How to Use Android Studio Rabbit 2: Bring Your Own AI Coding Agent (BYOA)
CodexFuse Team · 2026-11-03T13:54:02.650Z · 9 min read min read
What is Android Studio Rabbit 2's Bring Your Own Agent feature?
Bring Your Own Agent (BYOA) is the Android Studio Rabbit 2 feature that lets you run your own AI coding agent — Claude Agent, Codex, Junie, Google Antigravity, or any of 40-plus ACP-compatible agents — directly inside the IDE instead of using Google's built-in assistant only. Learning how to use Android Studio Rabbit 2 for BYOA means installing the stable Rabbit 2 release, opening Settings > Tools > AI > Agents, installing the agent you already pay for, and signing in once — the agent then works with the IDE's knowledge of your project, including the Emulator, build system, and Logcat.
Google moved BYOA to the stable channel in Android Studio Rabbit 2 on October 28, 2026, announced at Android Dev Summit. The key shift: the IDE now speaks the open Agent Client Protocol (ACP), so one integration covers every agent that speaks the protocol rather than requiring Google to build a bespoke connector for each. If Claude is the agent you bring, our Claude Code beginner guide covers the agent itself — this guide covers the Android Studio side.
What do you need before you start?
You need Android Studio Rabbit 2 on the stable channel and an account or API plan with at least one AI coding agent. That second requirement is the whole point of "bring your own": sign-in is per agent, and Google's setup step is simply to "authenticate," with each agent's own requirements applying. If you are starting from scratch, pick the agent you already use and pay for — BYOA connects it, it never substitutes for having the account.
How do you install your own agent inside Android Studio Rabbit 2?
The whole setup is a marketplace flow that takes a few minutes. Google's own instructions are one line: "open Settings > Tools > AI > Agents to browse and install agents from the marketplace."
- Update to Android Studio Rabbit 2 (stable channel) and open Settings > Tools > AI > Agents.
- Browse the agent marketplace, pick the agent you already pay for (Claude Agent, Codex, Junie, or Antigravity are the named examples), and install it.
- Authenticate the agent — sign in with your existing agent account or API key.
- Open your project and give it one real first task, like a small Compose UI change, so you can watch it drive the build and Logcat.
- Install a second agent later if you want — the picker supports more than one, and you switch between them as you work.
What can your agent actually do once it is inside the IDE?
Your agent gets everything that makes Android Studio Android Studio: a deep understanding of your project structure and build, symbol resolution, Compose Preview rendering, the Emulator, and Logcat. In practice that means the agent can run your app on a virtual device, read the crash from Logcat, and go fix the code — in one loop, without you copy-pasting anything between a terminal and the IDE.
Antigravity users get a bonus worth knowing: BYOA also gives you Google Antigravity directly in Android Studio — log in and you get access to the latest Gemini models as soon as they launch, plus new Antigravity capabilities as they roll out.
Should you use BYOA in the IDE or the Android CLI plugin?
Two different things shipped on October 28, 2026, and they solve different problems. Bring Your Own Agent puts the agent inside Android Studio; the Android CLI plugin puts the official Android toolchain inside your agent's terminal session, with no IDE open. Here is the split, built from Google's own two announcements:
| Question | BYOA in the IDE | Android CLI plugin |
|---|---|---|
| Needs Android Studio open? | Yes | No |
| Agents available | 40+ ACP-compatible agents | Claude Code, Codex, Antigravity |
| Project knowledge | Full: build system, symbol resolution, Compose Preview | Toolchain access via the android-cli skill |
| Emulator control | Through the IDE's Emulator | Create projects, manage virtual devices, inspect live UI layouts, run Journeys |
| Pick it when | You want previews and the project model in front of you while you review changes | Your agent already lives in a terminal, worktree, or CI pipeline |
To install the CLI plugin, from a Claude Code session run /plugin marketplace add android/android-cli-plugin then /plugin install android-cli@android-cli-plugin. For Codex from the terminal, the commands are codex plugin marketplace add android/android-cli-plugin and codex plugin add android-cli@android-cli-plugin — note the second Codex verb is add, not install. Google also ships an official Android plugin for Antigravity: install it from the plugin marketplace in the Antigravity app under Customizations > Build with Google to get Compose Previews and the Device Hub alongside your conversation. If you are weighing Antigravity as the agent itself, our Google Antigravity beginner guide walks through its install and first agent task.
How do you review what the agent changed?
You review agent-made changes with Change Review, a new agentic code review experience in Rabbit 2. Start a review session from pending agent changes or the Git changes window, and Android Studio generates a report of potential issues — or invoke the flow with the /review command from the agent tool window. You can ask for an AI-guided walkthrough to understand longer changes, then have your agent fix the issues it finds.
This closes the loop the whole feature is built around: the agent makes the change, the IDE's own review flags the risks, and the same agent resolves them.
Which agents work with BYOA?
