GitHub Copilot CLI
Native SDKYour installed Copilot CLI, with your local sign-in.
Copilot setup ↗
Your computer works. You stay connected.
Your coding agents.
Your coding agents.
Now within reach.
Take your coding conversations with you. Send work from Android, review progress and approvals, then continue the same chat at your terminal.
Android 8.0+ · Open source · Preview releases
Independent app. No Mochi installation required.
ONE CHAT / TWO SCREENSWorkflow illustration
01 / ON YOUR PHONE“Review the changes before I merge.”
AgentLink Bridge · authenticated
02 / BACK AT YOUR TERMINALContinue where you left off.Same session · Same task queue · Same approvals
Your repository and agent processes stay on your computer. The terminal shown represents AgentLink's shared UI, not a separate vendor CLI.
01 / Choose your agentFamiliar tools.
Familiar tools.
One mobile home.
Keep the coding agent you already use. Install and sign in on your computer; AgentLink brings its conversations to your phone. One configured agent is enough.
Claude Code
ACP adapterClaude on your computer, connected through its ACP adapter.
Claude setup ↗
Kimi Code
Native ACPThe current JavaScript CLI, using your Kimi login.
Kimi setup ↗
Qwen Code
Native ACPYour locally authenticated Qwen Code, ready to connect.
Qwen setup ↗
DeepSeek Harness
Developer previewDSH with your local provider configuration. Preview limitations apply.
DeepSeek setup ↗Brand marks identify compatible tools, not partnerships or endorsements. DeepSeek's mark identifies the provider, not a separate Harness logo. Icon sources & license
Compatibility, session recovery, and accounts
Copilot, Claude, Kimi, and Qwen support context + history recovery for sessions they expose. DeepSeek Harness: context only · no old-message replay. AgentLink retains messages already saved locally; switching agents does not migrate a conversation between providers.
Claude requires claude-agent-acp. Kimi uses the current JavaScript CLI and kimi login, not the archived Python CLI. Qwen connects through qwen --acp. Follow the compatibility guide for DSH version requirements and setup details.
Accounts, model access, fees, reasoning controls, permissions, and image support depend on the agent and provider. New integrations may require both a newer app and Bridge.
02 / Take the conversation with youFrom a quick request
From a quick request
to the next decision.
Follow work away from your desk, give your agent visual context, and pick up the same thread when you return.
See what is happening.
Stream replies, inspect tool activity, handle approval requests, and stop the current task. Stay involved without keeping a terminal in view.
A screenshot says more.
Attach one PNG/JPEG, preview it, and send it with optional text. New images are compressed to at most 1 MiB. Requires an image-capable agent/model and updated App and Bridge.
Image input & cache limits ↗Each project, its own space.
Manage multiple computers and chats from one app. Every chat keeps its own agent, project path, and context; your credentials stay configured on the computer.
Phone to keyboard.
No copy and paste.
The optional AgentLink terminal shares your session, messages, task queue, and approvals with Android. Return to the same conversation, not a separate vendor CLI.
03 / Get connected
Two installs. One connection.
An Android phone, a computer with Python 3.11+ and Git, and at least one installed, authenticated coding agent. Keep the computer awake, online, and running the Bridge.
- 01 / PHONE
Install AgentLink.
Choose the newest Android APK from AgentLink's Releases, including pre-releases. Use the same signing channel for in-place upgrades that preserve saved chats.
Get the Android APK ↗ - 02 / COMPUTER
Start the Bridge.
Install your chosen agent and sign in locally first. Then start AgentLink's server. One Bridge discovers multiple agents; you do not need a server per agent.
Windows quick start ↓ - 03 / CONNECT
Pair. Pick a project.
Open Machines → Scan QR. Compare the confirmation code and approve pairing on your computer. In Chats, choose your computer, agent, and a project path on that computer.
Pairing guide ↗
Windows PowerShell · first-time computer setup
Run in the folder where you want to keep AgentLink. This installs the Bridge, not the coding agents.
git clone https://github.com/gongpx20069/android-agent-link.git
cd android-agent-link
python -m venv .venv
.\.venv\Scripts\python.exe -m pip install -e .\bridge
.\.venv\Scripts\python.exe .\bridge\run.py startFollow the Dev Tunnels sign-in prompt if shown. Keep this terminal running; the server prints the pairing QR code. Next time, run only the last command from this repository folder.
Other environments and full setup ↗Optional · continue your chats in the terminal
Wait for active tasks to finish, then stop the server with Ctrl+C. From the same repository folder, install the terminal extra and restart:
.\.venv\Scripts\python.exe -m pip install -e ".\bridge[interactive]"
.\.venv\Scripts\python.exe .\bridge\run.py start --interactiveOpen your chat on Android. Use /chats to choose between chats in the terminal. This is AgentLink's shared terminal, not a mirror of an independently running vendor CLI.
Updating the APK does not update the computer.
Upgrade the app and Bridge separately. After remote tasks finish, stop the Bridge, update its checkout and installation using the update guide, then restart it. Reopen the terminal after installing a new agent so its executable is visible on PATH.
04 / Know the boundariesYour computer.
Your computer.
Your connection.
Private by default.
Authenticated Dev Tunnels avoid inbound firewall ports and an extra phone VPN app. Confirm pairing; never enable anonymous tunnel access or share pairing tokens.
Local workspace.
Not an offline model.
Your files and agent processes live on your computer. Model requests still follow the coding agent provider's data policy and may include project content.
Connected, not always on.
The computer must stay awake with the Bridge running. Credentials can expire. Android background monitoring is limited to one hour, not guaranteed permanent connectivity.
Do I need Mochi or all five coding agents?
No. AgentLink is an independent Android app and computer-side Bridge. One configured coding agent is enough; Mochi and terminal chat are optional.
What does “missing” mean for an agent?
The Bridge cannot find that agent's executable, or its required adapter, on its PATH. Install it on the computer, reopen the terminal, restart the Bridge, and refresh Machines. “Available” means an executable was found; it does not prove authentication or model access.
A missing installation is different from an absent picker entry. If Machines lists an agent but New session does not on the same computer, include your app/Bridge versions and a redacted screenshot in a bug report.
Can I resume any existing conversation?
Only sessions offered by the connected agent. Recovery support varies: DeepSeek Harness restores context without replaying earlier external messages. Switching agents does not migrate a conversation to a different provider.
Can I close the app or put my computer to sleep?
Keep the computer awake and the Bridge running. Closing a phone UI is not a cancellation request for remote work, but Android background updates are limited. Reopen AgentLink to reconnect and check authoritative task status.
Optional: can I use Mochi as a voice companion?
Yes. In Mochi Tools → AgentLink, open AgentLink and approve scoped access in its native screen. Enable the Provider, its Tools, and the AgentLink Skill. Mochi can discover authorized workspaces, create or continue shared chats, send tasks, and read results.
Credentials stay in AgentLink and are excluded from Mochi sharing. Results used by Mochi go to its model provider. Execution approvals remain human decisions; access is foreground-only, not enabled for scheduled Agents or Subagents. Background result delivery depends on Android scheduling. Keep both apps and the Bridge updated.
Explore Mochi →
Your next task.
Wherever you are.
Keep your computer working. Keep the conversation with you.