Corral

agent
Security Audit
Fail
Health Warn
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Fail
  • process.env — Environment variable access in scripts/shot.mjs
  • process.env — Environment variable access in scripts/shots-all.mjs
  • new Function() — Dynamic code execution via Function constructor in scripts/ui-checks.mjs
  • process.env — Environment variable access in scripts/ui-checks.mjs
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

Run and supervise many Claude Code sessions in one desktop app — Linux, macOS, Windows

README.md

Corral

A desktop app for running many Claude Code sessions at once. It finds every session on your machine (and on
your servers over ssh), shows at a glance which ones need you, and lets you answer, review and switch between
them as fast as terminal tabs.

Built with Tauri 2, React 19 and shadcn/ui. Runs on Linux, macOS and Windows.

What it does

  • Overview. One tree for everything that runs: a main agent on top, a column per repository, each
    session as a card with what it is doing right now. What needs you is the only thing in colour.
  • Needs you. Permission prompts, questions, plan approvals and errors are counted in the header.
    Ctrl+J jumps to the one waiting longest, and routine permissions can be answered right on the card.
    Requests that are hard to undo (force pushes, recursive deletes, secrets, deploys) get their own card:
    the effect spelled out, no "always allow", never a single keypress.
  • Calm notifications. Only what blocks a session notifies, after 10 quiet seconds (a prompt you answer
    in the terminal never pings), at most a few an hour; the rest arrive as one summary. Finished work is a
    dot in the list; working sessions spin.
  • Sessions like Claude Code desktop. A chat per session with readable work summaries
    ("Edited 2 files, ran a command +38 −17"), a live activity line, the plan as a checklist, and a details
    panel with the context window breakdown, changes, pull request status and plan limits.
  • Back in context. Return to a session after a while and a card says what you last asked and what
    happened since. The details panel tells whether tests or checks ran after the last edit, and a banner
    offers "Step back" when the same step has failed three times in a row. A long, much-steered session
    can start fresh: a new session in the same folder, opened with an editable brief built from the old
    one (no model call).
  • Parallel work without collisions. Two sessions editing the same files in one repository are flagged
    on their cards: in the same folder they overwrite each other, in separate worktrees a merge conflict
    waits. Dev servers a session started show as :5173 chips that open in the browser, and a working
    session that has written nothing for five minutes says so.
  • After a restart. The sessions that were running when the app closed are offered back in one go.
  • At the usage limit. Sessions that stopped at the plan's limit are listed with the reset time; tick
    the ones that should go on and they continue one by one after the reset.
  • Side by side. Any session opens in its own window (session menu or Ctrl+K), for a second screen.
  • More than one account. Other Claude Code config folders (CLAUDE_CONFIG_DIR) are read too, and a
    workspace can start its new sessions in one of them.
  • Pull requests. Status and checks in the details panel; a red check's failing log lines go to the
    session with one click.
  • Permissions you granted. Settings → Permissions lists every "always allow" rule from your user and
    project settings, the ones that allow something hard to undo first, and removes one with a click.
  • Finds sessions by itself. Sessions started in any terminal show up on their own, grouped by
    repository (worktrees, servers and cloud sessions of one repo share a column). Optional hooks make their
    states exact and let you answer a terminal session's permission prompt from the app.
  • Terminal sync. Turn on Settings → Sessions → Sync terminal sessions and every claude you start
    in any terminal (Zed, GNOME Terminal, iTerm, Konsole…) stays one session with the app: replies you
    write in the app are typed into that terminal, and nothing is closed or copied.
  • Built-in terminal. Ctrl+` opens a real shell in the session's folder. ! command in the prompt
    box runs there, and Claude Code commands that need its own interface (/mcp, /login, /config)
    open in it.
  • Usage. Tokens, value at API prices, prompts and sessions per day, week, month, year or all time,
    an activity heatmap, per-model breakdown, where the value goes (cache reads, cache writes, output), and
    your plan's 5-hour and weekly limits over time. The history is kept in its own ledger, so it survives
    Claude Code cleaning up old transcripts. "Last 5 hours by session" shows which session uses the plan's
    5-hour window. Next to the prompt box, a session whose prompt cache has expired says what the next
    message costs to re-read the conversation.
  • Review. Files that deserve a look first (secrets, CI, migrations, auth, dependencies) lead the list
    and tests come last; files not opened yet carry a dot; very large diffs suggest going turn by turn.
    Approving a plan can allow the checks it names (tests, builds) for the rest of the session.
  • Git loose ends. Worktrees and branches with uncommitted or unpushed work, per repository.
  • Find any old session. Every transcript's conversation is indexed once (seconds), then kept up to
    date: Ctrl+K finds what was said in any session, in any language, and opens the chat at that
    message; History lists every session by day; each project shows its latest finished sessions under
    its running ones; old chats continue when you type into them.
  • The chat of Claude Code desktop. Permission prompts, questions and plans in the conversation,
    queued messages you can edit or send now, Rewind (files and conversation, with Claude Code's own
    snapshots) and Fork from any message, Find in chat, prompt history, attached files by path.
  • Fast. Sessions switch instantly, chats load from the end of their transcript, and nothing polls
    harder than it has to.

Keyboard

Keys Action
Ctrl+Tab Previous session (hold Ctrl for the recent list)
Ctrl+J Session waiting longest for you
Ctrl+1 … Ctrl+9, Ctrl+0 Session by position, overview
Alt+↑ / Alt+↓ Up / down the session list
Ctrl+K Search sessions and what was said in them (opens the chat at the message)
Ctrl+Shift+H History: every past session, by day
Ctrl+F Find in the chat on screen
Enter / Ctrl+Enter / Ctrl+Alt+Enter Send · interrupt and send · send in a forked session
↑ / ↓ in the prompt box Earlier prompts of the session
Ctrl+Enter / Ctrl+Shift+Enter on a request Allow once · always allow
Ctrl+O Transcript view: Normal, Thinking, Verbose
Ctrl+N New session
Ctrl+` Terminal
Ctrl+B Show / hide the session list
Ctrl+Shift+M / I / E Permission mode, model, effort
? All shortcuts

Download

Installers for Linux (.AppImage, .deb, .rpm), macOS (Apple silicon and Intel) and Windows are on the
Releases page. They are not signed
yet: on macOS open it once, then System Settings → Privacy & Security → "Open Anyway"; on Windows choose
"More info → Run anyway".

Requirements

  • Claude Code installed and signed in (claude on your PATH)
  • Optional: gh (GitHub CLI) for pull request status, ssh for sessions on servers
  • Windows: Git for Windows (Claude Code needs it too); the built-in terminal opens PowerShell

Build and run

# Linux (Fedora) system libraries for Tauri
sudo dnf install -y webkit2gtk4.1-devel openssl-devel curl wget file libappindicator-gtk3-devel librsvg2-devel libxdo-devel
sudo dnf group install -y c-development

bun install
bun run tauri dev        # development window
bun run install:local    # release build, installed for the current user with a menu entry

bun run dev serves a browser preview with demo data at http://localhost:3000.

Development

bun run typecheck        # TypeScript
bun run lint             # ESLint
bun run check:ui         # end-to-end checks against the browser preview (needs `bun run dev`)
cd src-tauri && cargo test --lib

Credits

Built on create-tauri-ui and shadcn/ui.
Code adapted from other open-source projects is listed with its license in
THIRD_PARTY_NOTICES.md.

License

MIT. Code adapted from other open-source projects keeps its own license, listed in
THIRD_PARTY_NOTICES.md.

Reviews (0)

No results found