Skip to main content
Your cloud bill tells you what you spent last month, not which cluster or team caused it or what you could turn off tonight. Cloud Cost prices every cluster from the machines and volumes Ankra already sees, splits the money down to namespaces and Stacks, and points at the idle capacity worth cutting. It needs no extra cost tool.
All figures are estimates built from list prices for the resources Ankra can see. Use them for trends, right-sizing and chargeback, not for reconciling an invoice to the cent.
Closed beta. The portal’s Cost area (Cost in the sidebar, and the Cost page in each cluster’s menu) is in closed beta and enabled per organisation on request. Contact support to have it turned on for your organisation. Every organisation reads the same figures, sets budgets and rate cards with ankra cost, the API and Ankra AI, as shown on this page.

Get a number for every cluster

Ankra prices a cluster once two things are in place:
  1. A price source for its capacity. For AWS, Google Cloud or Azure, a cloud credential for that provider. Hetzner, OVHcloud, UpCloud, DigitalOcean, Scaleway and Ankra Cloud clusters need no credential of yours. For Proxmox VE, VMware, HPE Morpheus or your own hardware, a rate card.
  2. The cluster online, so the agent can sync its nodes and volumes.
Until then, ankra cost cluster my-cluster says what is missing, such as no credential, no rate card, cluster offline or pricing still resolving. A cluster that cannot be priced is shown as not priced, never as free. The state values are listed in the Cost API reference.

Where the prices come from

Hetzner. A server ordered before the 2026-06-15 price increase is priced from the earlier list. When Ankra cannot read Hetzner’s live list, it uses its built-in copy of the list and reports medium confidence instead of high. A server that ran for only part of the month reads slightly low, and servers outside the cluster, such as CI runners in the same project, are not in any cluster’s estimate.Ankra Cloud. When Ankra cannot read the price list, it keeps the last prices it read and lowers the confidence. Until it has read them once, the clusters read as pricing still being collected, never as free.

Private capacity: rate cards

Proxmox, VMware, Morpheus and imported hardware have no published price, so you state what the capacity costs. A rate card is three monthly rates in your own currency: per vCPU, per GB of memory and per GB of provisioned storage. Ankra multiplies them by what each cluster holds. The card lives on the credential that provisions the capacity, so every cluster built from one Proxmox datacentre inherits the same rates. Set it with the CLI; all three rates are required, because the whole card is replaced:
  • Proxmox ships with defaults of EUR 3.00 per vCPU, 0.75 per GB of memory and 0.02 per GB of storage per month. Setting a card makes the rates your own; ankra cost rate-cards delete homelab-proxmox returns the credential to the defaults.
  • Other private providers have no defaults, so their clusters stay not priced until you set rates.
  • Imported clusters with no provider credential use the organisation’s own rate card, if you set one: ankra cost rate-cards set organisation --currency EUR --vcpu 5 --memory-gb 1.5 --storage-gb 0.05. With the Cost area turned on, Cost → Settings also sets these organisation rates.
Changing a rate card needs the billing.manage permission. A rate change applies from the next hourly pass. It does not rewrite past figures.

Attribute spend to teams

Every hour Ankra allocates each cluster’s cost to its namespaces by their pods’ share of the CPU and memory requested on the cluster’s priced nodes, and joins that allocation to names you recognise: the Stack a namespace belongs to, the add-ons in it and the applications installed into it. That gives you a chargeback view per team.
ankra cost object credential reads spend the way the bill arrives: by the account a credential holds and the clusters it built.

By project

Every cost read is per project. A read shows the project you are working in, and the organisation-wide view carries a By project breakdown: each project’s run rate and cluster count, with an unattributed bucket for spend no project can claim, such as waste found on an organisation-wide credential rather than a cluster, so the projects reconcile to the fleet total only together with that bucket. On the API, a read without a project is your current project; project_id=all on GET /api/v1/org/cloud-cost/summary and the other cost routes reads the whole organisation, and a project id reads that project. The organisation-wide summary carries by_project, and every response carries a scope block saying which project it describes and whether it is a floor because some clusters are hidden from you. Cost is stamped with the cluster’s project every hour, so a cluster that moves projects mid-month shows up in both, each for the hours it was there, and a project’s history is not rewritten by a move.

See what changed

ankra cost trend draws the organisation’s run rate per day over the last 30 days (--days up to 34). When the set of priced clusters changes - a cluster is created, deleted, stopped or loses its pricing - the line moves without anything being spent, and those days are listed apart as coverage moves. ankra cost events lists what moved the run rate, newest first: application releases, node-count changes, stops and starts, executed cost decisions, clusters entering or leaving pricing, and resolved waste. Find out why your bill went up walks through reading them.

Cut waste

ankra cost savings reads the savings model, each recommendation with an estimated monthly saving and the basis it rests on:
  • Right-size idle capacity. When a large share of a cluster’s run rate sits idle.
  • Stop non-production clusters off-hours. A cluster with no enabled power schedule is offered weeknight and weekend stops only when a person chose its environment label, the label does not say production, and both its ingress traffic and its own workloads were measured as quiet in the stop window. The label a new cluster gets by default does not count as chosen.
  • Reduce unallocated spend. Raised when a quarter or more of a cluster’s run rate is not allocated to any workload.
The total counts each cluster once, at its best lever, and counts only measured and billed savings; estimates are shown beside it. Approving and running a change goes through cost decisions, and what each change saved is in ankra cost ledger.