Any ACP-compliant agent works — Google's stated number is "over 40 ACP-compatible coding agents directly inside Android Studio." The Rabbit 2 release post names four: Claude Agent, Codex from OpenAI, Junie, and Google Antigravity. The reason the number can be large with one integration is the protocol: the Agent Client Protocol standardizes communication between code editors and coding agents, with local agents running as subprocesses of the editor and talking via JSON-RPC over stdio. If your agent speaks ACP, it plugs in.
The open-source option in this lane is worth naming: our OpenCode beginner guide covers an open-source agent harness — and since Android Performance Analyzer's standalone app also accepts agents over ACP, the same protocol story now covers profiling as well as coding.
What does the data-terms lesson mean for third-party agents?
The privacy guarantees in the release post are written for Gemini Enterprise, and they do not automatically extend to Claude Agent or Codex. For Gemini Enterprise, Google says source code and workspace context are never retained or used to train global foundation models, with audit logging and central admin policy. A third-party agent inside Android Studio is still that vendor's agent on that vendor's terms — check your Anthropic or OpenAI plan's data terms before you switch either one on. This is the one lesson to take from the launch coverage and say out loud, once: BYOA brings the agent into your IDE; it does not bring your enterprise's data terms onto the agent.
What are the five beginner mistakes with BYOA?
Every one of these comes from the launch announcements themselves or from the first independent setup guides — none is hypothetical:
| Mistake | What happens | The fix |
|---|---|---|
| Trying BYOA on a Canary or older release | The marketplace path does not exist or behaves differently | BYOA is stable in Rabbit 2 — confirm the stable-channel update first |
| Assuming the data terms cover every agent | You grant a third-party agent project access under terms you never read | Read your agent vendor's data terms before the first authenticate |
| Confusing BYOA with the Android CLI plugin | You install the wrong half and wonder why the IDE shows nothing | BYOA = agent inside the IDE; CLI plugin = Android toolchain inside your terminal agent |
| Never switching agents | You pay for a marketplace you only browse once | Install a second agent and compare on the same task — switching is the feature |
| Skipping Change Review | Agent-made changes land unreviewed because "it looked right" | Make /review or the Git-changes review session your default before committing |
When should you NOT reach for BYOA?
Three honest no-goes from this run's sources. First, if what you want is the AI-guided XML Views to Compose Migration Assistant — that is still Canary-only, not part of Rabbit 2 stable. Second, if you are really asking "which model is best for Android work" — that question belongs to Android Bench 2.0, Google's LLM leaderboard that grades models and agent harnesses on long-running real Android tasks (migrating an app's UI, building an app from mocks), scored on whether the solution actually works rather than just compiles, and measured for cost and time. Third, if you never open Android Studio — the CLI plugin route gives your terminal agent the same Android toolchain without the IDE, and BYOA has nothing to offer you there.
Sources
- developer.android.com/blog/posts/android-dev-summit-26-developer-tools-performance (Google's own announcement, fetched November 3, 2026): BYOA stable in Android Studio Rabbit 2, 40+ ACP agents, the Settings > Tools > AI > Agents path, what the agent can access, Antigravity integration, Change Review and the /review command, the Claude Code and Codex plugin install commands, the Antigravity plugin install path, Android Bench 2.0, the Migration Assistant's Canary status.
- dev.to/gabrielanhaia/android-studio-now-runs-claude-and-codex-through-acp-android-cli-doesnt-need-the-ide-3kkp (November 2026, independent explainer): the named agents (Claude Agent, Codex, Junie, Antigravity), the ACP/JSON-RPC mechanism, the Gemini Enterprise-only privacy wording, per-agent authentication.
- developersdigest.tech/blog/android-studio-rabbit-2-claude-codex-acp-android-cli-plugin-2026 (November 2026, independent setup guide): the IDE-vs-CLI split, the
android skills list/android skills addcommands, 30+ official skills across 20+ areas.
Frequently asked questions
Is Android Studio Rabbit 2's Bring Your Own Agent free?
The BYOA feature itself ships with Android Studio Rabbit 2 — there is no IDE-side fee. What costs money is your agent: each installed agent runs on its own account or API plan, and you authenticate per agent. Budget for the agent, not the IDE.
Can I switch between agents in Android Studio?
Yes — that is the point of the marketplace design. Install more than one agent from Settings > Tools > AI > Agents and switch between them as you work, comparing them on the same task. You get new agent features as soon as they ship.
Does my agent get access to the Android Emulator?
Yes. BYOA agents get access to what makes Android Studio powerful for Android work: the Emulator, Logcat, the build system, symbol resolution, and Compose Preview rendering. The CLI plugin route gives terminal agents emulator and virtual-device management without the IDE open.
Is my code private when I use a third-party agent through BYOA?
The privacy guarantees Google published with Rabbit 2 cover Gemini Enterprise — source code and workspace context are not retained or used to train global foundation models under those terms. Those terms are not stated to extend to Claude Agent, Codex, or Junie. Treat each third-party agent under its own vendor's data terms.
What is the /review command in Android Studio Rabbit 2?
/review invokes Rabbit 2's agentic Change Review from the agent tool window: Android Studio generates a report of potential issues in the agent's changes, walks you through longer changes with AI guidance, and your agent can then fix the issues it finds.
