Ambient monitoring for AI coding agents

Run more agents.
Stop watching every terminal.

Glanceable Command Center is one live board for every AI coding agent you run, served from your Mac menu bar: the TV across the room, the phone in your pocket, any screen on your network. It catches blocked and failing runs before they cost you an afternoon of rework. No cloud.

Claude Code first. Codex, Gemini CLI, and OpenTelemetry-compatible agents next.

LIVE BOARD · GLANCE

payments-api is waiting on you.

infra-terraform is failing. 3 others nominal.

waiting 4m · wants to run the payments migration

AI

infra-terraform · 8 failed tool calls in a row

5h 38%G
G9:41 PM

Live board

Glance viewDetail view

Open in browser (localhost:8787)

Demo scenario

LLM verdict (Apple Intelligence)

Pause serving

GLANCEABLE · 5

payments-api

needs you · 4m

web-checkout

working

infra-terraform

failing

data-pipeline

working

docs-site

idle

localhost:8787· LAN only · no cloud

One view across the agents you use

  • Claude CodeAnthropiclive today
  • Codex CLIOpenAIfleet tier
  • Gemini CLIGooglefleet tier
  • OpenTelemetry-compatible agentsOTLP standardfleet tier

Vendor dashboards naturally focus on their own agents. This one doesn’t care who made yours.

  1. 1
    G9:41 PM

    Install the menu bar hub

    One small Mac app. Your Claude Code sessions appear on their own.

  2. 2

    localhost:8787· LAN only · no cloud

    It serves your board on your network

    It publishes to your LAN and stops there. No account, no cloud.

  3. 3

    LIVE BOARD · GLANCE

    payments-api is waiting on you.

    infra-terraform is failing. 3 others nominal.

    waiting 4m · wants to run the payments migration

    5h 38%G

    GLANCEABLE · 5

    payments-api

    needs you · 4m

    web-checkout

    working

    infra-terraform

    failing

    Open it on any screen

    Apple TV, any browser, the phone in your pocket. Same live truth everywhere.

One truth, two ways

Glance from the sofa.
Inspect from the desk.

The hub owns the view. Pick one in the menu bar and every screen follows: Apple TV, every browser tab, all of them. The menu on the left is live; switch to Glance to see how calmly the same moment reads from the sofa.

the Mac menu bar

Live board

Open in browser (localhost:8787)

Demo scenario

LLM verdict (Apple Intelligence)

Pause serving

LIVE BOARD · DETAIL

LIVE

payments-api

NEEDS YOU

Bash: npx prisma migrate deploy

Claude Code

web-checkout

WORKING

Editing src/checkout/session.ts

Codex CLI · 1,930 tok/min

infra-terraform

WORKING

8 FAILED IN A ROW

Retrying: terraform apply

Gemini CLI

data-pipeline

WORKING

Bash: pytest -q tests/

Claude Code

docs-site

IDLE

Finished 12 min ago

Claude Code

2 working1 needs you1 failing1 idle5h 38% · 7d 21%

With Apple Intelligence on your Mac, the Glance caption is phrased on device and marked AI. A validation gate rejects any wording the telemetry can’t prove; with it off, honest templates carry the wall.

Running is not the same as healthy. Glanceable shows both. See both views running on demo data →

Fidelity follows distance

Every screen gets a verb.

The farther a screen sits from you, the calmer it speaks. One hub feeds them all.

LIVE BOARD · GLANCE

payments-api is waiting on you.

infra-terraform is failing. 3 others nominal.

waiting 4m · wants to run the payments migration

5h 38%G

Across the room

Glance

TV or wall display

The calm sentence and the wash. Green until it isn’t, and when it isn’t, it names who needs you.

LIVE BOARD · DETAIL

LIVE

payments-api

NEEDS YOU

Bash: npx prisma migrate deploy

Claude Code

web-checkout

WORKING

Editing src/checkout/session.ts

Codex CLI · 1,930 tok/min

Arm’s length

Inspect

Browser or spare monitor

Detail view: every tile, sparkline, and dollar, with the exact command each blocked session wants.

