> ## Documentation Index > Fetch the complete documentation index at: https://docs.ankra.ai/llms.txt > Use this file to discover all available pages before exploring further. # Savings model for the organisation > The organisation's savings model: right-size and off-hours recommendations across the biggest priced clusters, the cost breakdowns they were read from, the clusters the fleet cannot price, and the waste scan, in the caller's display currency (`ankra cost savings`). One cluster counts once in the total, at its best lever. Bearer-PAT twin of the browser route of the same name under /org. ## OpenAPI ````yaml https://platform.ankra.app/openapi.json get /api/v1/org/cloud-cost/savings openapi: 3.1.0 info: title: Ankra Platform API version: v1 description: >- The Ankra platform HTTP contract. The /api/v1 surface is the public, bearer-authenticated API used by the Ankra CLI, MCP clients and automation; send a personal or service API token as `Authorization: Bearer `. Browser routes under /org and /chat are served on api.platform.ankra.app (each declares that server); they and /oauth/access authenticate with the ankra_session cookie, and mutations also require the X-Ankra-CSRF header. Every operation declares the credential it accepts, and no operation is served without one. This document is published deliberately: it describes the contract only and carries no organisation data or credentials. Report security issues to security@ankra.ai (see /.well-known/security.txt). servers: - url: https://platform.ankra.app security: - BearerAuth: [] tags: - name: Onboarding description: >- Generate a first website from a prompt and deploy it to a new organisation's cluster. - name: Access Gate description: >- Private service sign-in and membership checks for trusted Ankra proxy relays. - name: Canvas Datasets description: >- Read the fixed, named datasets canvas dashboards bind their charts, tables and stats to. - name: Canvases description: >- Keep dashboards and pages in the organisation's canvas library, with revision history, undo and pins. - name: Canvas Pages description: >- Publish a chat canvas artifact at a public pages address and take it down again. - name: Cloud Delegation description: >- Platform browser consent and organisation-bound connections for Ankra Cloud. - name: Organisation IAM description: >- Custom organisation roles and scoped role assignments for member and service-account identities. - name: Clusters description: Create, inspect and manage clusters, and the stacks deployed on them. - name: Managed Clusters description: Provider-managed control planes, driven through one common surface. - name: Imported Clusters description: Clusters that already existed and were connected to Ankra. - name: Cluster Access description: Kubeconfigs, service-account tokens and per-cluster access grants. - name: Kubernetes description: Read and act on the Kubernetes objects inside a cluster. - name: AWS Clusters description: >- Provision and manage self-managed k3s / kubeadm clusters on AWS EC2 in your own VPC. - name: Ankra Cloud Clusters description: >- Provision and manage self-managed kubeadm / k3s clusters on Ankra Cloud servers: a private network, a NAT router and a bastion per cluster. - name: DigitalOcean Clusters description: Provision and manage DigitalOcean Kubernetes clusters. - name: Hetzner Clusters description: Provision and manage Hetzner Kubernetes clusters. - name: OVH Clusters description: Provision and manage OVH Kubernetes clusters. - name: Scaleway Clusters description: Provision and manage Scaleway Kapsule clusters. - name: UpCloud Clusters description: Provision and manage UpCloud Kubernetes clusters. - name: Applications description: Deploy, configure and observe applications across the fleet. - name: Pipelines description: >- Pipeline definitions, the approval of the authority they declare, and the caches pipelines keep between runs. - name: Workspaces description: >- Personal workspaces on the organisation's CI pool where a repository's heavy commands run, and the repository profiles they are built from. - name: Deploy Targets description: >- Environments, the hosts registered to receive deployments, and the releases deployed to them. - name: Deploy Provenance description: >- Whether an application's deploys verify signed provenance from the organisation's own pipeline, the keys that provenance is signed with, and a dry-run that verifies images without deploying. - name: Backups description: >- Backup vaults, restore points, protection posture and captures for stacks and application deployments; a completed capture is not a verified restore. - name: Stack Profiles description: Reusable stack