Changelog
AI, Gemini and agent features in Android Studio, summarised from the official release notes. Newest first; every item links to its source.
Android Studio Rabbit 2 Canary
- Bring Your Own Agent (BYOA): Integrate your preferred coding agents
Building on Android Studio's support for custom AI models, Android Studio Rabbit 2 Canary 2 introduces Bring Your Own Agent (BYOA). Using BYOA, you can connect your preferred AI coding agents—including Claude Agent, OpenAI Codex, and Google Antigravity—directly into Android Studio using the open Agent Client Protocol (ACP).
- Model Assignment
The Model Assignment feature lets you pick different AI models for different types of tasks in Android Studio. For example, you can assign a large, pro-tier model for complex reasoning tasks in Agent Mode, and a faster, lightweight model for simpler, latency-sensitive tasks like commit message generation or next edit prediction.
- Migrate iOS, Flutter, and React Native apps to Android
You can now migrate existing iOS, Flutter, and React Native projects to native Android using automated migration tooling in Android Studio. From the New Project Wizard, you can call the assistant to convert the selected project to a Kotlin and Jetpack Compose-based Android app.
- Agentic fixes with LeakCanary
Starting in Rabbit 2, Android Studio directly integrates LeakCanary with the AI agent to help you diagnose and resolve memory leaks without leaving the IDE. This integration lets LeakCanary provide more meaningful narrative explanations of the underlying causes of memory leaks and suggest specific code changes to correct them.
- Discover, install, and manage Android skills in Android Studio
In Android Studio Rabbit 2, it's easier to discover, install, and manage Android skills directly in the IDE, giving you granular control over the specialized workflows and procedural guidance available to the agent.
Android Studio Rabbit 1
- Android Studio is part of the Gemini Enterprise bundle
Android Studio is now included in the Gemini Enterprise subscription bundle, which enables enterprise development teams to deploy Android Studio using models hosted directly in your organization's Google Cloud infrastructure. Every session runs under Google Cloud's enterprise security controls, data residency guarantees, and the Google Cloud Terms of Service.
Android Studio Quail 4
- Build full-stack apps with Firebase in Agent Mode
Firebase services like Authentication and Cloud Firestore databases can be enabled and configured directly in Agent Mode in Android Studio using Firebase agent skills. The agent can help you complete Firebase integration and configure backend services.
- Gemma 4 integration
You can now run the powerful Gemma 4 model directly within Android Studio for AI code assistance without relying on a third-party provider to host it. Powered natively by an integrated runtime, this local execution gives you full control to design new features, refactor code, and debug issues—all completely on-device.
- Recomposition state reads in the Layout Inspector
We've made it easier to diagnose high recomposition counts by adding Recomposition state reads to the Layout Inspector. Available in Panda 3 canary, this feature helps you identify the state variables that triggered a recomposition by providing a detailed list of state reads performed during that cycle.
Android Studio Quail 3
- Improvements to Planning Mode
Planning Mode facilitates a multi-stage reasoning process, giving the agent more space to evaluate its own proposed logic for potential issues before presenting it to you. This is especially useful for complex and long-running tasks which demand a high degree of precision.
- MCP Marketplace
Starting in Quail 3, we're making it easier for you to find and add MCP servers. Navigate to Settings > Tools > AI > MCP Servers to browse for a server and install it.
- Merge Conflicts action
Quail 3 introduces a new Merge Conflicts with Agent button into the Commit tool window that instructs the agent to automatically merge conflicts. When there are one or more conflicting files in the repositories, you can click this button to launch a new agent session targeting those files.
Android Studio Quail 2
- Parallel conversations
You can run multiple conversations in Agent Mode in parallel, enabling you to multitask more effectively inside the IDE. For example, you can kick off a long-running task like generating tests in one conversation, use a second conversation to plan a new feature, and use a third conversation to draft documentation—all at the same time.
- LeakCanary in Android Studio Profiler
Android Studio Quail 2 includes a LeakCanary integration directly in the Android Studio Profiler as a dedicated task.
Android Studio Quail 1
- Suggested fixes for crashes with Agent integration in AQI
The App Quality Insights tool window is now integrated with the AI agent to analyze crash data along with your source code to provide detailed explanations and suggest potential fixes. After selecting a crash in the App Quality Insights tool window, navigate to the Insights tab and click See more to see a detailed explanation of the crash.
Android Studio Panda 4
- Generate unit tests with Gemini
Gemini in Android Studio can generate comprehensive, compilable unit tests for your Kotlin and Java code. Gemini analyzes your source code to identify constructor dependencies, business logic branches, and edge cases, and then automatically creates a complete test class.
- Next Edit Prediction (NEP)
Android Studio Panda 4 introduces Next Edit Prediction (NEP), a major evolution of the editor's AI capabilities. While traditional AI Code Completion focuses on suggesting code at your current cursor position, NEP is designed for "away from cursor" updates.
- Ask Mode
In Android Studio Panda 4, we are removing the Ask tab and replacing it with a dedicated conversation mode, allowing you to ask the agent questions and get answers without prompting it to execute tasks.
- Planning Mode
The new Planning Mode prompts the agent to come up with a detailed project plan before it begins executing tasks. Instead of a single pass where the model directly predicts the next token of code, Planning Mode facilitates a multi-stage reasoning process, giving the agent additional space to evaluate its own proposed logic before presenting it to you.
- Agent Web Search
While Android Studio's agent already leverages the Android Knowledge Base for official documentation, modern Android development relies on a vast ecosystem of external libraries. Agent Web Search expands Gemini's reach, allowing it to query Google directly to fetch current reference material from across the web.
Android Studio Panda 3
- Manage permissions in Agent Mode
You can now manage specific permissions for the agent, giving you granular control over your workspace. Permissions let you control whether the agent can perform certain actions, including the following:
- Skills for Agent Mode
Starting in Android Studio Panda 3, you can use skills to enhance Agent Mode's capabilities with specialized expertise and custom workflows. Learn more at Extend Agent Mode with skills.