Sovereign AI
Your service platform was built for people filling in forms.
Agents do not fill in forms. They query, decide and write. Every platform that models work as screens is now charging you a premium to bolt an agent onto an interface that agent never needed.
We are building the alternative: service management that is headless and API-first, designed to run on a graph you own. It will be released open source, so leaving will always be available to you — including leaving us.
Build status
Meridian, our Enterprise Service Management (ESM) platform, is in development and not yet generally available. This page describes the architecture we are building and the licence it will carry, not a product you can download today. The exit-ramp engagement below is real and available now; the destination platform is not finished.
Why this is a finance conversation
The seat stopped describing the work.
Seat licensing made sense when a person was the unit of work: more people, more value, more licences. Once most routine transactions are executed by software, the seat count stops tracking anything real. You are billed on a number that no longer moves, for capability increasingly delivered by something that was never a seat.
The premium is not optional
AI capability arrives priced into the platform rather than alongside it. The increase lands whether or not your teams adopt it, and whether or not it works on your estate.
The debt is already on the books
Years of customization against a rigid schema is the real cost of ownership. It is also what makes each upgrade a project, and what a vendor is implicitly charging you to keep.
Nobody priced the exit
The cost of leaving is the number that never appears on the renewal paper. Not knowing it is not the same as it being zero — it is what makes the renewal conversation one-sided.
This is why the decision has moved out of IT. It is a total-cost question owned by the people who sign for it, and the technical merits of the platform are no longer what settles it.
The platform — Meridian ESM
Built by agents, for agents.
Build status
Meridian (Enterprise Service Management) is in development and is not yet generally available. What follows describes the architecture we are building and the licence it will carry, not a product you can download today. We would rather tell you that than have you find out during a migration.
Traditional Enterprise Service Management builds interfaces for humans and attaches AI to them afterwards. Meridian inverts that. Agents operate against the record layer directly through APIs and context protocols, and human judgment is invoked at explicit authority thresholds rather than at every step.
An API an agent calls directly
Legacy platforms model work as screens, then bolt an agent on top to type into them. Reverse it: the agent queries and writes to the record layer directly, and a screen is rendered only when a human is actually needed.
Per deployment
Seat licensing prices a world where a person is the unit of work. When most transactions are executed by software, seat counts stop describing anything and start being a tax on a number that no longer moves.
Workflows generated against your own topology
The customization debt on a legacy platform is the real cost of ownership, and it compounds. Generated workflow means the adaptation happens against your actual graph rather than being hand-built and then maintained forever.
A graph you hold
Your configuration, your incident history and your process knowledge are the substrate any future agent has to reason over. If a vendor holds them behind an export format, every future decision routes through that vendor.
Decide at the threshold
Agents sense, route, analyse and execute routine operations. Humans are invoked at explicit authority thresholds — the decisions that carry real consequence — instead of processing queues.
The engine underneath
One design decision carries most of the others. Service management is a graph problem wearing a ticketing costume, and splitting it across separate systems is where the coordination cost accumulates.
One engine, four models
SurrealDB
Graph, document, vector and key-value in a single engine. A configuration estate is a graph, an incident is a document, and semantic search needs vectors. Splitting those across three systems is where most of the coordination cost lives.
State that broadcasts
Live queries
When something crosses a threshold, the change is pushed to whatever is listening rather than polled for. Agents working elsewhere feel the state change and adjust, without a central coordinator instructing them.
Record-level permissions
Native RLS
Access is enforced at the record, in the engine, rather than in application code that has to be re-implemented for every new surface. Agents inherit the boundary rather than being trusted to respect it.
Workflows generated, not hand-built
The expensive part of a legacy platform is not the licence. It is the years of customization built against a schema that was never yours, which then has to be maintained by people who charge by the hour to understand it.
When the platform is agent-first and the topology is a graph you own, adaptation happens against your actual structure. Workflows are generated, compiled and refined from your own configuration rather than assembled by hand and then carried forward as debt.
The licence
The core is a gift. We are explicit about where that stops.
An argument for sovereignty is worth nothing if the thing making it carries the same lock-in risk you are trying to escape. So the operational core will be released under Apache-2.0— a permissive licence with an express patent grant, chosen because it is the one enterprise legal review objects to least. That is a stated commitment, not a published repository: nothing is on offer under that licence yet.
Open source
Meridian — Enterprise Service Management (ESM)
The operational core: workflow, ticketing, asset and configuration tracking, the Enterprise Service Management primitives. Everything you would need to run service management without us.
Commercial
System of Cognition · Judgment Infrastructure
The layers above: how the organization senses and decides, and how authority is made explicit across humans and agents. This is what The Value Network sells commercially: C4E subscriptions, network-routed engagement work under Network-First attribution, and the fixed-price Diagnostic. The open core is not one of them — we do not charge for Meridian.
Why give away the part everyone else charges for
Because ticket routing and asset tracking are not where the value is, and pretending otherwise is what produced the current market. The service-management primitives are commodity. What is scarce is the judgment layer above them: knowing what your organization should do, and who gets to decide.
A permissive licence is also what would make the sovereignty claim testable rather than rhetorical. Once the core is published you will be able to run it without us, fork it if we disappoint you, and audit what it does with your data. Until then it is a commitment you should hold us to, not something you can yet verify — and you are entitled to treat those differently.
The exit ramp
Nobody cuts over on a Friday night.
Replacement is not a switch. The new platform runs alongside the incumbent for a period, taking real work, until it has demonstrated on your own estate that it holds. Only then does anything get turned off.
Side-by-side is the proof
The migration period is not overhead to be minimized. It is the evidence a CFO needs: your own processes, your own data, running on both platforms, with the difference measurable before commitment.
Your contract sets the clock
The binding constraint is not technical readiness. It is where you sit in your current agreement, because the side-by-side period has to complete before the renewal decision, not after it.
Roles change, people stay
The political question in any migration is what happens to the people whose expertise is the incumbent platform. That gets answered explicitly, in writing, before the work starts.
When to start
Working backwards from your renewal date rather than forwards from today:
start = renewal date − side-by-side period − evaluation
Which means the honest answer is sometimes that you have already missed this cycle. If so we will say so, and the useful conversation is about the next one — the cost pressure that brought you here compounds rather than resolving, so the window comes round again in a stronger position.
We continue
The exit ramp
The licensing economics no longer justify the estate, or the platform cannot absorb what you need without another rebuild. We migrate you, side by side, on a schedule set by your renewal.
We decline
Respectfully, not now
The timing is wrong, the readings say you are not ready, or your intent does not require what we build. You get a short reasons-and-conditions document: why we are saying no, and what would have to change for the answer to be different.
There is a third answer, and it is not ours to give here. If your estate is genuinely worth deepening, that conclusion comes from field judgment rather than this page alone, and it ends with a referral to partners who do that work — we do not. Many organizations start with the Strategy Engineer AI agent; a Metabolic Health Diagnostic is a later path when a bounded loop needs a hard look.