RYMVARD

Product

One ledger.Two ways to read it.

Leaders need the answer and what it costs. Engineers need to check how it was reached. Rymvard shows both from the same numbers — switch between the business and the technical view at any time.

The screens below are the running product on an illustrative example estate — not a customer.

“Hosting providers lose money in the gaps between their data centers, their cloud bills and their licence audits. Rymvard closes those gaps. It reads an estate as it is — read-only, nothing to migrate — and shows what is used, what is paid for and what is licensed.”

Thorsten Meyer, founder of Rymvard

For business leaders

Decisions you can defend in the boardroom.

Capacity turned into sentences, prices and dates — every figure traceable to its source.

Rymvard's business view: the estate in a few sentences with sourced numbers, health per site and runway in months
The business view: where the estate stands, each number linked to its source.

Know where you stand in a minute.

A business view writes the estate in a few sentences — capacity, runway, recovery, money at stake. Every number links to where it comes from. A board pack prints the same facts with a fingerprint.

See what needs attention — and its price.

Findings arrive with an owner, a due date and their impact in kilowatts and euros. They close themselves when the problem is gone, so the list stays short and true. Alerts reach Teams, Slack, e-mail, PagerDuty or Opsgenie — and on-call incidents close by themselves.

Say yes to a deal, safely.

Check a request against power, cooling, space and recovery in one step, try it as a scenario, then turn it into an offer that holds the capacity for up to 30 days — won, it becomes the contract. Every hold is counted once, so two offers can never sell the same kilowatt.

Prove every recovery promise.

Each disaster-recovery commitment is checked against capacity reserved at its recovery site, with its last switchover test. The evidence pack is one download — for auditors and for each customer.

Find the megawatts you already own.

Sold-but-idle and powered-but-uncoolable capacity, idle servers and over-stated nameplates are named with their value. A refresh plan dates each end-of-life cohort with its cost and the kilowatts it frees.

Report and bill on time.

EED and German EnEfG figures from metered months, CUE where you set a grid factor, and draft statements per customer from meters and versioned tariffs that reconcile with your ERP — the invoice stays yours.

A what-if scenario in Rymvard: a new deal and how each site's runway moves
Scenarios: a deal, an expansion or a delay — and each site's runway, before and after.
Rymvard's findings inbox with severity, impact in kilowatts and euros, owner and status
Findings: one per root cause, with impact, owner and history.

Free is not free.

Every workspace home — and Overview, Runway and Halls — opens with two bars. Physical: the measured peak against what is commissioned. Allocatable: what is left after recovery reserves, contracts, held quotes and power without cooling. What remains is what you can really sell.

Customers see their own.

A customer signs in to a tenant view of their own contracts, recovery coverage and test status, energy statements, draft billing lines and the compute their workloads use — never another customer's.

Rymvard's compute view: free compute units per profile — how many more VMs of each size still fit, and which resource runs out first
Compute: what the clusters can still take — in compute units per profile, and what runs out first.

Sell compute the way you price it.

Per cluster: servers with their support status, vCPU, RAM, storage and GPUs — what is installed, what you can still promise after the failover reserve, and how many compute units are free per profile, with the resource that runs out first. Holds in compute units reserve it until the deal closes.

Approve first, act once.

People and AI agents propose a change — a hold, a release, an incident. A second person approves exactly that change, and it runs once. Licences, runway per resource and right-sizing advice come with the same numbers.

Rymvard's SAP & HANA view: memory per HANA host split into OS reserve, used, allocated, held for replication and free, with an overbooked host flagged
SAP & HANA: what each HANA host still takes — and whether every takeover fits.

SAP and HANA, by the gigabyte.

For SAP hosting: per HANA host the OS reserve, what each system may take and really uses, what is held for system replication — and what is free for the next system. A new system, by size or by profile S, M, L or XL, gets a host with room and a partner host for its replica; administrators get a billing basis per customer in weighted HANA hosts — counts, not prices.

Five homes for SAP teams.

SAP basis, hosting sales, customer management, capacity planning and resilience each open on their own figures — what still fits by profile, the weighted pipeline, every customer's protection, the S/4HANA wave and every takeover. Customer names stay with planners; SAP findings never name a customer.

A takeover that fits — checked before you need it.

Every replicated pair is checked: can the secondary host carry the production system after a takeover, once the test systems on it stop? Read-only from the SAP Host Agent, the HANA monitoring views and your landscape tools — Focused Run, ServiceNow, LaMa and Solution Manager. Where they disagree, a ranking you set decides.

Rymvard's Licences & Audit home: Licence positions short, Licence positions unknown, Vendors with a position short, Positions reported monthly — every vendor's positions counted by your own rules
Licences & Audit: every vendor's positions, short and unknown counted apart.