definitions, their versions and sharing. - name: Services description: >- Versioned service packages and explicit sharing. Runtime admission is separate from publication. - name: Charts description: Browse the chart catalogue behind stacks and addons. - name: Helm description: Helm registries, credentials and the charts they expose. - name: Executions description: Long-running platform executions and their jobs. - name: Operations description: Cancel in-flight cluster operations and their jobs. - name: Chat description: Conversational sessions, plans and confirmable actions. - name: AI Management description: >- Customer agent lifecycle, authenticated identity and organisation automation controls. - name: AI Agent Runs description: Autonomous agent runs and their outcomes. - name: AI Tickets description: The AI ticket board, its sync connections and settings. - name: AI Playbooks description: Reusable playbooks the AI lanes execute. - name: AI Conditions description: Conditions that gate AI autonomy. - name: AI Remediation description: >- The organisation's auto-remediation policy: what the AI lanes may fix by themselves, and who approves the rest. - name: AI Engineering Handoffs description: Work the AI lanes escalate to a human engineer. - name: AI Environment description: The environment and base stacks AI demos deploy into. - name: AI Memory description: >- What the assistant has learned about you: your cross-session memories and learned profile. - name: MCP Servers description: >- Register the MCP servers the organisation's AI agents can call, choose their tools and grant them to roles. - name: Security description: Findings, advisories, SBOMs, compliance and posture. - name: Cost description: Cluster and fleet cost, rate cards and cost settings. - name: Decisions description: >- The decision ledger behind the Security and Cost queues: proposals a surface computed, the approve and set-aside decisions people took on them, and the receipts of running them. - name: Usage description: >- What your organisation used of Ankra - vCPU-hours, playgrounds, managed services, hosted logs and metrics, and AI - per meter and period, with how completely it was measured and whether a published rate applies. - name: Billing description: Subscription and spend caps. - name: Organisation description: Members, invitations, audit logs and organisation settings. - name: Account description: >- Your own account: the contact email Ankra mails when it differs from your sign-in address. - name: Account Tokens description: Personal access tokens for the API and CLI. - name: Credentials description: The shared credential store. - name: AWS Credentials description: >- AWS credentials: access keys or CloudFormation-onboarded STS roles for cost, EKS and self-managed provisioning. - name: Ankra Cloud Credentials description: >- Ankra Cloud API tokens, shared by the self-managed and managed Ankra Cloud lanes. - name: Azure Credentials description: Azure credentials and SSH keys. - name: DigitalOcean Credentials description: DigitalOcean credentials and SSH keys. - name: Hetzner Credentials description: Hetzner credentials and SSH keys. - name: HPE Morpheus Credentials description: HPE Morpheus credentials and the capacity and quotas behind them. - name: OVH Credentials description: OVH credentials and SSH keys. - name: Proxmox Credentials description: Proxmox VE credentials and the capacity and quotas behind them. - name: Scaleway Credentials description: Scaleway credentials. - name: UpCloud Credentials description: UpCloud credentials and SSH keys. - name: Data Source Credentials description: Credentials for metrics and log sources. - name: DNS Credentials description: Credentials for DNS providers. - name: Object Storage Buckets description: >- Buckets Ankra creates and manages on an organisation's own provider credentials. - name: DNS description: DNS zones and records, including custom organisation zones. - name: Cloudflare description: Cloudflare domains and the credentials behind them. - name: Avura description: >- The linked Avura organization, the domains it shares with Ankra Platform, and their DNS records. - name: Variables description: Organisation- and cluster-scoped variables. - name: Secret Slots description: >- Store secrets in Vault and reference them by ${SECRET_SLOT:} sentinel instead of by value. - name: SOPS description: Encrypt and decrypt values with the organisation SOPS config. - name: Alerts description: Alert integrations and ingest credentials. - name: Notifications description: Notification routes and their delivery targets. - name: Support description: Support tickets. - name: AI Settings description: Organisation AI provider, model