For Vast.ai hosts · 14-day Ultra trial, no card

Run your GPU fleet
like a business.

HostDeck watches every machine, rental and dollar on Vast.ai — then tells you why earnings moved, what to charge, and what to fix before renters notice.

  • API key encrypted at rest
  • No price change without your consent
  • Set up in about two minutes
HostDeck overview: earned so far today against yesterday, Fleet Health, GPU occupancy and KPI tiles
1 minchecks on Ultra
5 yearsof earnings history on Ultra
14 daysof Ultra free, no card
0 SSHfor routine host chores

Monitor Every plan

Know what every GPU is doing, right now.

The Overview opens on what you earned since midnight against yesterday at the same time, where the day will end, one Fleet Health grade and the tiles that matter: net $/h after power, idle capacity, your price against the market median and 30-day uptime. Below it a live board you arrange yourself — today vs yesterday, the monthly goal, energy and net profit, demand by hour of the week — and a live card per machine. Every machine has its own page: GPU occupancy, rental sessions with start, pause, resume and end, earnings rate, reliability and the listing. Offline hosts, errors and machines Vast.ai drops from your account raise an alert — and an all-clear when they recover.

  • Updates stream in live; the LIVE badge on every page counts rented GPUs and opens your notifications
  • Discord, Telegram and Apprise channels, with quiet hours
  • Rules for low occupancy, low daily earnings, idle machines and more
  • A weekly digest: earnings vs last week, best and worst machine, open tasks
A machine page with live KPIs, GPU occupancy and rental sessions
Notification targets, alert rules and the weekly digest settings

Charts Every plan A few cards: Pro

The whole fleet over time — up to a year of it.

Thirteen cards on one page, from the last hour to twelve months or any dates you pick. Filter every card at once to a rack, a tag, a GPU model or a few machines, drag the replay slider to see the fleet as it was at any moment, pin notes to the time axis and turn what a chart shows into an alert rule — with how often it would have fired.

Stacked chart of the fleet's GPUs by rental state over seven days
Fleet occupancy over timeEvery GPU by state, GPU model or tag, with offline hosts, price changes, your notes and your alert thresholds on the time axis. Drag to zoom every chart on the page.
Hour of the week: seven rows of 24 hourly cells with the quietest and busiest hours
Hour of the week · 24×7The range, a week at least, folded onto weekday × hour: when the fleet sleeps, its busiest hour and weekends against weekdays — in occupancy or $/h.
Rental swimlane with a lane per machine, a row per GPU and a histogram of rental lengths
Rental swimlaneEvery rental of the range as a bar in its type's colour, a lane per machine and a row per GPU, the hours without a rental hatched — and how long rentals run: median, 75th percentile, longest.
The previous period's revenue, this range's and the change between them, split into downtime, occupancy, price, storage and mix
Why revenue changed ProThe range against the stretch before it or a year earlier: downtime, occupancy, price, storage and mix, with the machines that moved it most.
Machines that climbed and fell in the fleet's ranking by earned $/GPU/h, each with its place day by day
Machine rankingEach machine's place by the $/GPU/h it earned against its GPU model's median: who climbed, who fell and who led, with the place day by day beside each.
  • Fleet horizon: every machine against the fleet median, in thin bands
  • Where the potential went: what was realized, rented below the price, idle and offline, in one bar
  • Fleet temperature: one row per machine, hottest first
  • You vs the market: your $/GPU/h against the daily market band Pro
  • What a kWh earns: revenue per kWh against your tariff, per rack Pro
  • Views (Morning check, Prices, Hardware…) kept with your account, on every device
  • A table, CSV and PNG on every card; a link reopens the exact view
  • Days from your own midnight; a colour-blind-safe palette and full keyboard use

Earnings Every plan Autopsy: Pro

Earnings that explain themselves.

Today so far, ticking at the live rate against yesterday at the same time; a month ring measured against your monthly goal; and the Month race: this month added up day by day against the month before, your best month and the same month a year ago, with the forecast's 50% and 90% cone to the month's end. When the total moves, Earnings Autopsy breaks the change into downtime, occupancy, price, storage and bandwidth, and machines added or removed — as a waterfall, with the machines that moved it most.

  • How far ahead of or behind last month you are, and what the goal needs per day
  • The total, its GPU / storage / bandwidth split and the per-machine table add up
  • Import your full history from the Vast.ai CSV export
  • What-if sliders for occupancy, price and machines online on the forecast
  • "Why it changed" lands in your weekly digest too
Month race: cumulative earnings this month against earlier months, with the forecast to the month's end
Earnings Autopsy waterfall splitting a weekly change into its causes

Insights Pro

Only the signal, not the noise.

