Mainstream AI
Model
Product
Deployment
Adaptation
A systems-level framework for AI
AI, thought differently.
Constraint-native. Non-extractive. Viable beyond the cloud.
LowRes.AI is a systems-design framework for two connected challenges: making AI viable where the usual operating assumptions fail, and preserving human agency over technology and knowledge.
Start with reality01 / 13
Start with the operating environment
A system is only intelligent if the whole system can work where it is needed. LowRes.AI treats six interdependent dimensions as design inputs from the beginning.
01Connectivity
Whole-System Viability
02Compute + Power
03Data + Knowledge
04Interaction
05Operational Capacity
06Institutional Capacity
02 / 13
The constraint-native shift
The order of design matters. Instead of adapting a finished AI product to constrained settings, begin with the environment and the outcome people actually need.
LowRes.AI
Operating environment
Required outcome
Viable system
Appropriate AI
03 / 13
More than a language problem
Better support for low-resource languages is essential. It is not, by itself, enough.
A model may understand a language and still fail the people who speak it. Weak connectivity, limited electricity, low literacy, absent technical support, recurring subscriptions and inappropriate governance can make an otherwise capable system unusable.
Non-extractive by designIndigenous and community knowledge should not simply become another dataset to scrape. Knowledge may need to remain local, attributable, governed by its holders and outside model training.
04 / 13
Beyond the subscription model
Access assumptions shape architecture. A viable system reduces the number of dependencies that must keep working forever.
Default stack
accountemailcloudsubscriptioncontinuous connectivity
VERSUS
Constraint-native stack
shared accesslocal infrastructureoffline capabilitylocal or institutional ownershipminimal recurring dependency
The aim is minimum sufficient intelligence: enough capability for the required outcome, with fewer dependencies on remote infrastructure, recurring payments and specialist maintenance.
05 / 13
What qualifies as LowRes.AI?
Multiple constraints are first-class design inputs.
Those constraints materially change the architecture.
The system has a strategy for failure of mainstream assumptions.
06 / 13
Design principles
Not a checklist of features. A discipline for deciding what the system should—and should not—be.
- 01Constraint-first
- 02Minimum sufficient intelligence
- 03Graceful degradation
- 04Human-reality interfaces
- 05Bounded and attributable knowledge
- 06Lifecycle viability
- 07Human and local agency
07 / 13
Access is not agency.
AI concentrates the capability to turn codified knowledge into scalable technological power.
Wider access to that capability matters. But access to an AI service does not, by itself, redistribute authority over knowledge, infrastructure or technological decisions.
People may be able to use a system while having little say over what it does with their knowledge, who controls its underlying infrastructure, or which choices are made on their behalf. The question is not only who can use AI, but who can govern what it may do.
08 / 13
THE GOVERNING PRINCIPLE
The Membrane
Capability may flow in without knowledge having to flow out.
The Membrane is a principle of governed permeability. It allows AI capability and trusted external knowledge into locally controlled knowledge environments without making extraction the price of participation.
People and institutions with legitimate authority define what may enter, what AI may access or transform, and what may leave. Technical architecture supports those decisions; it does not replace human governance. Knowledge can remain under local authority while approved outputs move outward.
Explore The Membrane →09 / 13
FROM PRINCIPLE TO PRACTICE
What if AI came in a capsule?
Encapsulate AI. Keep knowledge in human hands.
AI Capsules are self-contained, locally controlled AI systems combining trusted Knowledge Libraries, local AI capability and human-governed workflows. They bring useful capability into a governed knowledge environment while keeping authority over local sources with the people responsible for them.
The architecture is model- and hardware-agnostic. AI Capsule 1.0 is currently under development, not yet deployed or commercially available.
We are also exploring how AI Capsules could support community communication: distinct knowledge sources, AI-supported dialogue, human review and distribution through community channels. Explore the workflow →
Explore AI Capsules →PUBLIC TECHNICAL PREVIEW · OCTOBER 2026
AI Capsule — Technical Preview
The published Qwen2.5 Technical Preview documents an experimental local, multilingual, source-grounded knowledge assistant, including a fictional demonstration and known limitations. It is shared for technical scrutiny, not as a ready-to-install application or a completed AI Capsule 1.0. The newer, separate Qwen3 dialogue experiment is not included in this public repository.
10 / 13
One framework, many domains
LowRes.AI is a systems-design framework, not a single implementation. The constraints differ across domains; the discipline of whole-system viability and human agency travels.
Climate resilience
Agriculture
Health
Humanitarian response
Education
Public services
11 / 13
Why LowRes.AI?
Because the people most poorly served by mainstream digital systems should not have to wait for ideal infrastructure—or surrender authority over knowledge to use AI.
10+years of Audiopedia Foundation practice designing accessible knowledge systems where literacy, connectivity, language and devices cannot be assumed.
“LowRes.AI did not begin with an abstract question about what AI might become. It began with years of practical work asking what knowledge systems must do when the usual assumptions fail.
Today, we're extending that question: How can AI become useful under those conditions while strengthening human agency and preserving control over knowledge?
AI Capsules are our first attempt to put that thinking into practice.”
13 / 13
Notes from the field
Ideas, tensions and lessons emerging as the framework meets practice.
Agency First
Those with the least power to shape AI should be where we begin.
Why I Started LowRes.AI
From accessible knowledge to constraint-native intelligence
Researchers · Implementers · Funders · Organisations
Let’s compare notes.
If you are working on AI in constrained environments—or questioning the assumptions built into today’s systems—we would like to hear from you.
Start a conversation