train and improve its artificial intelligence technology.

Beyond its consumer entertainment platform, Meta has also taken new steps in the professional generative AI field. Meta recently released its own AI coding assistant, Muse Code, powered by its new dedicated code model, Muse Spark 1.2. However, the precondition for enjoying this ultra-low fee is that developers must explicitly authorize Meta to access their proprietary codebases, prompts, and output content, which Meta will use as training data for its future AI models.