catalog and per-function model settings paths: /api/v1/org/cloud-cost/savings: get: tags: - Cost summary: Savings model for the organisation description: >- The organisation's savings model: right-size and off-hours recommendations across the biggest priced clusters, the cost breakdowns they were read from, the clusters the fleet cannot price, and the waste scan, in the caller's display currency (`ankra cost savings`). One cluster counts once in the total, at its best lever. Bearer-PAT twin of the browser route of the same name under /org. operationId: cloud_savings_api_v1_org_cloud_cost_savings_get parameters: - description: PAT organisation override. in: header name: x-ankra-organisation-id required: false schema: type: string responses: '200': content: application/json: schema: $ref: '#/components/schemas/CloudSavingsResponse' description: Successful response '400': content: application/json: schema: $ref: '#/components/schemas/ContractDetailError' description: >- Missing or malformed Authorization header, or business validation error '401': content: application/json: schema: $ref: '#/components/schemas/ContractDetailError' description: Unknown, expired, or revoked API token '403': content: application/json: schema: $ref: '#/components/schemas/ContractDetailError' description: >- Token may not use this surface (MCP-scoped token or service account restriction) '422': content: application/json: schema: $ref: '#/components/schemas/HTTPValidationError' description: Request validation error '500': content: application/json: schema: $ref: '#/components/schemas/ContractDetailError' description: Internal server error security: - BearerAuth: [] components: schemas: CloudSavingsResponse: description: >- The organisation's savings model. Each cluster is counted once, at its best lever, because idle spend sits inside what an off-hours stop reclaims: total_monthly_savings_cents is each cluster's best measured or billed lever, and estimated_monthly_savings_cents only what its best lever of any confidence adds above that, so the two never count one spend twice. Unanalysed clusters are priced clusters beyond the organisation's analysed limit or past the analysis time budget, and offline clusters are priced clusters whose agent has not checked in; neither is a cluster with nothing to save. properties: analysed_cluster_count: type: integer analysis_budget_exhausted: description: >- True when the analysis time budget (thresholds.analysis_budget_seconds), not the analysed-cluster limit, left some priced clusters unanalysed. type: boolean breakdowns: items: $ref: '#/components/schemas/CloudSavingsBreakdown' type: array currency: type: string estimated_monthly_savings_cents: description: >- What the clusters' best levers add, on estimates alone, above total_monthly_savings_cents: summed per cluster as its best lever of any confidence less its best measured or billed lever. Never counted in total_monthly_savings_cents. type: integer generated_at: type: string format: date-time namespaces: description: >- The costliest namespaces across the analysed clusters, at most twelve. items: $ref: '#/components/schemas/CloudSavingsNamespace' type: array priced_cluster_count: type: integer recommendations: items: $ref: '#/components/schemas/CloudSavingsRecommendation' type: array thresholds: $ref: '#/components/schemas/CloudSavingsThresholds' total_monthly_savings_cents: type: integer unanalysed_cluster_count: type: integer unanalysed_clusters: description: >- Every priced cluster unanalysed_cluster_count counts, costliest first, with why it was left out on this read: limit (past the organisation's analysed-cluster limit), budget (the analysis time budget ran out first), unreadable (its cost read failed; also in unreadable_clusters) or no_breakdown (its cost read had no priced breakdown). The analysed set is re-chosen on every read, so this is what explains a total that moved without anything being spent or saved. items: $ref: '#/components/schemas/CloudSavingsUnanalysedCluster' type: array unpriced_cluster_count: type: integer unpriced_clusters: items: $ref: '#/components/schemas/CloudSavingsUnpricedCluster' type: array stale_cluster_count: type: integer stale_clusters: description: >- Clusters priced before whose latest snapshot is older than the fleet's staleness window: metering has stalled, which is neither priced nor unpriced. items: $ref: '#/components/schemas/CloudSavingsUnpricedCluster' type: array unreadable_clusters: description: >- Priced clusters whose cost breakdown could not be read on this pass; counted as unanalysed, never as clusters with nothing to save. items: $ref: '#/components/schemas/CloudSavingsUnreadableCluster' type: array