Bring state-of-the-art agentic skills to the edge with Gemma 4

2026-04-02 · Source: Google Developers Blog - AI · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Internet of Things (IoT) & Connected Devices · Depth: Intermediate, short

Summary

Google DeepMind launched Gemma 4 on April 2, 2026, a family of open models under the Apache 2.0 license, designed for on-device AI development. Gemma 4 enables agentic and autonomous AI use cases directly on hardware, supporting multi-step planning, offline code generation, audio-visual processing, and over 140 languages without specialized fine-tuning. Developers can access Gemma 4 via Android's built-in AICore Developer Preview or Google AI Edge. The Google AI Edge Gallery, available on iOS and Android, introduces "Agent Skills" for on-device multi-step workflows, allowing Gemma 4 to augment knowledge bases, produce interactive content like summaries and visualizations, expand core capabilities through integration with other models, and create comprehensive end-to-end conversational experiences. For broader device deployment, LiteRT-LM offers enhanced performance with features like minimal memory footprint (<1.5GB for Gemma 4 E2B), constrained decoding for structured outputs, and dynamic context handling up to 128K tokens. LiteRT-LM also brings Gemma 4 to IoT and edge devices like Raspberry Pi 5 and Qualcomm Dragonwing IQ8, with a new Python package and CLI tool for Linux, macOS, and Raspberry Pi.

Key takeaway

For AI Architects and CTOs evaluating on-device AI solutions, Gemma 4 offers a robust, open-source foundation for developing agentic applications across mobile, desktop, and IoT. Your teams can leverage its multi-step planning, multi-language support, and efficient deployment via LiteRT-LM to build autonomous, privacy-preserving experiences. Explore the Google AI Edge Gallery and LiteRT-LM documentation to understand its full capabilities and performance metrics on target hardware.

Key insights

Gemma 4 enables powerful, multi-modal, agentic AI experiences directly on diverse edge devices.

Principles

• On-device AI enhances privacy and reduces latency.
• Agentic AI extends LLM capabilities beyond chatbots.

Method

Google AI Edge Gallery provides a platform for building and experimenting with on-device agent skills, while LiteRT-LM optimizes Gemma 4 deployment across various hardware with memory efficiency and structured outputs.

In practice

• Use Agent Skills to query external knowledge bases.
• Integrate Gemma 4 with other models for rich content.
• Deploy Gemma 4 E2B on devices with <1.5GB memory.

Topics

• Gemma 4
• On-device AI
• Agentic AI
• Google AI Edge
• LiteRT-LM

Code references

• google-ai-edge/gallery

Best for: AI Architect, CTO, VP of Engineering/Data, AI Engineer, Machine Learning Engineer, Robotics Engineer

Related on AIssential

• Google Gemma 4 is here! #ai #gemini #google — Gemma 4 is a multimodal open model optimized for local execution on mobile devices, offering efficient on-device AI capabilities.
• Gemma 4 Runs on Your Phone — Agentic AI models are becoming smaller, more efficient, and integrated into developer workflows.
• The Sequence AI of the Week #839: Gemma 4 and the Compression of Intelligence — Gemma 4 transforms frontier AI capabilities into practical, deployable infrastructure for diverse applications.
• Google just dropped Gemma 4... (WOAH) — Gemma 4 offers high intelligence per parameter, enabling powerful AI on consumer and edge devices.
• DeepMind’s New AI: A Gift To Humanity — Gemma 4 offers high performance and broad utility in a compact, open-source package.
• Easy Agentic Tool Calling with Gemma 4 — True LLM agency requires interacting with the local system through carefully sandboxed tools, enabling observation and deterministic computation beyond remote API calls.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Google Developers Blog - AI.