Independence
People should be able to keep their knowledge and normal AI workflow on computers and in folders they control, without making a remote service the permanent owner of their work.
GI JI PHILOSOPHY · ILLUSTRATIVE MODEL
Independence, intelligence, privacy—and visible resource choices.
GI JI is a direction for products that keep people in control of their knowledge, make intelligent work inspectable and treat resource use as something to understand. The model below explores assumptions. It does not measure GI JI or prove that local AI is greener than cloud AI.
ONE DIRECTION, MORE THAN ONE APPLICATION
Five principles for the GI JI family.
These principles guide current work and possible future GI JI applications. They are not a promise that every application or capability already exists.
Human-directed intelligence
AI can retrieve, compare and propose. A person remains responsible for meaning: inspecting evidence, correcting interpretation and deciding what becomes reusable knowledge.
Privacy
Local work is the default. When a connected service is useful, the boundary should be visible and the person should know what would leave before approving it.
Resource awareness
Model size, context, retries, network transfer and idle work are real design choices. They should be visible instead of disappearing behind a vague AI label.
Green AI as a direction
Environmental responsibility begins with measurement and honest uncertainty. It is a direction for GI JI products, not a badge or a claim that every local workflow is automatically greener.
FOUR POSSIBLE WORK PATHS
There is no universal winner.
The appropriate path depends on the task, computer, model, evidence and privacy boundary. The calculator compares four hypothetical mixes instead of declaring one architecture good.
Cloud AI only
Retrieval and generation may happen remotely. Local work is light, while network and data-centre work remain part of the path.
Local retrieval + connected model
Relevant evidence is selected locally before an approved request uses an online model. The remote generation still counts.
Local retrieval + local model
Search and generation run on the person's computer. The result depends strongly on the machine, model, runtime and workload.
Retrieval-only evidence
The workflow retrieves passages without generative answer writing. It uses less assumed model inference but provides less synthesis.
INTERACTIVE SCENARIO LAB
Change the assumptions. See how the scenario changes.
Every number below is an illustrative approximation calculated in this browser. It is not live product activity, telemetry, a carbon saving, or an environmental footprint for GI JI.
Scenario—not actual GI JI impactMethodology eco-scenario-1.0 · illustrative construction values
Every result on this page is an illustrative scenario, not a measurement. Actual impact varies with hardware, model, query length, utilization, electricity mix, data-center efficiency, network conditions, and user behavior. This simulator uses illustrative assumptions, not measured GI JI values.
METHODOLOGY ECO-SCENARIO-1.0
The assumptions stay visible.
These modest construction values come from the existing Eco Impact specification. They exist to make the model work and can be replaced when comparable measurements exist. They are not measurements of GI JI, a named cloud provider or a customer.
FORMULA
Requests, overhead, then electricity.
People × daily requests × days creates the base volume. Each workflow share adds its illustrative retry and correction overhead before applying its assumed kWh per 1,000 workflow requests. Rounding happens only on screen.
UNKNOWN IN THIS VERSION
No carbon, water or lifecycle total.
Grid intensity, embodied hardware, model training, cooling, water, complete network infrastructure and hardware replacement are not calculated. No invented value replaces missing evidence.
PRIVACY
The controls stay in this browser.
The scenario script makes no network request and stores no inputs. Loading the page still creates the ordinary web-server access record described in the privacy notice.
MEASURE BEFORE CLAIMING
Green AI is a direction, not a score.
A useful environmental claim would need comparable workloads, measured system energy, named hardware and models, uncertainty and a clear lifecycle boundary. Until that exists, GI JI can publish its philosophy, assumptions and unknowns—and keep improving the questions it asks.