AbteeX AI Labs · Aotearoa New Zealand

Building sovereign intelligence, end‑to‑end.

AbteeX AI Labs is a research lab developing artificial intelligence from foundations to product — silicon, model line, and the operational layer teams use every day.

Aotearoa signal Local roots, global work.

We build from Aotearoa New Zealand with place visible in the system: data rights, cultural context, maritime boundaries, and the Pacific edge are part of the architecture, not a footer.

LabAotearoa, NZ
StackSilicon → Product
Model1.05T MoE
AOTEAROA · NZ SOVEREIGN ZONE
Aotearoa New Zealand · sovereign zone marked
Thesis · Note 01

Most labs assemble. We descend — silicon, foundations, model, product.

/01Sovereign /02End-to-end /03Aotearoa-grounded
§ The stack · four layers, one direction
/01–04
/01 · Intelligence platform held in the light
1.05T · MoE
Platform · LumynaX

Sovereign intelligence, held in the light.

The intelligence platform: a 1T+ parameter mixture-of-experts that activates only what each task requires. Long context, alignment by construction, provenance on every weight.

1.05T MoE 512k context 18–28B active
lumynax.com Visit →
/02 · Agentic operations threads · 24
nodes · 8 live
Product · WeftlineAI

Threads of context, woven into action.

The agentic operations layer teams work through. Shared memory, approval-aware execution, and governed actions across Slack, Teams, and the systems your business already runs.

Live · ops layer Slack · Teams Cloud · on-prem
weftlineai.com Visit →
/03 · Aotearoa-trained model line Aotearoa corpus · cleared
1T MoE · 24 active
Model line · Te Mārama AI

A model that answers to its place.

Trained on a sovereign Aotearoa New Zealand corpus — consented, catalogued, and grounded in the cultural and legal context it serves. Built on LumynaX, 1T-parameter mixture-of-experts.

NZ · sovereign data 1T · MoE Te reo aware
temaramaai.com Read more →
/04 · The full stack end-to-end
4 layers · one direction
Discipline · Sovereign by construction

Built from the silicon up, not bolted on.

Most labs assemble. We descend. Hardware to foundations to model to product — every layer answers to a place, a people, and a record. Sovereignty is a design property, not a deployment option.

Hardware Foundations Model line Product
abteex.com — labs Brief us →

Notes from the lab, in the open.

04 / 2026 Sovereign AI NZ AbteeX explainer Sovereign AI in New Zealand A plain-language AbteeX position page on sovereign AI, specialist LLMs, data residency, governance, and why Aotearoa needs model infrastructure it can inspect and control. 04 / 2026 Healthcare Ops Emergency demand forecasting without theatre How short-horizon AI should support emergency departments without hiding uncertainty or replacing clinical judgement. Open article 04 / 2026 Justice AI in justice services should reduce friction, not decide outcomes A deployment framework for courts, legal aid, corrections, and case management where procedural fairness remains non-negotiable. Open article 04 / 2026 Police Operational AI for police must be auditable by design Why dispatch, evidence handling, intelligence support, and report drafting need strict provenance, review, and bias controls. Open article 03 / 2026 Defence Readiness intelligence for defence without black-box command A governance-first view of logistics, maintenance, training, and operational planning where AI supports readiness but does not own command authority. Open article 03 / 2026 Central Govt The public-sector AI layer should be explainable before it is scalable How central government agencies can use AI for service delivery, policy operations, and correspondence without losing accountability. Open article 03 / 2026 Insurance Claims intelligence that preserves trust at the moment of loss AI can accelerate claims, fraud review, and customer communication, but only if it separates assistance from final coverage decisions. Open article 02 / 2026 University Universities need research AI that respects provenance From grant discovery to ethics review and literature synthesis, the useful system is the one that protects attribution, consent, and academic judgement. Open article 02 / 2026 Local Govt AI for councils should make infrastructure trade-offs visible A practical operating model for water, roading, consents, complaints, and asset renewals where public value is more important than automation theatre. Open article 02 / 2026 Emergency Mgmt Crisis intelligence should compress time, not centralise judgement How AI can support civil defence, situational awareness, and recovery coordination while preserving local command and community context. Open article 01 / 2026 Education Assessment integrity in the age of generative AI A better university and school response moves beyond detection and redesigns assessment around evidence, process, and authentic demonstration. Open article 01 / 2026 Energy Grid operations need AI that understands constraint, not just demand Forecasting, maintenance, distributed energy, and outage response require models that expose uncertainty and operational limits. Open article 12 / 2025 Biosecurity Biosecurity AI must connect field evidence to national risk A field-to-policy architecture for pest detection, border intelligence, primary sector resilience, and traceable escalation. Open article 12 / 2025 Courts Caseflow intelligence without algorithmic justice AI can improve scheduling, document readiness, and backlog visibility while keeping legal determinations firmly with judges and authorised officers. Open article 11 / 2025 Procurement Public procurement AI should expose trade-offs, not hide them How agencies can use AI for market scans, tender analysis, risk review, and supplier questions without weakening probity. Open article 03 / 2026 Model Routing Similarity-first routing for sparse mixture-of-experts Why AbteeX treats expert routing as a coverage problem, not only a vector-distance problem: Tversky-style similarity, asymmetric evidence, and routing decisions that can be inspected after the run. Open article 11 / 2025 AgentOps Observability for agents: trace intent, not just tokens Token logs and span traces are necessary but incomplete. WeftlineAI records intent, evidence, proposed action, approval state, tool result, and rollback path as one operational object. Open article 10 / 2025 Sovereignty Why an AI lab from Aotearoa has a different centre of gravity A position note on building from Aotearoa for the world: local data rights, Pacific realities, Te Tiriti-aware governance, and a corpus that remains answerable to the place it came from. Open article
§ Lab · Aotearoa

Built by a small team in Aotearoa New Zealand.

AbteeX is a research lab, not a platform. We work with a handful of partners, publish openly, and answer to a place. Briefings, partnerships, and press inquiries are welcome.

Contact[email protected]
© AbteeX AI Labs · Aotearoa New Zealand
LumynaX WeftlineAI Responsible use Privacy Terms