offline_clusters: description: >- Priced clusters whose agent has not checked in within the platform's check-in threshold. Their synced inventory is the last one the agent sent, so they keep their price and breakdown but are offered no recommendation of any kind; never clusters with nothing to save. items: $ref: '#/components/schemas/CloudSavingsOfflineCluster' type: array waste: $ref: '#/components/schemas/CloudSavingsWaste' withheld: description: >- Every schedulable, priced cluster worth parking that was not offered an off_hours_schedule, with the first failing reason: purpose_unconfirmed (no person chose its environment label), production (its label or stored criticality says production), off_hours_traffic (it served verified ingress requests inside the stop window), off_hours_workload (its own workloads were busy there) or activity_unmeasured (either activity signal could not be read, which is unknown, not idle). And every right-size that offers no step every plan-time check passed, as kind right_size with no money figure: no_safe_step (usage was measured and no smaller size passes; detail carries the ladder notes), does_not_fit (a pod, or the group's requests with one node down, do not fit the next size), peaks_exceed (the busiest node's 30-day CPU max or p99 is at or over the next size), no_headroom (the group's p95 CPU with one node down would be at or over half the next size's capacity), inventory_stale (the nodes, pods, budgets and volumes the fit check needs could not be read fresh, so it failed closed) or metrics_missing (no usage could be read). Never clusters with nothing to save. items: $ref: '#/components/schemas/CloudSavingsWithheld' type: array required: - currency - generated_at - total_monthly_savings_cents - estimated_monthly_savings_cents - recommendations - breakdowns - namespaces - analysed_cluster_count - unanalysed_cluster_count - unanalysed_clusters - analysis_budget_exhausted - priced_cluster_count - unpriced_cluster_count - unpriced_clusters - stale_cluster_count - stale_clusters - unreadable_clusters - offline_clusters - waste - thresholds - withheld type: object ContractDetailError: example: detail: Cluster not found properties: detail: oneOf: - type: string - additionalProperties: true properties: {} type: object required: - detail type: object HTTPValidationError: properties: detail: items: $ref: '#/components/schemas/ValidationError' title: Detail type: array title: HTTPValidationError type: object CloudSavingsBreakdown: description: >- One analysed cluster's run rate split by billed component, with the idle and unallocated shares the recommendations read, the id of its best recommendation, its costliest namespaces, and its daily cost trend over the last thirty days (the same series GET /org/clusters/{cluster_id}/cost serves as trend), so the fleet's movement can be attributed to the clusters that moved it from this one read. properties: best_recommendation_id: anyOf: - type: string - type: 'null' cluster_id: type: string format: uuid cluster_name: type: string components: items: $ref: '#/components/schemas/CloudSavingsComponent' type: array idle_monthly_cents: description: >- The cluster's unrequested capacity, not unused capacity: the monthly compute cost of allocatable capacity no pod requests, of the binding resource per node group, with control-plane nodes and autoscaled or resized node groups left out, averaged over the last seven days of hourly snapshots. type: integer monthly_cents: type: integer provider: type: string top_namespaces: items: $ref: '#/components/schemas/CloudSavingsNamespace' type: array trend: description: >- The cluster's daily run rate over the last thirty days, one point per day, carried verbatim from its cost read; always present on an analysed cluster. items: $ref: '#/components/schemas/CostTrendPoint' type: array unallocated_monthly_cents: type: integer required: - cluster_id - cluster_name - provider - monthly_cents - components - idle_monthly_cents - unallocated_monthly_cents - best_recommendation_id - top_namespaces - trend type: object CloudSavingsNamespace: description: >- One namespace's allocated share of its cluster's run rate: the dependency cost a lane drills into. properties: cluster_id: type: string format: uuid cluster_name: type: string monthly_cents: type: integer namespace: type: string share_percent: type: integer required: - cluster_id - cluster_name - namespace - monthly_cents - share_percent type: object CloudSavingsRecommendation: description: >- One lever on one cluster. id is stable across reads (kind and cluster), so a ledger can reference it. properties: cluster_id: type: string format: uuid cluster_name: type: string environment: anyOf: - type: string - type: 'null' confidence: description: >- What monthly_savings_cents rests on. measured: a right_size_idle, whose usage the metrics store answered for and planned at least one ladder step that passed the plan-time fit and sizing checks (a right-size that offers no step is in withheld, never a recommendation). billed: an off_hours_schedule taken on the cluster's billed compute. estimated: inferred from capacity less requests or hedged on the whole run rate with nothing observed behind it (every reduce_unallocated, and an off_hours_schedule on a cluster whose breakdown was not read). Only measured and billed savings without plan_required count towards total_monthly_savings_cents. enum: - measured - billed - estimated type: string evidence: $ref: '#/components/schemas/CloudSavingsEvidence' id: type: string kind: enum: - right_size_idle - reduce_unallocated - off_hours_schedule type: string monthly_cost_cents: type: integer monthly_savings_cents: type: integer provider: type: string share_percent: type: integer ladder: anyOf: - items: $ref: '#/components/schemas/CloudSavingsLadderStep' type: array - type: 'null' description: >- A right_size_idle recommendation's full step plan: each node group onto the next smaller priced size of the same vCPU class and architecture, one wave per step, priced per step from the cluster's own price book, stopping where the group's p95 would lose its headroom or approach the verification limits. Null when no step could be planned; ladder_notes says why for each group left out. total_ladder_savings_cents: anyOf: - type: integer - type: 'null' description: >- The sum of the ladder's steps, never the unrequested-capacity ceiling; null when there is no ladder. ladder_notes: anyOf: - items: $ref: '#/components/schemas/CloudSavingsLadderNote' type: array - type: 'null' description: >- Why each node group left out of the ladder was left out (spiky, busy, silent, unpriced, spot, or bounded); a note with no group is about the whole cluster. autopilot_tier: enum: - hands-off - scheduled - managed - ephemeral type: string description: >- The cost autopilot tier this recommendation's change falls into at this read: the cluster's tier, except that an off_hours_schedule whose evidence.node_local_storage_state is not none is hands-off (proposal-only) whatever the cluster's tier. autopilot_tier_reason: description: >- Why the change is in that tier, as the card shows it: the cluster's sentence, followed for a hands-off off_hours_schedule by why it waits for a person (its stop deletes the worker servers, and claims keep data on their disks or the inventory could not tell). type: string plan_required: anyOf: - $ref: '#/components/schemas/CloudSavingsPlanRequired' - type: 'null' description: >- On a right_size_idle: non-null when the right-size is not one click - a person writes a maintenance plan (switchovers, order, window) first. Such a recommendation is in neither total_monthly_savings_cents nor estimated_monthly_savings_cents. Null for one-click right-sizes and the other kinds. preflight: anyOf: - $ref: '#/components/schemas/CloudSavingsRightSizePreflight' - type: 'null' description: >- On a right_size_idle: the plan-time fit check every offered step passed. A step it would refuse, or could not judge, is never offered; the right-size is then in withheld instead. Null for the other kinds. sizing: anyOf: - $ref: '#/components/schemas/CloudSavingsRightSizeSizing' - type: 'null' description: >- On a right_size_idle: the usage evidence the steps were sized on. Null for the other kinds. execution: anyOf: - $ref: '#/components/schemas/CloudSavingsRightSizeExecution' - type: 'null' description: >- On a right_size_idle: how the right-size runs. Null for the other kinds. required: - id - kind - cluster_id - cluster_name - provider - environment - monthly_savings_cents - monthly_cost_cents - share_percent - confidence - evidence - ladder - total_ladder_savings_cents - ladder_notes - autopilot_tier - autopilot_tier_reason - plan_required - preflight - sizing - execution type: object CloudSavingsThresholds: properties: analysed_cluster_limit: description: >- The organisation's analysed-cluster limit in effect (cost settings; 8 unless it set its own). type: integer analysis_budget_seconds: description: >- How long one savings read may spend on per-cluster breakdowns before the rest are counted as unanalysed. type: integer snapshot_stale_after_hours: description: >- How old a cluster's latest cost snapshot may be before its metering counts as stopped: such a cluster is named in stale_clusters and left out of every total. type: integer minimum_off_hours_monthly_cents: type: integer minimum_savings_cents: type: integer off_hours_share: type: number unallocated_share_threshold: type: number required: - minimum_savings_cents - minimum_off_hours_monthly_cents - unallocated_share_threshold - off_hours_share - analysed_cluster_limit - analysis_budget_seconds - snapshot_stale_after_hours type: object CloudSavingsUnanalysedCluster: description: >- A priced cluster the savings read did not analyse, and why: never a cluster with nothing to save. properties: cluster_id: type: string format: uuid cluster_name: type: string reason: enum: - limit - budget - unreadable - no_breakdown type: string required: - cluster_id - cluster_name - reason type: object CloudSavingsUnpricedCluster: properties: cluster_id: type: string format: uuid cluster_name: type: string kind: type: string required: - cluster_id - cluster_name - kind type: object CloudSavingsUnreadableCluster: properties: cluster_id: type: string format: uuid cluster_name: type: string required: - cluster_id - cluster_name type: object CloudSavingsOfflineCluster: description: >- A priced cluster whose agent has not checked in within the platform's check-in threshold: offered no recommendation, never read as having nothing to save. properties: cluster_id: type: string format: uuid cluster_name: type: string required: - cluster_id - cluster_name type: object CloudSavingsWaste: description: >- The open cloud-waste findings. available is false when the waste scan could not be read, which is unknown waste, not no waste. total_monthly_cost_cents and unpriced_finding_count cover the findings that count toward this organisation's savings: a resource several organisations' credentials see counts only in the organisation its Ankra cluster label names, and in none without one, so it is counted once between them. finding_count is every finding listed. properties: available: type: boolean finding_count: type: integer has_data: type: boolean scanned_at: anyOf: - type: string - type: 'null' total_monthly_cost_cents: type: integer unpriced_finding_count: type: integer shared_not_counted_monthly_cost_cents: type: integer description: >- The priced monthly cost of the listed findings whose resource other Ankra organisations also see and whose cost counts in another organisation or in none; it is not in total_monthly_cost_cents. shared_not_counted_finding_count: type: integer description: >- How many of the listed findings count in another organisation or in none, priced or not; none of them is in unpriced_finding_count. required: - available - has_data - scanned_at - total_monthly_cost_cents - finding_count - unpriced_finding_count - shared_not_counted_monthly_cost_cents - shared_not_counted_finding_count type: object CloudSavingsWithheld: description: >- A lever the savings model withheld, and why, with no money figure: an off_hours_schedule the purpose gate did not offer, or a right_size that offers no step every plan-time check passed. properties: busy_hours: description: >- Hours inside the stop window over window_days in which the cluster's own workloads were busy. Omitted when not measured. type: integer cluster_id: type: string format: uuid cluster_name: type: string detail: description: >- For a right_size, why in sentences: the planner's ladder notes, each prefixed with its node group. Omitted for an off_hours_schedule. items: type: string type: array kind: enum: - off_hours_schedule - right_size type: string reason: enum: - purpose_unconfirmed - production - off_hours_traffic - off_hours_workload - activity_unmeasured - no_safe_step - does_not_fit - peaks_exceed - no_headroom - inventory_stale - metrics_missing type: string requests: description: >- Verified ingress requests inside the stop window over window_days. Omitted when not measured. type: integer window_days: description: >- How many days back the activity signals looked. Omitted when the purpose withheld the schedule before any activity was read. type: integer required: - cluster_id - cluster_name - kind - reason type: object ValidationError: properties: loc: items: anyOf: - type: string - type: integer title: Location type: array msg: title: Message type: string type: title: Error Type type: string required: - loc - msg - type title: ValidationError type: object CloudSavingsComponent: properties: key: type: string label: type: string monthly_cents: type: integer required: - key - label - monthly_cents type: object CostTrendPoint: properties: day: type: string format: date-time monthly_cost_estimate_cents: type: integer fully_priced: type: boolean description: >- Every node and billed resource of the cluster was priced that day, so the run rate is the cluster's whole cost rather than a floor under it. A point exists only for a day