BeyDock
Every agent. One dock.
Claude, Codex, Antigravity and OpenCode don't just sit side by side in BeyDock: they hand work to each other. Bring the accounts you already have, work on this Mac or over SSH, and keep every session, limit and approval in one place.
- Claude
- Codex
- Antigravity
- OpenCode
Better together
Agents that work together.
A session can hand work to an agent on another provider and account. Claude plans, Codex reviews, Antigravity writes the tests, and every answer comes back to the session that asked.
- Any provider, any account. Each agent runs on the plan you pick for it.
- In parallel. Children work at the same time, each in its own session.
- Back where you asked. Results land in the parent session, ready to act on.
Bring your own accounts
Every plan you already pay for.
Sign in with the accounts you have. BeyDock picks one per session, shows how much of each plan is left, and lets any of them hand work to another.
- Claude
- Codex
- Antigravity
- OpenCode
Every account can hand work to every other.
BeyDock also runs your sessions on remote machines over SSH ↗
Get to know BeyDock.
Sessions
Side by side, not tab by tab.
Open a session with any agent on any account. Claude, Codex, Antigravity and OpenCode share one sidebar, one composer and one set of shortcuts, so switching models never means switching apps.
Every account, one switch. Pick the account per session. Claude and Codex sessions can move to another account of the same provider mid-way.
Limits before they bite. Five-hour and weekly windows for every plan, so you know which account has room.
This Mac or any SSH host. The agent runs where the code lives. BeyDock connects to it.
Branch and rewind. Try another direction from any message, or roll the files back with the conversation.¹
Worktrees, commits, PRs. Give a session its own worktree, then push, open a PR or merge locally.
Find any session in a snap.
Search every session. Titles come from the agent itself, so last week's work is one word away.
See where the tokens went. Tokens, cost, cache hits and active time for every session, updated as it runs.
Delegation
Second opinions, built in.
Any session can hand a task to an agent on another account or provider and read the result. Run side tasks in parallel, ask a different model for a review, or split a big change across agents.
Parallel side tasks. Child sessions run on their own and report back when they're done.
Any model, any account. Choose the provider, account, model and effort for each task.
Edits in their own worktree. A delegated edit works on a separate worktree. You decide what comes back.
Results you can open. Every child is a real session you can read and take over.
Approve what matters. Nothing more.
Permission modes decide what runs on its own. Everything else stops at a card that shows the full command, or every changed file with its diff.
Workflows
Agents on autopilot.
Build automations on a canvas. A schedule, a pull request, a failed CI run or a push starts them, and each agent step runs on the provider and account you pick, next to shell commands, approvals and GitHub actions.
How workflows runDraw it, or ask for it. Edit the graph on the canvas, or let the assistant change it while you watch.
Triggers that fit. Manual, schedule, pull request, CI run, comment or git push.
Every run is a session. Open any step, read what the agent did, and take over when you want.
A human where it counts. Pause a run for an approval, and get a notification when it's done.
Runs with the lid closed.
Keep Awake holds the Mac awake while agents work, even with the lid closed, and lets go on its own when the battery runs low or the Mac gets too warm.
Connections
Set up MCP once. Use it everywhere.
The BeyDock Hub connects to your MCP servers one time, for every provider and account. Agents get a single entry, and secrets stay in the Keychain.
One hub. Servers connect through the hub, not through each agent.
Policies and approvals. Allow, ask or deny per tool, with an activity log of every call.
Shared skills. Your skills, linked into every provider's folder. Turn it off and only our links go.
Your other apps, too. Claude Desktop, Cursor and other MCP clients can use the same hub.
Built for the Mac
Feels like it came with macOS.
Liquid Glass, native menus and the shortcuts you already know. BeyDock lives in your menu bar, on your desktop and in Control Center.
Liquid Glass. Glass, tint and accent follow your Mac, or your taste.
Menu bar items. Limits, sessions and Keep Awake one click away.
Widgets and controls. Usage on the desktop, controls in Control Center.
Remote machines, first class.
Add an SSH host and BeyDock starts a small daemon there. Sessions keep running when you disconnect, and pick up where they were when you come back.
Settings
Make it yours.
Choose the look, the accounts, what agents may do on their own, and what the menu bar shows.
Accent color, glass and tint, from fully clear to almost solid.
Add accounts for every provider and see each plan's limits in one list.
Decide which accounts other agents may delegate to, and with which access.
Idle sessions that aren't on screen are put to sleep, and wake on your next message.
Better together
Plays well with your tools.
Terminal, changes and a browser sit next to the conversation, so checking the agent's work never means leaving it.
A real terminal in the session's folder, on this Mac or the SSH host.
Every file the agent touched, with its diff, ready to commit.
Preview the app the agent is building without switching windows.
Questions? Answers.
What is BeyDock?
A native Mac app that runs coding agents (Claude, Codex, Antigravity and OpenCode) side by side, with several accounts, on this Mac or on your SSH hosts.
Which agents does it support?
Claude through the Claude Agent SDK, Codex, Antigravity and OpenCode. When a provider can't do something, BeyDock switches that feature off for it instead of leaving a button that does nothing.
Can one session use more than one provider?
No. A session stays on its provider (Claude and Codex sessions can move to another account of the same provider). To bring in another provider, the session hands a task to an agent on that provider's account, and the result comes back into the session.
Do I need new subscriptions?
No. BeyDock uses the accounts you sign in with and doesn't sell model access.
Does it work on remote machines?
Yes. Add an SSH host and sessions run there, with the host's folders, worktrees and files.
Where do my secrets go?
MCP secrets stay in the macOS Keychain. Agents get a token for the BeyDock Hub, never the secrets themselves.
Which Mac do I need?
A Mac with macOS 26 or later, and Node.js (brew install node).
How much does it cost?
Pricing comes with the public release. Early access is open now.