aime™ unveils aimeFUSION™ — the orchestration intelligence that fuses the education AI stack

aime today introduced aimeFUSION™, the orchestration intelligence that fuses the full aime stack — from a compact reasoning model running on a teacher's laptop to a national supercomputer running in-country — into a single education system. aimeFUSION is the layer that decides, for every task, how much intelligence to apply and where to apply it.

The problem with one big model in the cloud

Most AI in education today is a single large model behind an API in a distant data centre. Every learner question travels out and back. It is a clean design until it meets a real classroom in a real province: the connection is intermittent, the power is not guaranteed, the per-token bill grows with every learner, and data about children leaves the country to be processed. That architecture puts intelligence as far as possible from where it is needed, and makes it most fragile exactly where conditions are hardest.

aime takes the opposite position. Its foundation is efficient small reasoning models, structured knowledge, agent-driven workflows and offline-first delivery — intelligence that lives close to the learner. Distributing intelligence across many locations creates a hard coordination problem; aimeFUSION is the layer that answers it.

Three tiers, one system

• Device tier — aimeTEACH and aimeJAM run Leo 4B (teachers) and Leo 2B (students) directly on the CPU, in 2–4 GB of RAM, with no GPU and no internet required.
• Classroom tier — aimeHUB runs on the OPS compute module seated in the interactive panel's OPS slot, with on-board NPU acceleration, the full agent fleet and a classroom content library served to fifty or more devices over the room's own network.
• Cloud tier — aimeCLOUD™ runs on an in-country NVIDIA DGX supercomputer for the heaviest reasoning, content generation, model fine-tuning and the national Intelligence Graphs that connect every classroom into one evolving picture.

These are not three products. They are one system expressed at three scales — and aimeFUSION is what makes them behave that way.

Vertical fusion — compute, data and sync

Vertically, aimeFUSION fuses the stack along three dimensions at once, with one strict design rule: each must have a local version, not just the model. A hybrid system that keeps only the model local but still reaches across the internet for its data or to save its work is not offline-capable — it only looks that way until the link drops.

• Fusion of compute — the same intelligence is available at three scales, and a task runs as far up the stack as it needs to and no further. Kern™, aime's micro-agent framework, lets identical agent logic execute on a 2B model on a laptop CPU and on a frontier model in the cloud.
• Fusion of data — ThinkBook™, aime's structured, library-inspired knowledge architecture, is replicated and cached down the stack, so the device and the Hub already hold what the classroom actually uses. Lessons travel as self-contained .aimepack™ files, rendered offline.
• Fusion of sync — Loom™, aime's orchestration engine, runs durable, resumable workflows that tolerate the connection coming and going. Work done offline queues locally and reconciles automatically the moment a link returns.

How aimeFUSION decides

For every request, aimeFUSION walks a cheapest-first decision path: deterministic builders before any model; ThinkCache™ for repeated and near-repeated requests; Leo on the device for fresh inference; the Hub's larger model on the OPS NPU for heavier work; and only the hardest tasks — deep reasoning, large generation, cross-school intelligence — reach aimeCLOUD. Connectivity is an input to that decision, never a precondition. Every feature has a defined offline behaviour — sometimes at reduced quality, but never an error screen.

“Teaching continues fully offline at good-enough quality, and quality improves when the connection returns. That is the promise of aimeFUSION — precise, honest, and built into every feature as a contract, not an aspiration.”

Why vertical fusion matters

• Responsiveness — most interactions resolve on the device or in the room, in the time it takes to think.
• Economics — small-first means the majority of inference runs on hardware already present in the classroom, at no marginal cost per query. The expensive supercomputer tier is reserved for the few percent of work that genuinely needs it.
• Privacy and sovereignty — most learner interaction never leaves the device; what does reach the cloud stays in-country. Sovereignty is a property of the architecture, not a policy bolted on top.
• Resilience — the classroom is independent of the link. An outage or a dead connection slows nothing and stops no lesson.
• Quality — because ThinkBook™ grounds every answer in structured curriculum knowledge and EduRule™ encodes how a concept should be taught, a small model's output is curriculum-accurate and pedagogically sound. The cloud raises the ceiling without ever being the floor.

Horizontal fusion — a model fabric in the cloud

Vertical fusion decides which tier should handle a task. Horizontal fusion decides which model should handle it — and at the cloud tier, that is rarely a single model. aimeCLOUD runs a fabric of specialised models, each chosen for the kind of work it does best, and aimeFUSION routes across them and composes their outputs: agentic reasoning and orchestration through Kern™ and Loom™; STEM problem solving on models tuned for it; educational visuals from aime's Text-to-Diagram pipeline rather than a generic image generator; interactive simulations produced as runnable code; and conversational tutoring drawn from the model best matched to the subject and the language, including Filipino and regional languages.

A single lesson is often the product of several of these at once — explanatory text from one model, a diagram from Text-to-Diagram, a runnable simulation from a code model, an assessment from another — assembled by aimeFUSION into one coherent .aimepack™. This is composition, not merely selection.

“Vertical fusion places the right amount of intelligence in the right place. Horizontal fusion places the right kind of intelligence on each task. aimeFUSION operates on both at once.”

Continuing innovation

The specific routing policies, escalation thresholds, reconciliation rules, model-selection heuristics, composition strategies and offline-tier contracts that make aimeFUSION work are aime's proprietary IP and are shared with partners under NDA. aimeFUSION extends concepts already proven inside the Loom Workflow Engine™ and the Kern micro-agent framework, and is what allows a small-model-first, offline-capable, sovereign education AI to be viable at national scale — not as a compromise, but as a better architecture for the conditions real schools face.

Availability

aimeFUSION is in phased rollout across the aime stack from June 2026 — coordinating Leo, ThinkCache™, Kern, Loom, ThinkBook, EduRule™, Text-to-Diagram, .aimepack and the AIME Renderer across the device, Hub and cloud tiers, and sequenced inside the aime Loom Workflow Engine for deterministic, replayable execution.

About aime

aime builds the operating system for educational intelligence — the foundational infrastructure layer that future education systems will run on. aime's stack combines structured curriculum knowledge, pedagogy-aware reasoning, compact education-tuned models, agentic orchestration and offline-capable deployment, and is designed for ministries of education, universities and national school systems.