Every licence, counted where it runs.

Oracle processors and middleware, SQL Server and Windows Server cores, Broadcom per-core subscriptions, Red Hat and SUSE subscriptions and IBM PVUs — counted by the rules and entitlements you enter, next to the capacity they run on. Counts, never prices.

Before the VM lands.

Database and Oracle placement name the new processor licences a host would need; the monthly SPLA and provider reports come from nightly records, and a missing night is named, not guessed.

Rymvard's Customer registry: every customer with its aliases, business groups and the data rows that name it, unconfirmed names marked
Customer registry: one customer under every spelling the ledger holds.

One customer, every spelling.

The customer registry holds each customer with its aliases and a tree of business groups; every row that names a customer resolves to it, and billing and energy statements count it once. A name nobody knows waits as an unconfirmed customer.

Every cloud resource, with its owner.

AWS, Azure, Google Cloud, OCI and OpenStack — sovereign partitions included — and on-prem sources discovered with their identities; tag rules give each resource its customer, business group and contract, and an ambiguous identity is quarantined, never merged by a guess.

Workspaces

Forty-four teams. One ledger.

Sales, finance, the NOC, the board and the teams of every solution — SAP, Oracle, licences and subscriptions, private cloud, databases, GPUs, recovery, sovereign cloud, platforms, hyperscaler hosting, transition and cloud commitments — each open Rymvard on their own home: their question, their figures, their findings. A solution's homes appear once the ledger holds its data. Underneath, the numbers are the same for everyone.

Rymvard's Sales workspace: sellable capacity now, the first short quarter, quotes and holds, what changed since the last visit and suggested next steps
The Sales home: what can be sold today, and what changed since the last visit.

A home for every team.

An admin gives each person a default workspace, or your identity provider does through SCIM. Its home answers the team's question with its own figures and only the findings that concern it — the rest of the ledger stays one click away.

Rights stay with the role.

A workspace only hides pages; it never grants access. Money is for admins: everyone else sees the same home without the euros, and a note says so.

  • IT infrastructureWhat can the clusters still take, and what is coming?
  • SalesWhat can I sell today, and does this request fit?
  • ManagementWhere does the estate stand?
  • FacilityDoes the infrastructure hold, and where must I go?
  • C-LevelWhere do we grow, and what must we invest?
  • FinanceDo we earn money, and where is money lying around?
  • Capacity planningWhen does it get tight, and where?
  • Account managementHow are my customers doing?
  • Sustainability / ESGHow efficient and clean are we?
  • Compliance & auditCan we prove it?
  • NOC wallIs anything burning right now?
  • ProcurementWhat must be ordered, and when?
  • SAP & HANA · five homesOperations, sales, customers, capacity and the S/4 wave, resilience.
  • Oracle · four homesOperations, sales, customers, upgrades and lifecycle.
  • Licences & subscriptions · five homesMicrosoft and SPLA, the VMware exit, Red Hat and SUSE, IBM, Licences & Audit.
  • Private cloud · four homesOperations, migration and onboarding, platform and licences, sales.
  • Databases · four homesOperations, licences and audit, customers, capacity and placement.
  • GPU & AI · four homesGPU operations, sales, AI facilities, customers.
  • Recovery ServiceCan all customers recover together — and can we prove it?
  • Cloud, platform & takeover · five homesSovereign cloud operations, platform engineering, hyperscaler hosting, transition, cloud commitments.

Since your last visit.

What changed for your team — new and resolved findings, edits to the records you care about — phrased in a few sentences, every number checked.

Why? on every figure.

Each number on a home explains itself: how it is computed and where it comes from, in plain words.

Next steps you approve.

The rules list the open findings; the AI picks the one that helps most. One click turns it into a proposal that a person approves — accepting or dismissing teaches the AI.

Documents in, typing out.

Energy invoices, capacity requests and switchover test reports are read in. Rules find every value in the document; the AI only picks among them; you confirm.

Rymvard's Management workspace: sold share, first short quarter, DR backed and sold but unused, with the situation in sentences
Management: four figures, the situation in sentences, the risks that matter.
Rymvard's NOC wall: urgent items, live load and halls by load on a full screen
The NOC wall: urgent first, refreshed every 15 seconds.

For technical teams

Every number, open to inspection.

Standard protocols in, standard interfaces out, and the method printed next to every calculation.

Rymvard's failure impact view: if a hall fails, the customers, contracted and live load, tier-0/1 workloads and recovery reservations it takes down
Failure impact: who goes down if a hall fails — and whose recovery capacity it takes with it.

Live readings, in-country.

