Google Gemma releases gemma-skills to accelerate agentic workflows with multi-token prediction

Google GemmaGoogle Gemma

· Updated

Google launched the first iteration of gemma-skills, an open-source library of reusable capabilities for AI agents. By standardizing how agents select model sizes and use performance optimizations like MTP, Google is making it easier to build efficient, autonomous workflows on top of the Gemma ecosystem.

Google Gemma released the first iteration of gemma-skills, an open-source library of reusable capability packages for building with the Gemma model family. The initial release features the gemma-dev skill, providing agents with technical resources and support for Multi-Token Prediction (MTP) (predicting multiple tokens at once to increase speed).
Initial skill
gemma-dev
Optimization support
Multi-Token Prediction (MTP)
License
Apache-2.0
Installation tools
Vercel skills CLI, Context7 skills CLI
Programming languages
Python, JavaScript

This release extends the Agent Skills library pattern to Google's open-weight models, addressing the knowledge layer bottleneck in agentic engineering. By automating model size selection, the library helps agents navigate the Gemma 4 lineup—ensuring they use the most cost-effective model for a task without manual developer intervention.

Integrate these skills into your workflow using the Vercel or Context7 command-line interfaces. The library is available on GitHub under an Apache-2.0 license, allowing you to add the gemma-dev skill to specific projects. This enables agents to autonomously locate documentation and optimize their own performance using MTP during multi-step tasks.

View the full update on github.com
Google Gemma
Google Gemma
@googlegemma
X

The first iteration of gemma-skills is officially out! 🛠️ It enables agents to build with Gemma, including using MTP to improve speed, choosing the right model size for your use case, and locating up-to-date resources. https://t.co/VZHhvwH0Zw

96retweets904likes
View on X

Still wondering? A few quick answers below.

Google gemma-skills is an open-source repository of agent skills designed for the Gemma model family. These skills are reusable capability packages that allow AI agents to perform specific tasks, such as retrieving technical documentation or optimizing model performance, without requiring the developer to write custom logic for every interaction.

You can install gemma-skills using standardized command-line tools. The repository supports the Vercel skills CLI using the npx skills add command and the Context7 skills CLI via npx ctx7 skills install. These tools allow you to browse available skills interactively or install specific ones like gemma-dev globally.

The gemma-dev skill is the first capability released in the gemma-skills library. It is specifically designed for building applications with Gemma or handling general knowledge inquiries about the models. It helps agents locate up-to-date resources and documentation to ensure they are operating with the most current technical context.

Yes, the library includes support for Multi-Token Prediction (MTP). This optimization allows Gemma models to predict multiple tokens simultaneously during inference, which can significantly improve the generation speed of AI agents. The library also includes logic to help agents select the most appropriate model size for their current task.

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

Google

See all AI news & updates from Google →

Share this update