Orphaned and idle resources

A waste scan runs every six hours and lists resources that bill while doing no work: Each finding shows its monthly cost, or “price unknown”. The provider-side scan covers Hetzner accounts; the released-volume check covers every cluster. A credential that could not be listed is shown as a partial scan, not as “no waste”. ankra cost waste resolve deletes or releases the cloud resource a finding names: a stopped server gets a snapshot kept for seven days before it is deleted, and an unattached volume is deleted only with written consent (--consent). ankra cost waste resolved totals what the cleanup resolved, by kind; see the ankra cost reference for what each needs.

What if the node groups changed?

ankra cost estimate prices a change before you make it: change a node group’s count, resize or add a group, or add a load balancer, and it returns today’s bill, the predicted bill and the change. Nothing is saved or applied.
The prediction is today’s bill plus the priced difference between the draft and today, so only what you changed moves the number. A resized group lands on the cheapest published size that holds its vCPU, memory and GPUs, keeping its CPU architecture, and the estimate names the size. Private capacity is priced from the cluster’s rate card. Name another provider to price the same draft there, including from private capacity. A group with GPUs is placed only on a GPU size and keeps its card where a size carrying it fits; a group without GPUs is never placed on one. On UpCloud every GPU plan in the price book is a size, and a GPU group marked spot lands on the matching spot plan. Where a provider’s price book records no GPU sizes, the group reads Not priced instead of taking a CPU size of the same shape. A group nothing can price reads Not priced with the reason. The request and response are in What-if estimate.

Compare providers

A comparison prices the whole bill (node groups, the control plane Ankra provisioned, volumes and load balancers) on every provider Ankra can build on, side by side. With the Cost area turned on, it is the Compare options tab.
  • DigitalOcean, Hetzner, OVHcloud, Scaleway and UpCloud price from Ankra’s list-price book.
  • AWS, Google Cloud and Azure price from their per-region catalogues, and each column names the region it used. Which sizes each hyperscaler offers.
  • Proxmox VE and HPE Morpheus price per unit of capacity from a rate card, and the column says which card.
Matching decides how sizes are chosen: Like-for-like capacity keeps each group on the same vCPU class, shared or dedicated, where the target sells one (the class of each provider’s sizes is in the reference); Explore resizing and cheaper classes ignores class. A node group with GPUs is priced only where the target’s price book records GPU sizes. A target with an unpriced line shows a “from” figure, names what is missing, and never sorts as cheapest. Latency, storage class, availability zones, load-balancer features, egress terms and negotiated discounts are not priced. Weigh them next to the number.

Settings

Organisation admins set the defaults every figure uses:
  • Currency - the organisation default: EUR (the default), USD, GBP, SEK, NOK, DKK, CHF, PLN, CZK, JPY, CAD, AUD or INR.
  • Effective discount - a flat 0-100% discount on list prices, to approximate committed-use or enterprise agreements.
  • Network egress estimate - stored, not metered: no figure includes network egress, whichever way it is set.
With the Cost area turned on, the same settings are on Cost → Settings, beside Clusters and how they are priced and Cost automation.

In the portal (closed beta)

With the Cost area turned on, Cost in the sidebar has six tabs: Each cluster’s menu then has a Cost page: How the monthly rate is built, Servers in this cluster, Cost by namespace and What Ankra found on the cluster.

From the CLI and Ankra AI

Each read is your current project. The global --project <name> reads another project; --all-projects on cost summary, savings, ledger, trend, events, reconcile and cost budgets list reads the whole organisation, and cost summary then shows the By project table; ankra cost object project <name> prices one project:
Every command takes -o json. See the ankra cost reference for every command, and the Cost API reference for field meanings. In chat and over MCP, Ankra AI reads the same figures with get_fleet_cloud_cost, get_cluster_cost, get_stack_cost, get_cloud_savings and get_cost_settings. Changes need an mcp:write token and the same permissions as everywhere else.

Budgets and the cost autopilot

A budget is a monthly amount for the organisation, one project, one cluster, one environment label or one application. It shows spend so far, the projected month end and the hour it would cross the budget. Ankra raises a Budget crossing notification when the projection reaches the notice threshold (80% unless you set another); route it with notification routing. Ankra also emails the budget’s owner, or the organisation’s admins when the budget has no owner or the owner is no longer an active member, when the month first reaches the threshold, again when it is projected past the budget, and again when it is spent past it, each at most once per budget per month. Anyone who muted the notification for that scope is left out. The cost autopilot decides how much Ankra may do about cost without asking. Each environment kind (production, staging, development, preview, unknown) takes a tier: hands-off, scheduled, managed or ephemeral. An organisation that never set a policy is hands-off everywhere, so every change waits for a person. You can override one cluster’s tier and set daily quiet hours, and give a project its own defaults, which sit between the organisation’s policy and a cluster’s override.
Every member can read budgets and the autopilot policy. Changing a budget needs the billing.manage permission; changing the autopilot also needs clusters.write. Ankra AI uses get_cost_budgets, set_cost_budget, get_cost_autopilot and update_cost_autopilot. With the Cost area turned on, budgets are on Cost → Budgets & alerts and the autopilot on Cost → Settings, under Cost automation. Cost budgets and the autopilot walks through budgets, autopilot tiers, approving decisions and reading the ledger step by step. Next: stop idle clusters on a schedule with power schedules.