I
Determinism at the boundary
Models may reason probabilistically. Permission to move money, expose data, or invoke infrastructure should resolve through explicit rules.
Dixit AlgorizmiThus spoke the algorithmStart a conversation Independent applied research company · Ontario, Canada
We build the deterministic layer autonomous systems need.
AI can generate possibilities. Our work is about governing what happens next through explicit authority, bounded execution, and infrastructure that can be audited without asking a model to explain itself.
Scroll to investigate
01 / Research thesis
Our thesis
Intelligence should expand what software can propose, not what it can do without permission.
We research and build infrastructure for the moment an AI system crosses from language into action. The objective is practical: keep critical authority outside the model, make policy enforcement deterministic, and preserve evidence of every consequential decision.
II
Authority stays outside the agent
An agent should propose an action without automatically possessing the credentials or authority required to execute it.
III
Context before execution
A harmless-looking call can become dangerous in sequence. Systems should evaluate the plan, state, and cumulative effect, not only one request.
02 / Flagship product
UBAG
The authorization gateway for agents that act in the real world.
UBAG sits between an autonomous agent and the systems it wants to change. The agent proposes. UBAG evaluates identity, policy, plan context, and live state before any protected credential is introduced.
Its execution boundary has been evaluated on AgentDojo and InjecAgent, two public academic prompt-injection benchmark suites, with methodology and limitations published alongside the results.
UBAG / DECISION TRACEPOLICY ACTIVE
01Agent identityVERIFIED
02Requested capabilityIN SCOPE
03Plan-level exposureEVALUATED
04Credential releaseBOUNDED
DECISIONALLOW / AUDITDeterministic policy result · no model approval
03 / Intellectual property
Patent portfolio
Four connected inventions. One infrastructure thesis.
Our provisional applications cover complementary layers of agent identity, authority, data normalization, and web routing.
Credential Injection
Keeping sensitive credentials outside the agent boundary and introducing them only after an explicit authorization decision.
UBAG Gateway
A deterministic control plane for evaluating, constraining, and auditing actions proposed by autonomous systems.
Universal Converter Engine
Turning inconsistent legacy inputs into stable, machine-readable structures before they reach an AI system.
Web-Layer Agent Routing
A web-layer protocol for distinguishing, challenging, and routing automated actors using verifiable identity signals.
Provisional patent applications have been filed. “Patent pending” describes application status and does not imply that a patent has been granted.
Open work
Useful standards improve when they can be inspected.
We publish the UBAG web-layer protocol and reference SDKs so developers can evaluate the identity and routing model independently. The commercial gateway remains a separate product.
04 / Company
Dixit Algorizmi Inc.
A small company working on a large boundary.
Dixit Algorizmi is an Ontario-based applied research company building deterministic infrastructure for autonomous software. We focus on the interfaces between identity, authorization, data, and execution.
We are currently working with design partners on UBAG pilots and continuing research across the wider patent portfolio.
mohamed@dixitalgorizmi.comBuild with us