VERDICT MEMORY
Your memory. Governed.
One memory that every AI assistant you use can read and write, with a record of who wrote what, when, and on whose authority.
Verdict Memory is live, in limited early access for individuals.
An assistant that forgets is a tool you reintroduce yourself to every morning. An assistant that remembers across clients, without mixing what belongs to whom, is a different kind of instrument. The hard part is not storage. The hard part is isolation.
HOW IT WORKS
Connect your assistants.
You connect assistants such as Claude, ChatGPT, or Cursor, and they store and retrieve information through your own isolated instance. Untrusted sources are held in quarantine until you or a trusted agent you designate approves them; superseded information is retained and linked to what replaced it.
Your own isolated instance
Your database, your storage, your authenticated endpoint. Your data is not stored in a shared table with other customers' data. Instances run in Frankfurt, Germany by default.
Governance, recorded
Untrusted writes are quarantined, approvals are recorded, nothing is overwritten. Complete version history and audit trail: who wrote what, when, and on whose authority.
Yours to take
Export everything, anytime. Deletion is self-serve, and deletion means deletion.
PRICING
One plan.
Start with 14 days free, then $7.99 for your first month, then $19.99/month. Connect up to 5 assistants. Unlimited memories.
Teams, agencies, and regulated organisations: dedicated instances, data-residency options, and compliance features are available.
Interested in Verdict Memory?
We are talking to a small number of organisations about early use.
THE COMPANY
One model's answer is one opinion.
We build infrastructure that makes AI outputs verifiable, and memory that makes AI systems continuous. For people and organisations where being wrong is expensive.
Disagreement between rival systems is signal, not noise.
A single model is fluent when it is right and fluent when it is wrong. Confidence is not verification.
We build the layer that turns that disagreement into something you can act on, and the memory that keeps it.
PORTFOLIO
Two products, one thesis.
Verification and continuity are the same problem seen from two sides: an AI system that cannot be checked, and an AI system that cannot remember. We build for both.
Verdict
Adversarial debate between rival-vendor frontier models, with a judge. Returns a verdict, a confidence score and a disagreement map.
On the web, and as a connector inside Claude, ChatGPT, Cursor and Perplexity. Mobile app shipping.
Verdict Memory
A hosted memory service for AI systems. Privacy and non-mixing are design constraints, not features added later.
Limited early access for individuals.
Reserved
The portfolio is built to hold more than two.
PROOF
What we can show, and what we cannot.
Intentionally empty.
Customers
Intentionally empty.
Funding
Intentionally empty.
Published benchmarks
Intentionally empty.
Press
This section is intentionally empty. We have no customers, no funding, no press coverage and no published benchmarks of our own. When we have them, they will appear here with sources. Until then, the only claims on this site are ones you can check yourself.
METHOD
How a claim gets checked.
01
A question enters.
One that is expensive to get wrong.
02
Rival systems argue it.
Frontier models from rival vendors answer independently and attack each other's reasoning.
03
A judge rules.
You get the answer that survived, a confidence score, and a map of exactly where they disagreed.
COMPANY
Independent by construction.
We are not a model vendor and we are not owned by one. That is the only position from which cross-vendor verification means anything: a judge with a stake in one side is not a judge.
Enterprise verification at volume and AI implementation for high-risk environments are handled by our services arm.
There is a direct line for each audience.
Investors, enterprise buyers, partners and press each reach a different inbox.