Each developer’s AI workflow and wishes are distinctive, and it is vital to have the ability to select how AI helps your growth. In January, we launched the power to decide on any native or distant AI mannequin to energy AI performance in Android Studio, and as we speak, we’re asserting the supply of Gemma 4 for AI coding help in Android Studio. This new native mannequin educated on Android growth gives the most effective of each worlds: the privateness and cost-efficiency of on-device processing alongside state-of-the-art reasoning and tool-calling capabilities.
AI help, domestically delivered
By working domestically in your machine, Gemma 4 provides you AI code help that does not require an web connection or an API key for its core operations. Key advantages embrace:
- Privateness and safety: Your code stays in your machine. Gemma 4 processes all Agent Mode requests domestically, making it a really perfect alternative for builders working with knowledge privateness necessities or in safe company environments.
- Price effectivity: Run complicated agentic workflows with out worrying about hitting quotas. Gemma 4 is optimized to run effectively on fashionable growth {hardware}, using native GPU and RAM to offer snappy, responsive help.
- Offline availability: Use the agent to write down code even while you don’t have an web connection.
- State-of-the-art reasoning: Gemma 4 delivers best-in-class reasoning, able to complicated multi-step coding duties in Agent Mode.
Highly effective agentic coding
Gemma 4 was educated for Android growth with agentic software calling capabilities. When you choose Gemma 4 as your native mannequin, you’ll be able to leverage Agent Mode for a wide range of growth use instances, reminiscent of:
- Designing new options: Builders can ask the agent to construct a brand new function or a whole app with instructions like “construct a calculator app” and the agent is not going to solely generate the UI code however will use Android greatest practices like writing in Kotlin and utilizing Jetpack Compose.
- Refactoring: You can provide high-level instructions reminiscent of “Extract all hardcoded strings and migrate them to strings.xml.” The agent will scan your codebase, establish situations requiring modifications, and apply the edits throughout a number of recordsdata concurrently.
- Bug fixing and construct decision: If a undertaking fails to construct or has persistent lint errors, you’ll be able to immediate the agent to “Construct my undertaking and repair any errors.” The agent will navigate to the offending code and iteratively apply fixes till the construct is profitable.
Really useful {hardware} necessities
The 26B MoE is advisable for Android app builders utilizing a machine with the minimal {hardware} necessities. Complete RAM wanted consists of each Android Studio and Gemma.
| Mannequin | Complete RAM wanted | Storage wanted |
|---|---|---|
| Gemma E2B | 8GB | 2 GB |
| Gemma E4B | 12 GB | 4 GB |
| Gemma 26B MoE | 24 GB | 17 GB |
Get began
To get began, guarantee you’ve gotten the most recent model of Android Studio put in.
- Set up an LLM supplier, reminiscent of LM Studio or Ollama, in your native laptop.
- In Settings > Instruments > AI > Mannequin Suppliers add your LM Studio or Ollama occasion.
- Obtain the Gemma 4 mannequin from Ollama or LM Studio. Seek advice from {hardware} necessities for mannequin measurement choice.
- In Agent Mode, choose Gemma 4 as your energetic mannequin.
For an in depth walkthrough on configuration, take a look at the official documentation on the best way to use an area mannequin.
We’re excited to see how Gemma 4 allows extra non-public, safe, and highly effective growth workflows. As all the time, your suggestions is crucial as we proceed to refine the AI expertise in Android Studio. When you discover a bug or subject, please file a problem. Additionally you could be a part of our vibrant Android developer group on LinkedIn, YouTube, or X. Pleased coding!