Every machine's earnings, reliability and GPU temperature are checked against their own 7-, 14- or 30-day baseline. A finding needs both a real Z-score and a real change in the value, so a flat baseline cannot turn a cent into "many σ". Each one is drawn against its baseline band and placed on a fleet timeline of the last 24 hours.

  • Findings tracked as episodes: how long it has lasted, what is new since your last visit
  • Why? splits an earnings drop into downtime, occupancy and price
  • One click to the next step: open the machine, review the price, schedule maintenance or add an alert rule
  • Fleet Health score trend and a machine × component matrix
Insights page with anomaly cards, a fleet anomaly timeline and summary tiles

Pricing Pro

Price against the market, not a guess.

Every machine is placed against the current market band for its GPU model, with occupancy, $/GPU/day and a demand curve. Apply a suggestion in one click, or let the Pricing Autopilot lower idle machines toward the median and raise busy ones — in dry run first, then live when you trust it.

  • Floors, ceilings, a maximum step and a cooldown per machine
  • An energy floor from your tariff, so it never prices below the electricity
  • Every change logged with 72 h before and after — and one-click rollback
  • Nothing is written to Vast.ai until you allow it for that machine
Pricing Autopilot policies and the price change log with attribution

Profitability Pro

Net profit, after the electricity bill.

Enter your tariffs — the Polish G11, G12, G12w and G13 come as presets, with zones by weekday, weekend, holiday and season, one per datacenter — and your hardware costs. HostDeck measures energy from the host agent (or estimates it from board power) and shows what the fleet actually makes. Where the money goes walks from GPU, storage and bandwidth revenue through energy and fixed costs down to net profit; the Daily net profit calendar costs every day on its own, so weekends and outages show at a glance.

  • Profit by hour of week against your tariff's zones
  • Payback timeline with energy-price and occupancy what-ifs
  • What a GPU-day earns per GPU model, split into energy, fixed costs and net
  • The break-even $/GPU/h per GPU model next to your price and the market's
Waterfall from GPU, storage and bandwidth revenue through energy and fixed costs to net profit
Daily net-profit calendar on the Profitability page

Inventory Every plan Rack view, map and fleet planning: Pro

Know what every GPU model earns for what it is.

Your hardware, counted and priced: which models earn more than their share of the fleet, what the room draws this minute, how worn every disk is and how the fleet grew. Click a bar, a model or a rack and the whole page filters to it.

Hardware economics: GPU models' share of GPUs, VRAM and compute against their share of revenue, with a demand and price chart
Hardware economics ProEach model's share of GPUs, VRAM and compute against its share of the revenue; revenue per GPU, per GB and per TFLOPS; occupancy and the price a rented hour really earns.
Rack view grouped by GPU model, with hosts and GPUs lit by rental type
Rack view by GPU model ProGroup the hosts by datacenter, rack, location or GPU model and see how many of each model are rented right now.
Energy now card with power, cost, circuit headroom and power per rack
Energy now ProThe room's draw this minute, its cost at each datacenter's tariff, the headroom to every datacenter's and rack's circuit, room for more GPUs, power over time and what another tariff would have cost.
Disk health: endurance used, temperature and projected life per disk
Disk health UltraEndurance used, TB written, write rate and life left for every disk, with its temperature and link — ranked by wear.
Fleet history chart of GPUs by model over time with a time slider and a list of changes
Fleet history ProReplay the fleet day by day with a time slider: what it owned then against today, and every host that joined, left or changed.

Fleet planning Pro

Your server room, and what to buy next.

Every host as a chassis in its rack, every GPU lit by what it is doing. Then the planning tools: what holds each host back, what an upgrade would earn, when each machine paid for itself.

Rack view with racks of hosts and one slot per GPU
Rack viewRacks from your tags, standing under their datacenter, one slot per GPU, lit by rental, temperature, power, load or VRAM.
Bottlenecks card ranking what the fleet loses points on
Bottlenecks ProOld drivers, slow upload, low reliability, a price in the market's top quarter — ranked, with an offer score per host.
Upgrade simulator with the cumulative-profit chart
Upgrade simulator ProOther GPUs, more of them or more RAM — priced from the market, with the day it pays off.
Hardware lifecycle timeline per host
Hardware lifecycle ProPayback, warranty end, disk endurance and the fleet's book value over time.

Host agent Ultra

See inside every host. Fix it without SSH.

