Splitlane

A native desktop app for CLI coding agents

See which agent is waiting for you.

Run Claude Code, Codex and 14 other CLI agents side by side, and see which one is waiting for you. For Claude Code and Codex, Splitlane reads the agent’s own records instead of guessing from the screen.

brew install --cask ivkan/splitlane/splitlane
Download for macOS Linux

Rust on GPUINo accountTelemetry only if you opt inGPL-3.0-or-later

Three agents in one project. One is working, one has finished, and one is asking to run npm test. The rail, its pane header and the title bar all say so.

How it knows

Read from the agent, not from the screen.

A permission prompt and an idle prompt can look alike in a terminal. A status dot you cannot trust is worse than none, so each agent's state comes from the best source that agent offers, and Splitlane says which source that is.

  • starting
  • running
  • waiting for you
  • idle
  • failed

The title bar counts who is waiting; the pane that needs you says so in its header.

  • Their own session records

    Claude Code's status file and transcript, Codex's rollout, and the process tree under the pane. A permission prompt, a running tool call and a finished turn are told apart from what the agent recorded.

    Claude Code, Codex

  • Hooks, for eight more

    They report through hooks that a small shim on the pane's PATH installs for the length of the session.

    Gemini, Cursor, OpenCode, Pi, Hermes Agent, Grok, CodeBuddy, Qoder

  • Only whether it runs, for the last six

    They expose no hook surface Splitlane can use, so their row can say running while the agent is in fact waiting for you.

    Copilot, Kiro, Factory, Antigravity, Openclaw, Amp

What is in the window

Every agent in a real terminal, and the diff beside them.

A project's panes are a row, a column, or a grid of four. You answer an agent in its own terminal; there is no second approval UI to keep in sync.

The same project as a grid of four: the three agents and the diff of what they changed.

  • The rail

    One list of projects, each expanding to its sessions with their status. ⌥⇥ jumps to the next session waiting for you.

  • Groups

    Projects sit under named labels. Fold a group away, or keep its project names out of notifications.

  • A branch per session

    In a git repository each agent session gets its own worktree and branch, so two agents never edit the same checkout, plus ten ports for a dev server.

  • Diff

    The project's changes against its base branch, uncommitted and untracked files included, in a pane next to the agents that made them.

  • Hand an answer over

    A session's last answer goes to another agent on screen, onto its input line; you press Enter.

  • Limits

    Your Claude and Codex plan usage in the rail, with when it resets.

One agent working through others

An agent can open sessions of its own. You still see each one.

An agent in a pane can hand other sessions a task, wait for their turn to end, read the answer and close them. Each of those sessions is a row in the rail with a status you can see, and a question meant for you still reaches you.

Splitlane with one project in a two-by-two grid. In the rail a session named plan carries the count 4, and a bracket in the margin joins it to the four sessions it opened: tests, api-docs, lint and fixtures. The api-docs pane shows a permission prompt and its header reads 'waiting for you'.

One session and the four it opened. One is working, one is waiting for you.

  • Eight at most, one level deep

    A session that was opened by another cannot open any, and it works in its opener's worktree.

  • A session you opened is yours

    When an agent wants to send to one, nothing is written until you allow it, in that agent's own pane.

  • What an agent typed says so

    Text one session sends another starts with a line saying it was sent by an agent session, not typed by a person.

Sessions outlive the window

Close the window. The agents keep working.

On macOS and Linux the terminals run in a separate process, the session host. Closing the window, an update or a crash of the window leaves every agent working, and the next start shows each one in its pane again.

Where a session does end, its conversation resumes on the next launch; a turn that was running does not. On Windows the sessions still live in the window's process.

Running (⌥⌘P) lists every agent and every command running in a shell, and stops the ones you mark.

Your phone

Answer a waiting agent from your phone.

A paired phone follows the agent sessions on the computer and answers them. There is no app to install from a store: it is a page at app.splitlane.dev that you add to the home screen.