the cluster was priced; a missing day is unknown, not zero. required: - day - monthly_cost_estimate_cents - fully_priced title: CostTrendPoint type: object CloudSavingsEvidence: description: >- The numbers a recommendation was computed from; members that do not apply to its kind are null. properties: idle_monthly_cents: anyOf: - type: integer - type: 'null' description: >- Unrequested capacity, not unused capacity: the monthly compute cost of allocatable capacity no pod requests (binding resource per node group, control planes and autoscaled or resized groups left out), averaged over seven days. Usage is never read into it; unrequested_basis names the measure. unrequested_basis: anyOf: - enum: - requests_vs_allocatable_7d_avg type: string - type: 'null' description: >- How idle_monthly_cents was measured: requests_vs_allocatable_7d_avg. Null where the evidence carries no idle figure. off_hours_share_percent: anyOf: - type: integer - type: 'null' start_cron: anyOf: - type: string - type: 'null' stop_cron: anyOf: - type: string - type: 'null' stoppable_monthly_cents: anyOf: - type: integer - type: 'null' unallocated_monthly_cents: anyOf: - type: integer - type: 'null' unallocated_share_percent: anyOf: - type: integer - type: 'null' metrics_state: anyOf: - enum: - present - absent - unavailable type: string - type: 'null' description: >- On a right_size_idle recommendation: whether the cluster's usage could be read from the platform metrics store (present), the store holds no node series for it so no group can be sized (absent: a prerequisite, not a saving), or the store is not configured or could not be read (unavailable: unknown). Null for the other kinds and for a cluster priced by rate card or by its own bill. node_local_storage_state: anyOf: - enum: - present - none - unknown type: string - type: 'null' description: >- On an off_hours_schedule recommendation: whether the cluster keeps persistent data on its worker nodes' own disks (local or hostPath volumes, or a local provisioner such as local-path, OpenEBS LocalPV or TopoLVM), read from the synced cluster inventory. The schedule's scale_to_zero stop deletes the worker servers and the start creates new ones, so that data is lost at every stop. present: at least one claim does; none: the inventory is synced and none does; unknown: the inventory holds no nodes for the cluster or could not be read. Only none says the cluster is safe, and neither other state is ever read as none: on present the decision ledger runs the schedule only with written consent; unknown is a warning shown to the person approving it. Null for the other kinds. node_local_pvc_count: anyOf: - type: integer - type: 'null' description: >- On an off_hours_schedule recommendation: how many persistent volume claims keep their data on a worker's disk. Null for the other kinds and when node_local_storage_state is unknown. node_local_pvc_names: anyOf: - type: array items: type: string - type: 'null' description: >- On an off_hours_schedule recommendation: up to 25 of those claims as namespace/name, sorted. Null for the other kinds and when node_local_storage_state is unknown. environment_source: anyOf: - enum: - person - default - unknown type: string - type: 'null' description: >- On an off_hours_schedule recommendation: how the cluster's environment label was set. Always person on an offered schedule: a schedule is offered only for a cluster whose purpose a person confirmed (see withheld). Null for the other kinds. environment_set_by: anyOf: - type: string - type: 'null' description: >- On an off_hours_schedule recommendation: who chose the environment label (email, or the auth subject when the user has none), when known. Null for the other kinds. environment_set_at: anyOf: - type: string format: date-time - type: 'null' description: >- On an off_hours_schedule recommendation: when the environment label was chosen. Null for the other kinds. off_hours_window_days: anyOf: - type: integer - type: 'null' description: >- On an off_hours_schedule recommendation: how many days back the activity signals looked (14). Null for the other kinds. off_hours_requests: anyOf: - type: integer - type: 'null' description: >- On an off_hours_schedule recommendation: verified ingress requests the cluster served inside the would-be stop window (weeknights 19:00-07:00 and weekends, in the schedule's timezone) over off_hours_window_days. An offered schedule is below the gate's small threshold, since an uptime probe counts as use. Null for the other kinds. off_hours_busy_hours: anyOf: - type: integer - type: 'null' description: >- On an off_hours_schedule recommendation: hours inside the would-be stop window over off_hours_window_days