A small, dependency-free agent on each host reports what the Vast.ai API cannot: the container and Docker image of every rental, per-GPU temperature, power, throttling and ECC errors, and SMART health of every disk. Worn disks and overheating GPUs open a maintenance task on their own.

  • Remote Hands: diagnostics, speedtest, Docker cleanup, Vast daemon restart, reboot when idle
  • A closed allowlist compiled into the agent — no free-form commands, opt-in per host
  • Disk health: endurance used, TB written, projected life, thermal events
  • Signed self-updates; tokens hashed at rest and revocable
Per-GPU telemetry charts from the host agent
Remote Hands card with the host actions allowed on a machine

Mission Control Every plan

Put the fleet on the TV in the server room.

A full-screen live view: a card per machine with a tile per GPU lighting up as it is rented, an earnings ticker that moves every second, each datacenter's and rack's power against its circuit and the latest rentals and alerts. Share it through a revocable kiosk link — no login on the TV.

  • Sized to fit the whole fleet on one screen
  • Dark, light and ash themes, pinned per screen from the link
Mission Control wall with a tile per GPU and an earnings ticker

And the rest

Everything around it, done properly.

Your team, with roles

Invite people into your workspace as admin, member or read-only viewer; each signs in with their own login.

Datacenters and racks

Mark a tag as a datacenter or a rack (or use rack:A): racks stand under their datacenter, each with its own circuit and tariff.

Click to filter

Click a bar, a GPU model, a datacenter or a rack on the Inventory and the whole page follows; the filter lives in the URL.

Every chart exports

A table view, CSV and PNG export on every Charts and Inventory card, and a full-screen view.

Maintenance board

A kanban for hardware work. On Pro, announce the window to renters through Vast.ai in the cheapest hours.

Five themes

Light, dark, stone, ash and neon, a colour-blind-safe palette and a streamer mode that blurs your numbers.

API keys

Long-lived keys for your scripts, acting with their owner's role, each revocable on its own.

Installable app

Add HostDeck to your phone's home screen; the command palette is one Ctrl K away.

Tour

Every screen, in your theme.

Switch the theme at the top and the screenshots follow. Click one to open it full size.

Overview: earned so far today, Fleet Health, KPI tiles and what needs attention

earned so far today, Fleet Health, KPI tiles and what needs attention

How it works

From sign-up to your first alert in minutes.

  1. 1

    Connect Vast.ai

    Paste your Vast.ai API key. It is encrypted at rest and only used to read your machines — and to change prices when you explicitly allow it.

  2. 2

    Pick your machines

    Choose the hosts to monitor and where alerts go: Discord, Telegram or any Apprise channel. Import your earnings history from the Vast.ai CSV.

  3. 3

    Add the agent (optional)

    One command per host installs the agent for telemetry, disk health and Remote Hands. Python 3 standard library only.

    curl -fsSL …/agent/install.sh | sudo bash -s -- --token …

Pricing

Start with everything. Keep what you need.

Every account starts with 14 days of Ultra — no card. Then pick the plan that covers your GPUs. Cancel or switch any time.

Starter

Monitoring and alerts for a small fleet.

$7/ month

billed monthly

  • Up to 8 GPUs
  • 7 days of telemetry
  • 1 year of earnings
  • Checks every 5 min
  • 2 people
  • Live overview, machine pages and rental sessions
  • Charts: rental swimlane, 24×7 clock, replay, ranking
  • Earnings analytics, CSV import and forecast
  • Alerts for offline, errors and vanished hosts, plus rules
  • Email, Discord, Telegram and Apprise alerts, weekly digest
  • Inventory: host table, GPU models and hardware specs
  • Your price against the market median
  • Mission Control TV wall, maintenance board, tags
  • Maintenance windows announced to renters via Vast.ai
Start free trial

Ultra

Run the server room without SSH.

$39/ month

billed monthly

  • Up to 128 GPUs
  • 90 days of telemetry
  • 5 years of earnings
  • Checks every 30 s
  • 15 people
  • Everything in Pro, plus:
  • Host agent: container and image of every rental
  • GPU telemetry: temperature, power, throttling
  • Disk health and predictive maintenance tasks
  • Energy measured per GPU, circuit load per rack
  • Remote Hands: diagnostics, cleanup, daemon restart, reboot when idle
  • Machine ID backups
Start free trial

Prices in USD. The trial is Ultra for 14 days; nothing is charged until you choose a plan. Email, Discord and Telegram alerts are in every plan. Your price is locked: a subscription keeps the price it started at for as long as it runs — our first customers still pay their founder prices.

