Edge AI enables fully autonomous robots and local AI agents on compact devices like Raspberry Pi, offering zero cloud dependencies, ultra-low latency, and total data privacy. Google AI Edge’s LiteRT, a production-proven on-device inference runtime, simplifies deploying classical ML models and state-of-the-art LLMs across platforms, optimizing execution and memory usage on both CPU and GPU. Paired with Gemma, Google’s family of lightweight open models, LiteRT powers the Reachy Mini robot on a Raspberry Pi 5 to perceive and react to its environment entirely locally in real time. Gemma models support autonomous agents, smart cameras, and social robotics, with several options: Gemma 3 270M, a hyper-efficient base model for task-specific fine-tuning; EmbeddingGemma 300M, a text embedding model for RAG, semantic search, and classification; Gemma 3 1B, a multilingual text-only model for summarization and content creation; Gemma 4 E2B, tailored for tight edge environments with memory-mapped per-layer embeddings for continuous monitoring and speech processing; and Gemma 4 E4B, balancing performance and size for complex multi-step planning. Through LiteRT-LM, a specialized orchestration layer on top of LiteRT, developers can seamlessly deploy these models.