in which the cluster's own workloads (every namespace outside the platform's) used at least a quarter of a CPU core on average. Null for the other kinds. type: object CloudSavingsLadderStep: properties: wave: type: integer group: type: string from_instance_type: type: string to_instance_type: type: string count: type: integer monthly_savings_cents: type: integer cumulative_monthly_savings_cents: type: integer verify_days: type: integer required: - wave - group - from_instance_type - to_instance_type - count - monthly_savings_cents - cumulative_monthly_savings_cents - verify_days title: CloudSavingsLadderStep type: object CloudSavingsLadderNote: properties: group: type: string reason: type: string code: description: >- The withheld reason the note stands for when a safety rule stopped the group (does_not_fit, peaks_exceed, no_headroom, inventory_stale, metrics_missing); omitted for a note that is no safety refusal. type: string required: - group - reason title: CloudSavingsLadderNote type: object CloudSavingsPlanRequired: description: >- Why a recommendation needs a written maintenance plan before it can run. The decision ledger recomputes the same reasons from the cluster as it is when an approved right-size runs, and refuses it on any of them. properties: reasons: items: enum: - production - purpose_unconfirmed - stateful_volumes - zero_disruption_pdb - single_instance_database - cross_plan_family type: string type: array detail: type: array items: type: string required: - reasons - detail type: object CloudSavingsRightSizePreflight: description: >- The plan-time fit check of an offered right-size, read from a fresh inventory: every pod fits one new node beside its DaemonSets, and the group's requests stay at or under 90% of what the remaining new nodes leave with one node down. state is always fits on a recommendation; a step it refuses or cannot judge is withheld. properties: state: enum: - fits - unknown type: string checked_at: type: string format: date-time n_minus_one_cpu_percent: anyOf: - type: integer - type: 'null' description: >- The tightest step's group CPU requests as a percent of what the new nodes leave with one node down. n_minus_one_memory_percent: anyOf: - type: integer - type: 'null' description: The same for memory requests. largest_pod: anyOf: - $ref: '#/components/schemas/CloudSavingsRightSizePod' - type: 'null' description: The pod with the largest CPU request on a changed group. required: - state - checked_at - n_minus_one_cpu_percent - n_minus_one_memory_percent - largest_pod type: object CloudSavingsRightSizeSizing: description: >- The usage an offered right-size was sized on, over window_days, each the tightest step's figure after the change: the busiest node's CPU max and p99 against one new node (under 100%), and the group's p95 CPU against the new capacity left with one node down (under 50%). properties: window_days: type: integer node_max_cpu_percent_after: anyOf: - type: integer - type: 'null' node_p99_cpu_percent_after: anyOf: - type: integer - type: 'null' group_cpu_percent_after_n_minus_one: anyOf: - type: integer - type: 'null' required: - window_days - node_max_cpu_percent_after - node_p99_cpu_percent_after - group_cpu_percent_after_n_minus_one type: object CloudSavingsRightSizeExecution: description: >- How an offered right-size runs. method: in_place_resize (drain, power off, change the plan, power on) or replace (the nodes or the managed pool are re-created). nodes_at_once: how many nodes of the first wave's group are down at the same time, from the platform's resize lane rule. groups_changed of groups_total node groups. cpu_change and memory_change describe the first wave (null when neither applies). canary: the waves inside the first wave's group, as node counts (one node, a quarter, the rest). properties: method: enum: - in_place_resize - replace type: string nodes_at_once: type: integer groups_changed: type: integer groups_total: type: integer cpu_change: anyOf: - enum: - halves - reduces type: string - type: 'null' memory_change: anyOf: - enum: - same - reduces type: string - type: 'null' canary: items: type: integer type: array required: - method - nodes_at_once - groups_changed - groups_total - cpu_change - memory_change - canary type: object CloudSavingsRightSizePod: description: >- One pod as a right-size plan names it, with its summed container requests. properties: name: type: string namespace: type: string cpu_millicores: type: integer memory_bytes: type: integer required: - name - namespace - cpu_millicores - memory_bytes type: object securitySchemes: BearerAuth: bearerFormat: PAT scheme: bearer type: http ```` This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.