A Rust collector per site reads Redfish, SNMP v2c/v3, Modbus TCP and BACnet/IP — and, for SAP, the Host Agent and the HANA monitoring views; for Oracle databases, Oracle REST Data Services with a fixed list of read-only queries — buffers to disk when the link drops, and ships signed batches over HTTPS or MQTT. Raw data stays where it was measured. In a customer's network a collector is bound to the sites and accounts it may write for, installed with Terraform or Helm.

What if a hall fails?

Pick a hall or a site: the customers and workloads it takes down, whether each recovery commitment still holds when every landing is called at once, and whose recovery capacity it takes with it.

Power math you can check.

Per rack: if feed A fails, can feed B carry the peak? Side capacity is V × A × √3 × utilisation, set against the measured 24-hour peak. Runway per constraint shows whether power, cooling or rack positions run out first.

Every write, on record.

A change log records who changed what — a person, an import file, a connector — with the values before and after. Model-versus-measured checks flag ratings, contracts and nameplates that disagree with the meters.

AI that shows its work.

Ask a question in plain English: the answer names the query, the sources and how many of its numbers were checked against the ledger. Engines run locally by default; Anthropic, OpenAI, OpenRouter or your Codex subscription only on your opt-in, with customer names pseudonymised — and each engine earns its own track record in shadow mode.

Your tools, your data.

REST and MCP for agents, OIDC single sign-on and SCIM provisioning. Read-only connectors for NetBox, openDCIM, vCenter, Proxmox, Prometheus (Kubernetes, vSphere, NVIDIA DCGM), DCIM, CMDB and CRM — for SAP landscapes Focused Run, ServiceNow with SAP options, LaMa and Solution Manager — and Oracle Enterprise Manager. ServiceNow incidents on opt-in, Docker install on any Linux, the interface in English with a German switch, a handbook in seven languages — and an export in the same templates, no lock-in.

Capacity counted once.

Contracts, recovery reservations and holds are claims on a resource in its own unit — kW, cross-connects, Gbps, vCPU, GiB, GPU — for a time window. A new claim is checked under lock against everything that overlaps, so parallel deals cannot oversell.

Supply paths to the source.

Power and liquid-cooling paths are traced from each rack to the utility. Shared UPS, chillers and failure domains are named, and the surviving path must carry the load at its derated rating. A missing rating makes the verdict unknown, never green.

Compute from the BMC up.

The collector reads each server's Redfish inventory — CPUs, cores, threads, memory, drives, GPUs, BIOS and BMC firmware. vCenter, Proxmox, Kubernetes and NetBox add clusters and workloads, read-only. Where a declared value disagrees with the hardware, you decide.

Evidence that replays. A ledger that restores.

A snapshot freezes a calculation's inputs and result with SHA-256 and replays them later. Background jobs run on one worker at a time with fencing tokens. Each nightly backup is copied to a second machine, its SHA-256 compared at the target and kept 30 days — a failed copy, or one older than 26 hours, is named on the Operations page — and a restore drill brings a backup up in isolation and measures how long it takes.

New in 0.28 · Enterprise access

Each person sees their part.

Scopes, tenant views and sessions are enforced where the data is — so one ledger can serve many customers and many teams without showing everything to everyone.

Staff scoped by the database.

A staff user can be limited to customers, business groups or sites — enforced by the database's row-level security, not just the screen. Everyone sees their own scope on My scope.

SCIM without surprises.

SCIM groups map to scopes; a user that SCIM creates sees nothing until an admin or a group mapping gives it a scope.

Tenants per business group.

A customer's tenant view can be pinned to one business group; a value that names another customer is hidden whole.

Sessions that end.

Signing out ends every session; an admin can sign a user out everywhere, and a role change or a deactivation ends a user's sessions at once.

Ask the ledger in Rymvard: an answer with the route, parameters, sources and verified numbers
Ask: the answer, the facts table and how it was produced.
Rymvard's command palette searching racks, assets and customers
Search everything in two keystrokes — racks, assets, customers, deals, commitments.

What the AI does — and does not.

  • It routes questions and makes 26 typed decisions — 13 for the platform (is this the same asset, does this reservation still serve a commitment, will this deal close?) and 13 for the workspaces (which next step helps most, which alternative to offer, which cluster or rack to place on).
  • Every decision is logged with its inputs, confidence and the person who reviewed it.
  • Each one stays in shadow mode — it suggests, it never acts — until at least twenty reviews agree nine times in ten.
  • It never produces a number. Figures come from the ledger; a sentence the model writes is checked number by number.
  • By default no data goes to a third-party AI model: anything that would leave your environment needs your opt-in, and a hosted engine sees customer names only as pseudonyms.

See it on your own estate.

Start with one site: your inventory, your meters, your commitments.

Write to contact@rymvard.com