A local dashboard for running Claude

Run, queue and orchestrate Claude from one dashboard

See every isolated Claude Desktop instance and which account it's on, live-tail all your Claude Code sessions, and build a queue of claude runs, each with its own prompt, model, effort and account, dispatched on demand or on a schedule. One daemon, your browser, a tray icon.

Download View source

Free & open source (MIT) · Windows, macOS, Linux · no account needed to run it

localhost:7787

CC Manager UI

Sessions Instances

Scheduler off Queue

Search sessions…

running

Add retry + backoff to the API client

acme-checkoutfeat/retry-backoff47 · 5m ago

Fix the flaky auth test on CI

billing-servicefix/flaky-auth19m ago

queued

Migrate the settings page to the design system

dashboardredesign/settings54m ago

Wire up the Stripe webhook handlers

paymentsfeat/webhooks2h ago

Track down the memory leak in the worker pool

ingest-workermain4h ago

Refactor the CSV importer for streaming

data-toolsrefactor/csv6h ago

Add retry + backoff to the API client

9e3779b1C:\work\acme-checkout6 turns shown

Our API client hammers the server and gives up on the first 429. Add retry with exponential backoff, and treat 429 vs 529 differently.

Good call. Those are two different failures. A 429 is your own rate cap and comes with a Retry-After header we should honour exactly. A 529 is the server being overloaded; that one we retry on our own schedule with backoff + jitter.

Cap it at 5 attempts and make the jitter full, not equal.

Done. Base delay 500ms, doubling each attempt, full jitter over [0, delay], hard cap of 5 tries. On a 429 I sleep for exactly Retry-After when present. Everything else fails fast.

Nice. Add a test that a 529 then a 200 resolves, and that a 400 never retries.

Added three: 529→200 resolves after one retry, a 429 waits for Retry-After, and a 400 throws with zero retries. All green.

Message this session. Enter to send, Shift+Enter for a new line

Opus 4.8 Effort: high Permissions work@acme Send

Instances (4) Create instance

Name	Instance account	PID	Uptime	Memory	Usage	Actions
work ~/.claude-instances/work	work@acme.com · Max	21840	2h 14m	512 MB	5h 12%wk 63%	Focus
personal ~/.claude-instances/personal	me@gmail.com · Pro	8104	41m	388 MB	5h 4%wk 21%	Focus
client-a ~/.claude-instances/client-a	client@studio.io · Max	-	-	-	Not checked	Open
Claude DesktopExternal your regular install	you@work.com · Max	4620	3h 02m	690 MB	5h 30%wk 71%	Focus

Settings

Appearance

ThemeLightDarkSystem

Show tooltips

Instances

Show desktop instances

Show CLI instances

Auto-refresh usage

Keep quota numbers current in the background.

Scheduler

Enabled

Work through the queue automatically.

Max concurrent3

Spacing (s)20

Auto-resume monitor

Enabled

Resume sessions after a 5-hour limit resets.

Updates

Current version0.6.0 · up to date

Cloud sync

Sync settings with ConnectionsConnect

Run queue

Scheduler off New run

Add retry + backoff to the API client

acme-checkoutOpus 4.8highrunningStop

Migrate the settings page to the design system

dashboardOpus 4.8highqueuedRun

Fix the flaky auth test on CI

billing-serviceSonnet 5mediumqueuedRun

// how it works

One daemon, your whole Claude workspace

Start the daemon

Run the tray app (or bun run start). It serves the dashboard and opens it in your browser, with no console to babysit.

See everything

Every isolated Desktop instance and its account, and every Claude Code session across them, live-tailed in one place.

Queue & dispatch

Build a queue of claude runs and fire them on demand or on a schedule, or hand the whole thing to your agents over MCP.

ccmanagerui: first run

$ bun run start

[ccmanagerui] reading ~/.claude/projects … 128 sessions

[ccmanagerui] discovering Claude Desktop instances … 4 found

[ccmanagerui] scheduler off · auto-resume off (opt in from Settings)

[ccmanagerui] dashboard → http://localhost:7787

// what you get

Many Claudes, one pane of glass

Multi-instance manager

Every isolated Claude Desktop instance on the machine, with its account (email + plan), live memory and uptime, in one sortable table. Open, quit, create, delete, rename, or drop a per-instance desktop shortcut.

Sessions across every instance

Browse and live-tail your Claude Code transcripts, filtered by instance, with "thinking" hidden by default. Open the raw .jsonl, download it, or copy the file, named after the session, not its uuid.

Chat composer

Type at the bottom of a transcript and dispatch straight to that session: now, queued, or scheduled for later. Multi-select to message many sessions at once.

Queue + scheduler

Build a queue of claude runs (prompt, model, effort, permission mode, account) and dispatch on demand, or let the scheduler run them with concurrency and spacing controls. Off by default.

Runs that survive a restart

Dispatched runs run under a detached supervisor and reattach after a quit or an auto-update, instead of dying mid-flight.

Auto-resume past a rate limit

Sessions stopped by a 5-hour limit are picked back up once the window resets, including ones you started yourself in a terminal. Gated on your weekly usage, capped per session, off by default. A 529 overload retries in seconds, automatically.

// one daemon, two front doors

Built for you and your agents

You get a dashboard

Click through your instances, sessions, queue and accounts in your regular browser. Message a session, schedule a run, watch it stream back, with no terminal juggling.

Your agents get MCP

The same daemon is exposed over MCP stdio, so Claude Code, Claude Desktop or Cursor can drive sessions, the run queue, the scheduler, accounts and instances: the exact same state you click.

sessionslist · get · tail

queueadd · run · cancel

schedulerget · set

instanceslaunch · quit

accountssecrets masked

check_usagequota, no human

claude_desktop_config.json

{
  "mcpServers": {
    "ccmanagerui": {
      "command": "bun",
      "args": ["run", "--cwd", "<path-to-ccmanagerui>", "mcp"]
    }
  }
}

Start the daemon first; the MCP server follows whichever port it actually bound. Mutating tools say MUTATES: in their description, and there is deliberately no shutdown tool.

// check_usage

Agents that know when to stop

Any MCP-speaking agent can read an account's remaining quota before it starts a heavy fan-out, with no human in the loop. Reading it costs nothing.

check_usage → work@acme.com

Session (5h)12%

Week (all models)63%

The binding cap. A fresh session % is a red herring when weekly is near 100.

Week (Opus)41%

Pace the fan-out, don't crash into the wall

Switching the flagship model doesn't dodge the shared weekly bucket, so an agent that checks first can route heavy work to whichever account has the most left.

And when a session does hit the 5-hour wall, the auto-resume monitor picks it back up once the window resets, including runs you started yourself in a terminal.

Local & private

Runs entirely on your machine. One daemon on your localhost, your regular browser, and a tray icon. No cloud, no account needed to run it, nothing leaves your computer. Open source under MIT.

Built with

Bun + Hono Vue 3 + Vite shadcn-vue Tailwind v4 bun:sqlite MCP stdio

Stop tab-juggling your Claudes

One daemon, one dashboard for every instance, session and run, plus an MCP door for your agents.

Download a release