GLANCEABLE · 5

payments-api

needs you · 4m

web-checkout

working

infra-terraform

failing

In your pocket

Triage

Phone

The same board, one column, the session that needs you first. Answer the question from the hallway.

Live board

Glance viewDetail view

Open in browser (localhost:8787)

Demo scenario

LLM verdict (Apple Intelligence)

Pause serving

Where you work

Control

Mac menu bar

The G is the status light, and the menu is the room’s remote: views, demo scenarios, AI verdict, pause.

What it watches

Everything you glance at,
nothing you babysit.

It answers one question from across the room: is anything blocked, going wrong, or burning money right now?

  • Know exactly which agent needs you

    Blocked sessions rise to the top of the board in amber, carrying the exact command waiting for approval. "Write: src/auth.tsx", not "waiting for input".

    • The blocked tile names the tool and the target, not just the state
    • One alert per episode, so an hour-long block is one chime and not sixty
    • The menu bar G turns amber the moment anything needs a human
  • Catch a bad run while it is still running

    A wrong turn and a runaway bill are often the same event seen from two angles. Failed tool calls, token velocity, and your plan quota land on the board together, while the session can still be stopped.

    • A run of failures paints the whole tile, readable across a room
    • Your plan’s 5-hour and 7-day windows, with an arrow when they’re climbing
    • Every cost estimate traces to a source, or it is not shown
  • Stay in flow, and keep it on your network

    The quiet state is the point. When nothing needs you the board says one calm sentence, and none of it leaves the building to say it. Even the AI that phrases the wall runs on your Mac.

    • Local-first: processing and storage stay on your Mac; board views are served only across your local network
    • Apple Intelligence phrases the verdict on device, marked AI; a gate rejects anything the telemetry can’t prove
    • The board survives restarts: the hub replays its own event log, so nothing forgets who your sessions are

Shipping in the open

What's built now, and where it goes.

Watching first, then every screen and vendor, then further from home. Solid cards are built; outlined ones are the plan.

Built · the launch board

Start with your own fleet

Mac menu bar hub and web board, zero terminal setup. Glance and Detail views, on-device AI phrasing, quota trends, restart-surviving history. Up to two Claude Code sessions, free. TestFlight first; the launch list below hears the moment it opens.

Next · Fleet

Every screen, every vendor

Apple TV and iPhone apps, needs-you alerts, and adapters for Codex, Gemini CLI, and OpenTelemetry agents. The one-time purchase.

Later · Anywhere

Off the network, still watching

See your fleet from anywhere, with longer history in one place. Still watching, not driving.

The next act · Teams

The same board, pointed at a whole team

On-prem by design, one board for every vendor your team runs, live spend across the floor. Built on what a single developer already gets, and built for coordination, not surveillance: team and cost signals, never a scoreboard over individuals.

The honest map

Monitoring isn't orchestration. Orchestration isn't visibility.

Most tools people line up side by side are in different categories. Here is the map, so you can compare like with like.

Glanceable Command Center

observation plane

Watches or runs
Observes
Which agents
Claude now; multi-vendor with Fleet
Lean-back?
Yes, ambient
On your network
Always, on-prem

Claude Agent View / Remote Control

first-party

Watches or runs
Observes
Which agents
Claude only
Lean-back?
No, lean-in
On your network
Cloud account

Omnara, Claude Squad, orchestrators

control plane

Watches or runs
Runs and controls
Which agents
Some multi-vendor
Lean-back?
No, lean-in
On your network
Varies

ccusage and usage dashboards

analytics

Watches or runs
Analyzes, after the fact
Which agents
Some multi-vendor
Lean-back?
No, lean-in
On your network
On your machine

Each of these is good at its job. Ours is the one none of them do: the ambient, multi-vendor, on-prem observation layer that sits above whatever you use to run and act.

Commitments, not features

Why you can trust the numbers.

A dashboard that lies once gets double-checked forever, and then it isn't glanceable any more.

  • If we can’t measure it, the widget doesn’t exist.

    No first-token latency chart, because no honest source exists. Near-neighbors are shown and labeled as what they are.

  • The board alerts on its own blindness.

