01
When the cloud is the adversary
Every request routed through foreign infrastructure reveals what an institution is monitoring, planning, or concerned by. Running inference inside your own perimeter removes that exposure.
Your datacan't leave.What runs?.
Regnant builds intelligence institutions own outright: the weights, the hardware spec, the audit trail. On-premise, air-gap capable, and fluent in the language you actually govern in.
Ownership, held
01
Private Inference
02
Decision Intelligence
03
Semantic Ground Truth
04
Workflow Orchestration
05
Agentic Engineering
06
Native-Language Models
Regnant · FIELD RECORDSystem№ 00 / 07
01The three constraints
Three constraints
we solve for.
Institutions don't shop for AI. They arrive with constraints that rule out the public cloud: a regulatory mandate, a liability the board carries, or a working language the prevailing stack was never built to handle.
01
Data that can't cross the border. Models that can't touch foreign servers.
→ Cordon / Infrastructure
Complete sovereign stack. Hardware root of trust to encrypted weights to air-gapped deployment. Your institution's perimeter is the hard boundary. What runs inside never phones home.
02
Decisions that carry liability can't be made on vibes and probabilities.
→ Wallgarden / UAMUZI
Deterministic decision systems that rehearse outcomes before committing capital or policy. What it models, you interrogate; what you approve, it executes; what it executes, you own.
03
The working language of government isn't English. Translation breaks institutional precision.
→ LUGHA / Language Models
A research programme aimed at models that hold Swahili, Amharic, and the other languages institutions govern in at the weights, rather than reaching them through a translation layer. Early, and published as it goes: KW5-Lite is the first release and its limits are on the card.
Research
Research papers will appear here when published.
02Why it matters
Three failures
the cloud
cannot fix.
Not shortcomings of a vendor. Properties of the arrangement.
02
When the model cannot read the language
A decision-maker briefed by a system that can't reliably work in their working language is constrained by the tool. Language proficiency is not optional in governance.
03
When the decision cannot be undone
Infrastructure, procurement, and public-health decisions unfold over decades. Model the alternatives before committing.
03The portfolio
Seven systems.
Handed over
whole.
Weights, hardware spec, and architecture included with every system. Nothing is rented back. Nothing phones home.
Cordon: private inference. The ground everything else stands on.
01/07Sovereign InfrastructureIn Production
Cordon
Private Inference Engine
Exfiltration is architecturally impossible.
Our AI cannot touch the public cloud. What do we run?
02/07Sovereign InfrastructureDeployment Ready
Knott
Workflow Orchestration
Automation you can actually be accountable for.
We need automation we can actually be accountable for.
03/07Sovereign InfrastructureEarly Access
Matta
Enterprise Semantic Platform
A semantic platform, not a file.
Our systems can't agree on what the data means.
04/07Decision IntelligenceEarly Access
Wallgarden
Autonomous Business Operating System
The decision arrives before the problem does.
Information is everywhere. Why are decisions still late?
05/07Decision IntelligenceIn Development
IIN
Industrial Intelligence Platform
Hear the failure before the floor does.
Every machine is talking. Who hears the failure coming?
06/07EngineeringIn Production
Bubbly
Self-Hosted Coding Agent
Your own coding agent. Not a chatbot with autocomplete.
Our engineers want AI coding agents. The code can't leave.
07/07EducationOpen Source · MIT
Milkshake
Sovereign Education Ecosystem
The whole classroom, in your language, offline.
What does sovereign AI mean for a school with no internet?
▣ Weights included◆ Hardware spec included◧ Air-gap ready⬡ Tamper-evident audit✕ Zero telemetry⚑ Runs in your perimeter
Common ground: 04
What every
system shares.
01
Handed over whole
Weights, hardware spec, and architecture included with every system. Nothing is rented back, nothing phones home.
02
Developer-ready platform
SPARQL, GraphQL, REST, and native SDKs across the stack.
03
Audit-first design
Every decision, approval, and transition lands in a tamper-evident, offline-verifiable log.
Developer surface. SPARQL · GraphQL · REST · native SDKs
▮Session opened inside the perimeter. Audit chain attached.
Containment enforced in hardware, not policy.
05Deployment
Choose your
perimeter.
Posture 01
Air-Gapped
Fully offline. No internet required. Models updated via secure media.
Best for: Governments, central banks, defence, critical infrastructure.
Posture 02
Private Network
Runs on your LAN. Data stays on-premise; updates sync over a controlled tunnel.
Best for: Standard enterprise and institutional deployments.
Posture 03
Hybrid
Core inference on-premise. Heavy training offloaded to a private cloud you control.
Best for: Large organisations with dedicated IT teams.
Dispatch
What moved.
PRODUCT2 min
Milkshake: two sides, one classroom→
The education district is one ecosystem with two halves: Kairos for the teacher — shipping today — and Weave, the studen…
01 Jul 2026
RESEARCH4 min
Why Air-Gapped Matters→
Every AI query an institution sends to a cloud provider is a clue about what it is watching, planning, or worried about.
20 Jun 2026
RELEASE4 min
Kairos is open source→
The teacher copilot for Tanzanian secondary schools ships under the MIT license — free for schools, NGOs, and government…
02 Jun 2026