7 instruments across 4 families. Each answers a specific failure of cloud AI, and each is handed over to run inside your perimeter, with weights, hardware spec, and architecture included.
“We don’t sell software as a service. We hand over the whole instrument, to live inside your perimeter.”
Cordon
Inference that stays inside your walls.
Cordon is our private inference engine. Rather than a privacy policy, it is built as a set of technical constraints, from a hardware root of trust through a trusted execution environment to a client-verifiable, Merkle-chained audit log. The four properties we build toward are containment, verifiability, sovereignty, and auditability.
- Hardware root of trust: TPM 2.0, Secure Boot, dm-verity
- AMD SEV-SNP, Intel SGX v2, or ARM TrustZone execution
- AES-256-GCM weights with a client-held key hierarchy
- Offline-verifiable Ed25519 Merkle-chained audit log
Matta
Meaning is infrastructure.
Enterprise semantic infrastructure for organizations that cannot afford ambiguity. Matta provides full lifecycle management of OWL 2 DL ontologies, knowledge-graph operations, and LLM grounding, giving you machine-verifiable shared meaning across every system you govern.
- Four-layer ontology architecture with strict dependency governance
- SHACL validation enforced before every persistence event
- SPARQL, GraphQL, and natural-language query interfaces
- LLM grounding: validate facts against the knowledge graph
Knott
Workflows that tie your systems, your AI, and your people together.
Knott is a sovereign workflow automation platform for organizations that want capable automation without a cloud dependency. It runs every AI decision on Cordon, our private inference engine, so models execute on your own hardware. It keeps all data inside your perimeter, escalates to human reviewers when model confidence is low, and records an immutable audit trail of every decision, approval, and node transition.
- Visual workflow designer with eight composable node types
- Confidence-scored AI decisions running privately on Cordon
- Human-in-the-loop escalation with required justification and SLA tracking
- Immutable audit trail across every decision, approval, and transition
Wallgarden
The executive intelligence layer your enterprise never had.
Wallgarden is an Enterprise Decision Operating System. It sits above every system you run, including ERP, CRM, finance, procurement, and operations, and continuously analyses the business across finance, operations, procurement, legal, strategy, and risk. It surfaces decisions before problems arrive, and builds institutional memory from every outcome.
- Continuous coverage across finance, operations, procurement, legal, strategy, and risk
- Organizational memory that retains every decision, assumption, and outcome
- A living knowledge graph of the entire enterprise
- Decision simulation: model a move before money is spent
Bubbly
The coding agent that never leaves your perimeter.
Bubbly is a sovereign agentic software factory. It is a full coding environment, with an IDE, a CLI, and a mission-control room, where autonomous agents understand your entire codebase, plan a change, implement it across files, run the terminal and tests, debug failures, and review the result. Every model runs locally on your own hardware via Ollama, so your source, prompts, and secrets never leave your perimeter.
- Whole-codebase intelligence: symbol, semantic, and call-graph retrieval, not file-by-file guessing
- Runs the full agentic loop: plans, edits across files, runs the terminal and tests, then reviews its own work
- Every model runs locally via Ollama, so your code, prompts, and secrets never leave your hardware
- Spec-driven and fully audited, with every action permissioned, logged, and reversible
Motif
The whole classroom, in your language, offline.
Motif is a sovereign education ecosystem built for the Tanzanian classroom. It pairs a local-first student tutor with an offline teacher copilot under one roof, curriculum-aligned, in Kiswahili and English, running entirely on hardware a school already owns. No internet, no telemetry, and no student data leaving the device.
- Student tutor: seven learning modes with spaced-repetition flashcards and auto-generated quizzes
- Teacher copilot: curriculum-aligned lesson plans, exams with marking schemes, and class reports in seconds
- Kiswahili and English instruction, with native understanding and no translation layer
- 100% offline on modest school hardware, and no student data leaves the device
KW5-Lite
A Kiswahili-first model, in research.
KW5-Lite is a compact Swahili-language conversational model and an early research release. It is built to hold natural, helpful exchanges in Kiswahili, covering everyday conversation, general knowledge, and language instruction, with careful handling of sensitive topics. It is deliberately lightweight and best suited to short, self-contained exchanges rather than long sustained conversations. We publish its strengths and its known limitations openly.
- Swahili-first, with native understanding and no translation layer
- Compact and low-latency, so it runs on modest, on-premise hardware
- Trained for everyday conversation, general knowledge, and language teaching
- Open weights under Apache 2.0, with a transparent, published evaluation record