Solutions · Living Architecture · In development · pilots open soon
A map of your systems that stays alive.
Your body keeps a map of itself and updates it as you move. Living Architecture does the same for your infrastructure: your systems, what connects to what, the evidence behind every fact, kept current by your own people and your own AI.
Live software we run on our own estate. Customer pilots run on a mirror of three of your systems that we build and run.
Illustration ·A fact with its sourceOut of date after a change
Your body keeps a map of itself.
Biology: proprioception. You know where your hand is in the dark.
Every fact on the map has a source.
Which systems you run, what they connect to, which data sits where, who owns each. Living Architecture holds them as a map, and behind every fact sits the document, the answer or the record it came from. Nothing unconfirmed is shown as fact; an unknown stays marked unknown.
In your estate
The map is built from your documents, from workshop answers your Owner gives an agent, and from read-only connectors to tools you already run. Each import quotes its source and is checked against the source text.
Biology: a nerve carries the warning to the place it concerns.
A warning lands where it matters, not in your inbox.
A new vulnerability or a new rule arrives every day. The map shows which of your systems it touches, which data sits behind them and who owns them, so the person who decides sees the question in their own terms.
In your estate
Vulnerability feeds are matched against the software versions your map declares. A match is affected, potentially affected or unresolved; against a declared list it reads “declared version potentially affected” until your owner confirms what actually runs. Regulations and standards are matched the same way, through scoped assessments on your frameworks.
Biology: healing under control. The body repairs itself without rewriting itself at random.
Every change to the map is a proposal.
Your people and your agents propose changes; your Owner approves them in one place, Today. One setting decides how much agents may apply on their own: Strict, Balanced or Open. Balanced is the default.
In your estate
Strict holds every change for a person. Balanced lets agents apply what a source states and holds what an agent inferred. Open lets agents apply more, above a fixed floor of changes that always needs a person, in every mode. Each agent run reaches your Owner as one entry, not a pile of items. Every applied change carries who proposed it, which source, and when, and can be reverted item by item.
Biology: scar tissue remembers.
One threat model for the whole estate, that knows when it is stale.
Threats are recorded once, by what they attack and how, and applied per system. When a system changes, the assessments that depend on it are marked out of date, so your threat model never quietly goes wrong.
Re-checking stale assessments automatically is in development.
In your estate
The register holds one scenario per mechanism and protected object; look-alikes are held for a person, never merged by a machine. Every conclusion is recorded against the exact inputs it rests on, and when one of those inputs changes, the map flags the conclusion on Today.
“The declared one, not yet updated.” Confirmed by the owner.
Exposed now
Version matches
Route reachable from the internet
Owner confirmed what runs
Proposed: update PHP to 8.3.36, with the advisory's interim mitigation
Approving records the decision. Nothing on the network changes.
Approved by the owner, recorded
Exposed now
Version matches
Route reachable from the internet
Owner confirmed what runs
The version matches, the route is reachable, the owner confirmed what runs: exposed now. Next door sits the order system, shown for planning only. Exposed now: version matches, route reachable from the internet, owner confirmed what runs. The order system could be reached next; it is not part of the verdict.
Step 3 of 4 · The answer
Version matches
The declared PHP 8.3 on the customer website, confirmed running by the owner, against the advisory as served by NVD.
Route reachable
The map records a public route from the internet through the proxy to the website.
Owner confirmed
The service owner's answer, recorded with their role and the date.
Could reach next
The order system, through the website's link to it. Shown so you can plan, not part of the verdict.
The advisory rows are a recorded Ansvar run from 28 October 2026. The company and its map are fictional; the route and the owner's confirmation show what the pilot builds for your systems. Nothing here changes a system.
We run Ansvar on it.
Our own estate lives in a Living Architecture twin. Our threat register has been kept there, as the authoritative one, since 2 November 2026, and every change to it goes through the same proposals and the same Owner approval we offer you.
Financial entities assess cyber threats continuously, and all but microenterprises repeat the risk assessment after each major change to their ICT systems, processes or procedures.
Under 24 hours from a flaw found in the wild to a working exploit (median), and 30 to 60 days for a company to fix a critical external vulnerability. A threat model reviewed once a year cannot keep pace.
Where it stands, in November 2026.
Live today
Running on our own estate.
The map, the Today inbox, the three change-control modes, the threat register, scoped assessments on your frameworks and vulnerability matching over declared software are live software we run on our own systems every day.
In the pilot
Your map, built with you in about two weeks.
We build your map from your documents and two one-hour sessions with your Owner, match your declared software, run the assessments for your frameworks, and attack and defend a mirror of three systems you choose. A person approves every change.
In development
Re-checking, learning, responding.
Re-checking stale assessments automatically. Learning your systems' normal from their own traffic. The response layer, the Ansvar Immune System, with its own page.
A pilot of about two weeks, without touching production.
You describe your systems. We build the map and a mirror of three systems you choose, attack the mirror, and defend it live with you in the room.
You bring descriptions.
Architecture documents, a list of the software you run, two one-hour sessions with the person who owns the systems, and the three systems you care about most. No access of any kind.
We run everything.
Two servers we operate for the pilot only: one holds your documents and your map, the other holds the mirror we attack. Torn down at the end, with the deletion recorded.
You keep the results.
Your map, exported to you. A readout of what the attack-and-defend sessions showed. A plan naming which of your tools would replace each stand-in.
About six hours of your time.
Over about two weeks. The shape is a proposal and we confirm it with you before we start.
A pilot starts with an NDA and a pilot data-processing agreement. We never touch your production systems, logs or credentials.
You stay in control.
Every claim has a source.
Follow any finding back to the record behind it.
Unknowns stay unknown.
Nothing unconfirmed is presented as fact. A link nobody described is marked “auth: unknown”.
People decide.
A named person approves each response. Ansvar does not change your systems by itself.
It runs where you decide.
The pilot runs on a mirror we operate, never on your production. After the pilot the product runs in your environment, standalone or on your own platform, and your map comes with you, exported. Only the pilot runs on Ansvar's side.
Start with three systems.
One conversation about the systems you care about most. Bring a diagram and one recent change.