  • The list is the rail's

    It opens with Needs you: sessions waiting for an answer or failed. A permission prompt comes with its command, and each numbered choice is a button.

  • End-to-end encrypted

    Neither side opens a port. The relay sees only that a message passed, its length and when. A phone writes only into agent sessions, never into a plain shell.

  • Pairing

    Scan the QR code from Settings → Phone, then type on the computer the six digits the phone shows, so a photo of the code is not enough.

The phone's list for one computer: a line with the Claude plan limit, Active and All tabs, a Needs you section holding the session npm test failures with 'Claude needs your permission', and below it the Work group with the project ledger.
The session npm test failures on the phone: Claude Code's permission prompt for 'npm test 2>&1 | tail -60' as its screen shows it, and its first choice, Yes, as a button.

Next to its neighbours

The column it was built around.

Several tools put coding agents side by side. What sets them apart is how each one learns that an agent needs you. Taken from each project's own documentation as of October 2026; they change quickly.

ToolWhere the agents runHow it learns an agent needs youPlatforms
SplitlaneTerminal panes in a desktop app, one git worktree per agent sessionThe agent's own session records for Claude Code and Codex, hooks for eight more; for the rest, only whether the process is alivemacOS, Linux; Windows from source
herdrPanes inside the terminal you already useRules matched against what the agent draws on screen; hooks or the agent's own reports for some agentsLinux, macOS, Windows
cmuxTerminal panes in a native macOS appTerminal notification sequences, a cmux notify command and hooks it installsmacOS
ConductorIts own chat interface, one git worktree per workspaceIts documentation does not saymacOS; no new sign-ups since November 2026
Where the neighbours are ahead, and where Splitlane differs →

Questions

Questions people ask first.

Why not tmux with one agent per pane?

tmux shows you the panes; it cannot tell you which agent is blocked on a question. If you work over SSH or want something headless, use tmux. Splitlane is for supervising agents on your own machine, where knowing who is waiting is the whole job.

Is it an Electron app?

No. It is Rust on GPUI, Zed's UI framework, rendering with Metal on macOS, Vulkan on Linux and DirectX on Windows.

Does it run agents for me, or in the cloud?

Neither. The agents are the CLIs installed on your machine, signed in with your own accounts, running in ordinary terminals. Splitlane never sends a prompt on its own.

An agent can send one to a session it opened itself, and to a session you opened only after you say it may. A paired phone writes what you type on it, and only into agent sessions.

Does it phone home?

Only if you opt in. Telemetry never includes terminal contents, paths or prompts, and SPLITLANE_NO_TELEMETRY=1 turns it off regardless of settings.

Its other connections: it asks GitHub for updates ("check_for_updates": false stops it), reads your Claude plan usage with the login Claude Code already stored, and, once you pair a phone, keeps a connection to the relay, which carries only encrypted messages.

Can other agents read my terminals?

Only agents you register with the read-only MCP bridge, and only to read. Output it hands to an agent is marked as untrusted content.

Install

Get Splitlane.

Download the file for your platform from the latest release. Splitlane is early: it is used daily on macOS (Apple Silicon); Linux and Windows are built and tested in CI but have had far less hands-on use.

PlatformFileHow
macOS (Apple Silicon).dmg, signed and notarizedOpen the image and drag Splitlane to Applications, or run brew install --cask ivkan/splitlane/splitlane. Intel Macs are not built yet.
Debian, Ubuntu.debsudo apt install ./splitlane-<version>-<arch>.deb
Fedora, openSUSE.rpmImport the signing key once, then install with dnf or zypper. Steps
Other Linux.AppImage or .tar.gzThe AppImage runs as is after chmod +x; the tarball installs under ~/.local without sudo.
WindowsNone yetNo signed installer yet; build from source.

Linux builds need glibc 2.35 or newer and a Vulkan driver. Every file has a .sha256 and a minisign .minisig beside it. Splitlane checks for a newer version at startup and every four hours, and a click in the title bar installs it.

See which agent is waiting for you.