Android Studio is no longer a single-assistant IDE. Rabbit 2 (2026.2.2), stable as of October 28, 2026, lets you plug in Claude Agent, OpenAI Codex, Google Antigravity or any other agent that speaks the Agent Client Protocol (ACP), and Google says more than 40 coding agents qualify. What sets it apart from running them in a separate terminal is that the IDE hands them the project graph, build diagnostics and emulator controls.
What Bring Your Own Agent changes
Up to Rabbit 1, Android Studio already let you bring your own model, but the agent loop was Google's. With BYOA (Bring Your Own Agent) the agent is yours and the IDE becomes its workspace. According to the release notes, a connected agent gets:
- Build diagnostics and the Android SDK tools.
- Terminal and shell command execution.
- Emulator management and Logcat access.
- The Android skills installed in the IDE.
Google's argument is that feeding the agent this information directly, instead of letting it infer it from files, speeds up tasks and cuts token usage. At the Android Dev Summit it shared internal test results for the Android CLI, the same idea outside the IDE: rendering Compose previews made agents finish UI tasks 77% faster with up to three times fewer tokens, and with LeakCanary and Claude Opus 5.5 they found and fixed memory leaks nearly twice as fast while using about 25% fewer tokens.
Sign-in is flexible: a personal or enterprise subscription, or your own API key. Android Studio's built-in agent remains available at no extra charge.
What else ships in Rabbit 2
Skills manager. Android Studio bundles Android skills and Firebase agent skills for jobs such as migrating XML views to Jetpack Compose, upgrading to AGP 9, adapting layouts to foldables and tablets, or wiring up CameraX and Firebase. You discover, install and manage them under Settings > Tools > AI > Skills Manager, or by typing @ in the agent chat.
Change Review. AI code review starts from agent changes, pending commits or the Git Log. It leads with recommended fixes and, with Create Walkthrough, groups diff hunks by theme rather than walking through files alphabetically. That is what makes it practical to review what an external agent did across twenty files.
LeakCanary meets the agent. From the LeakCanary task in the Profiler, the agent explains the root cause of each leak and suggests the code change.
Gemini Enterprise. For teams that don't want every developer picking their own agent, there are admin policies to approve skills, agents and model endpoints, plus audit logging.
How to update and connect an agent
Rabbit 2 is on the stable channel: Help > Check for updates (Android Studio > Check for updates on macOS), or download it from developer.android.com.
Then:
- Open the agent window and pick one of the defaults (Claude Agent, Codex or Antigravity).
- Sign in with your subscription or enter an API key.
- For other ACP agents, browse the registry under File > Settings > Tools > AI > Agents (Android Studio > Settings on macOS).
If you also want the agent to use the same tooling from a terminal or in CI, update the Android CLI with android update, and android init installs the android-cli skill for your agent:
android update # fetch the latest CLI
android init # install the android-cli skill for your agentBefore you roll it out to the team
The sensitive part is that BYOA gives the agent shell execution on your machine, right next to your Gradle, signing and Git credentials. Before opening it up to everyone:
- Decide which agents are allowed and, if you have Gemini Enterprise, enforce it with admin policies instead of leaving it to each developer.
- Keep personal and company API keys apart: the IDE accepts both, and the bills land in different places.
- Run everything the agent proposes through Change Review before committing, especially changes to
build.gradle.ktsand signing config. - Start with a narrow skill (the AGP 9 migration, say) on a branch and compare the result with your current workflow.
If you already run Claude Code or Codex in a terminal next to the IDE, the gain is context rather than capability: the same agent, but one that sees the emulator, Gradle errors and Logcat without you pasting them in.