    If the feed goes quiet, the board says so loudly. A stale dashboard that looks live is the worst failure a glance can have.

  • Restarts never fabricate a spike.

    Rates are computed from stamped receive times. Replayed history can update totals; it can never mint a burn rate.

  • One owner per number.

    When first-party telemetry is available it owns the counts outright. Two sources never feed one average.

  • Uncertainty wears an asterisk.

    A count we couldn’t fully verify says so on the tile, instead of rounding to confidence.

  • Alerts fire once per episode.

    An hour-long block is one alert and one escalation, not sixty. A board that cries wolf gets muted, and then it’s worthless.

No required subscription to watch your own agents

Own the board outright.

Watching your fleet is a one-time purchase, forever. One buy across Mac, Apple TV, and iPhone.

Free

€0forever

See it work with your own sessions.

  • Up to 2 concurrent sessions
  • Claude Code support
  • Mac menu bar hub and web board
  • Honest telemetry, same as paid
Download free

Fleet

€59one-time

Unlimited sessions, every screen, and multi-vendor adapters.

  • Unlimited sessions
  • Apple TV and iPhone apps
  • Needs-you alerts and Live Activity
  • Multi-vendor adapters: Claude Code, OpenAI Codex, Gemini CLI, OpenTelemetry-compatible agents
  • Yours forever; optional €22/yr for updates and new adapters
  • Token burn, cost, and context telemetry
Get it free in the beta

Launch pricing is provisional while App Store economics are finalized. Nobody will be charged before the product ships.

No surprises

Fair questions.

What is Glanceable Command Center?

Glanceable Command Center is a local-first observation plane for AI coding agents. It puts every agent session on one live board across your Mac menu bar, iPhone, and Apple TV, so you can see at a glance which agents are working, blocked, finished, or burning budget. It runs on your own network, and nothing you monitor leaves your Mac.

Which agents does it support today?

Claude Code, fully: session state, the exact blocked command, token burn, cost, and context telemetry. The Fleet tier adds OpenAI Codex CLI and Google Gemini CLI adapters, plus anything that speaks OpenTelemetry, and every future adapter is included, never sold separately.

Does my code or terminal output leave my network?

No. Processing and storage stay on your Mac. Board views are served only across your local network, never through our cloud. Screens on your LAN read from the small hub you run; the live board never depends on a cloud service. A future Anywhere option may add off-network access and longer history, but it would be opt-in and kept clearly separate from local monitoring.

What do I need to run it?

A Mac running Claude Code. The hub is a small menu bar app; the Apple TV and iPhone apps are optional screens. No server, no Docker, no terminal setup. If your Mac has Apple Intelligence, the wall’s phrasing runs on device; without it, honest templates do the same job.

What are the Glance and Detail views?

Glance is one calm sentence for the whole fleet, sized for a TV: who needs you, what they want, everything else summarised as a number. Detail is the full board, every tile, sparkline, and dollar. The hub owns the switch, so picking a view in the menu bar flips every screen at once, Apple TV included.

Is it a subscription?

No. Fleet is a one-time purchase and the version you buy is yours forever, offline. Continued updates and new adapters are an optional €22 a year, and if you lapse it you keep what you have and stop paying.

When can I get it?

TestFlight first, then the App Store. Join the launch list below and you’ll hear exactly once when each happens.

Phone alerts when I’m away from home?

Version one alerts on your own network only, because that keeps the privacy story absolute. An opt-in relay for away-from-home alerts is under consideration, and would carry alert metadata only.

Does it slow my sessions down?

No. Reporting is fire-and-forget with hard timeouts measured in milliseconds, and heavy work is throttled at the source. The board going stale is always preferred over your session waiting.

What about teams?

Single-operator first, and shared team boards come next. Everything in the teams section of this page is built on what one developer already gets, so it is not a different product. Join the launch list to hear when team access opens.

Know when Fleet ships.

The beta is out now and it’s free, so there’s nothing to wait for. This is the other list: one email when Fleet ships and stops being free, so the price never surprises you. Nothing else, ever.

Would rather just use it? Download the beta free →