Compare every feature
FeatureStarterProUltra
Plan
GPUs Up to 8 GPUs Up to 32 GPUs Up to 128 GPUs
Telemetry 7 days of telemetry 31 days of telemetry 90 days of telemetry
Earnings history 1 year of earnings 2 years of earnings 5 years of earnings
Checks Checks every 5 min Checks every 1 min Checks every 30 s
People 2 people 5 people 15 people
Monitoring
Live overview and machine pages (SSE) ✓ ✓ ✓
Rental sessions: start, pause, resume, end, per GPU ✓ ✓ ✓
Charts: swimlane, 24×7 clock, ranking, replay ✓ ✓ ✓
Offline, error and vanished-host alerts ✓ ✓ ✓
Conditional alert rules ✓ ✓ ✓
Email, Discord, Telegram and Apprise channels ✓ ✓ ✓
Weekly fleet digest ✓ ✓ ✓
Team members with roles ✓ ✓ ✓
Earnings
Earnings analytics and CSV import ✓ ✓ ✓
Earnings forecast with what-if ✓ ✓ ✓
Earnings Autopsy — ✓ ✓
Pricing and profit
Your price against the market median ✓ ✓ ✓
Pricing suggestions against the market band — ✓ ✓
Pricing Autopilot (beta) — ✓ ✓
Profitability: tariffs, net profit, payback — ✓ ✓
Insights: anomalies and fleet health — ✓ ✓
Fleet
Inventory: host table, GPU models, hardware specs ✓ ✓ ✓
Server room rack view and fleet map — ✓ ✓
Side-by-side host comparison — ✓ ✓
Maintenance board ✓ ✓ ✓
Bottlenecks and hardware economics — ✓ ✓
Upgrade simulator and hardware lifecycle — ✓ ✓
Energy now, circuits, tariff simulation and fleet history — ✓ ✓
Mission Control TV wall ✓ ✓ ✓
Maintenance windows via Vast.ai ✓ ✓ ✓
Host agent
Container and image per rental — — ✓
GPU telemetry and disk health — — ✓
Predictive maintenance tasks — — ✓
Remote Hands — — ✓
Machine ID backups — — ✓

FAQ

Questions hosts ask us.

What happens when the trial ends?

For 14 days you get everything in Ultra. When the trial ends, monitoring and alerts pause until you choose Starter, Pro or Ultra — your machines, notification targets, settings and earnings history stay, and the last 7 days of telemetry are kept. Subscribe during the trial and nothing changes on the day it ends.

Do I need a credit card to start?

No. You only enter payment details, through Stripe, when you choose a plan.

Will HostDeck change my prices on its own?

Only if you ask it to. Nothing is written to Vast.ai until you tick Allow price changes for a machine. Autopilot policies start in dry run, so you see what they would do before they do it, and every change can be rolled back.

What does the host agent do, and is it safe?

It reads container metadata from Docker, GPU telemetry from nvidia-smi and disk health from smartctl, and reports them to your account. Remote Hands actions are a closed list compiled into the agent — there is no way to run an arbitrary command — and each host has to opt in. Rebooting additionally needs its own flag and waits until no rental is running.

What happens to my Vast.ai API key?

It is encrypted at rest and only used from our servers to read your machines, earnings and the market — plus the price and maintenance changes you explicitly start or allow.

Can I switch plans later?

Yes, any time from Settings → Billing. Pay for 6 months up front and one of them is free.

Will my price go up?

Not while your subscription runs. Every subscription keeps the price it started at: the hosts who joined before today's prices still pay their founder prices, with the limits their plan had then. A new price only applies to a new subscription.

Which plan fits my fleet?

Plans count GPUs, not hosts: Starter covers up to 8 GPUs, Pro up to 32 and Ultra up to 128 — one 8-GPU server or eight single-GPU hosts are the same 8 GPUs. When your trial ends, the plan picker marks the plan that covers the hosts you set up.

Do you support Polish energy tariffs?

Yes. G11, G12, G12w (the cheap weekend from Friday 22:00 to Monday 06:00, and statutory holidays) and G13 (its afternoon peak moving with the season) come as presets on the Profitability page, each datacenter can be on a tariff of its own, and the Energy card simulates what a datacenter would have paid on another one. Prices are entered in USD, the currency Vast.ai pays in.

Your fleet already earns. Make it earn on purpose.

14 days of Ultra, free. No card, set up in minutes.

Start free trial