Loading research content…
Loading research content…
A two-face spatial workfield for moving from intent to governed, human-accepted work.
This is the v0.5 standalone browser build. It begins with empty local containers: no account, repository, chat, file, or project adapter is connected. Anything you capture exists only in this page's memory and disappears on reload.
Select DWELL to exercise the gaze-selection state machine with the pointer. An eye-tracker adapter may feed viewport coordinates through CorollaGaze.feed(x, y); this public build does not start a camera, transmit gaze data, or claim a completed hardware eye-tracking integration.
This compact harness exercises the authored v0.5 states and reset path without replacing the full interactive workfield above.
751265ccEach control executes a declared browser-local transition. Repeated dual-outcome controls alternate deterministically.
NOT YET EXERCISEDField and Making reuse the same six-axis spatial topology while changing semantic role.
NOT YET EXERCISEDDwell nomination traverses targeting, locked, and dissipating states before activation.
NOT YET EXERCISEDNo account, repository, chat, file, or project adapter is connected in the public fixture.
NOT YET EXERCISEDReturned work never becomes Done without a separate human verdict.
NOT YET EXERCISEDThe default public path performs no external model request until a user configures an endpoint.
No transition yet. Operate a declared control.
This transparent browser-local instrument teaches the authored state contract. Exercising an assertion demonstrates fixture behavior only; it is not evidence of production performance, compression ratio, model uplift, cognition, safety, or universal correctness.
Corolla is an intent-centered spatial workfield. It replaces the idea that a serious AI workspace should be a chat transcript, file tree, or dashboard of disconnected applications.
The stable object is a six-axis bloom. The meaning of that bloom can turn over without forcing the user to learn a second geometry. One face helps a person locate the state of existing work. The other helps them declare what should exist next.
The operating sequence is:
locate → nominate → shape → route → return → human verdict
Language may propose and organize the work. It cannot silently mark the work complete.
The Field face is the operational-memory view:
The public fixture starts with empty browser-local containers. No actual account, chat, repository, file system, or project is connected. A production adapter would project governed objects into these locations while canonical state remained outside the visual surface.
The Making face reuses the same six spatial axes as contribution modes:
An intent also receives a human-scope coordinate from Myself through Family, Friends, Contacts, Neighborhood, Town or City, Country, and Earth. The result is not merely a category. It is a coordinate that changes the trust basis, privacy expectation, human radius, plan language, and acceptance criteria.
Typing an intent can nominate a mode immediately. Entering that axis folds the spatial array into a shaping surface: intent, scope, contribution mode, five-step plan, provenance, readiness, and Chat / Plan / Make lane. Committing folds the surface back into the workfield as a bounded job.
Jobs move from BUILDING to RETURNED. They do not move directly to Done.
A returned result requires one of three human verdicts:
That distinction is load-bearing. A builder may return a result. Only a human acceptance act closes the work order.
The interaction state machine is input-agnostic:
IDLE → TARGETING → LOCKED → IDLE
↘ DISSIPATING
The current public build includes:
window.CorollaGaze.feed(clientX, clientY).An eye tracker can feed the same viewport coordinates used by the selection machine. The embedded public build does not start a camera, infer gaze by itself, transmit gaze data, or claim completed hardware validation. It demonstrates the gaze/dwell interaction contract and its adapter boundary.
Consequential completion remains separately gated by the returned-work verdict.
Corolla can shape plans through two optional OpenAI-compatible engines:
If neither engine is configured or reachable, Corolla falls back to built-in deterministic mode patterns and labels that provenance honestly as template. The public fixture configures no endpoint by default. API keys, if a user enters one for evaluation, remain in memory only.
The standalone build also demonstrates:
All fixture state is ephemeral. Reloading the page resets it.
Radial menus, gaze input, dwell selection, project dashboards, workflow states, and AI plan generation all have substantial prior art. Glyphd does not claim those fields.
The candidate contribution is the specific arrangement:
This is a dated technical contribution statement, not a legal conclusion that any patent claim is novel, non-obvious, eligible, valid, or enforceable.
The repository is the canonical implementation home and is currently access-controlled. The single-file browser build above is the public, directly downloadable evaluation artifact.
IMPLEMENTED for the standalone v0.5 interaction fixture.
SPECIFIED for production adapters, durable project state, HID, and hardware eye-tracker integration.
AWAITING_LOCAL_RUN for measured gaze accuracy, false-commit rate, correction time, task time, accessibility performance, and productivity comparisons.
This page publishes the visible interaction grammar, public evaluation source, behavioral contract, and dated authorship record. It excludes nonpublic adapter credentials, personal project state, patent-claim drafting, and restricted implementation refinements.
Public access is not an open-source grant. The public demo may be run unmodified for evaluation under the applicable Glyphd proprietary license. No patent, implementation, derivative-work, model-training, redistribution, or commercial-use license is granted.