For SLPs, educators and communication partners
Partners model on the communicator's stable topology without authoring their message. The communicator remains the sole author of their own utterances.
EyeSay in about 90 seconds
The Voice Canvas has 48 stable motor positions. Forty-eight is the number of learned coordinates, not the expressive vocabulary ceiling: category positions can open additional language and then return to the same base map.
The Context Halo can surface changing possibilities beside the motor map. It does not move permanent vocabulary, and each suggestion remains a possibility until the communicator selects it. SOMETHING ELSE remains an explicit escape path.
The communicator remains the sole author of their own utterances. Suggestions and partner models remain possibilities until the communicator acts, and partner events stay outside communicator speech and history.
Model with EyeSay · quick start
Aided-language modelling means using the AAC system yourself while communicating naturally. We keep partner modelling technically separate from the communicator's own message: partners demonstrate on the same topology without authoring the communicator's message.
Model on the same stable Voice Canvas the communicator uses. A permanent position stays in the same motor location whether it is selected by the communicator or demonstrated by a partner.
Touch one or two useful words while you speak. You do not need to model every word in the sentence or turn the interaction into a lesson.
Demonstrate refusal, comments, questions, humour, social language, routine talk-back and self-advocacy as well as requesting.
There is no requirement to imitate, answer or prove understanding. A partner model is a demonstration, not a target response for the communicator to complete.
EyeSay treats learned motor coordinates as an invariant. Permanent positions are not re-ranked by context, routines or prediction. Additional vocabulary can be reached through expansion paths without silently relocating the base map. The design question is therefore not “which 48 words?” but “which stable coordinates form the motor-language plane, and how can more language be reached without moving them?”
The Context Halo is a separate adaptive plane for situationally useful possibilities. Suggestions remain possibilities until the communicator deliberately selects one. Nothing in the Halo becomes authored communication on its own, and contextual assistance cannot reorder the learned Voice Canvas.
Model Mode can create transient partner-side visual demonstrations on the mirrored topology. A partner model cannot become the communicator's utterance or become something the communicator is recorded as having said. Partner events cannot append to the communicator's utterance ribbon, cannot trigger speech on their device, cannot enter communicator history and cannot be counted as communicator selections. The separation is an implementation boundary, not merely an instruction to staff.
The current web build supports direct touch and keyboard-accessible controls, visible focus, high-contrast and reduced-motion patterns, and includes scanning work in development. Dedicated switch profiles, formal eye-gaze validation, keyguard-specific layouts and broad specialist-hardware compatibility should not be assumed complete until they are explicitly documented and tested.
See the Accessibility page for the current published capability and limitations.
Product-behaviour statements are supported by implementation and regression evidence. Broader AAC practice is not presented as EyeSay-specific outcome evidence, and hypotheses about possible communication or participation outcomes remain hypotheses until appropriate EyeSay-specific evaluation exists. Internal evaluation, external validation and independent replication are treated as different evidence states.
The Research and Evidence page carries the current public evidence boundary.
Step 1
Everyone supporting the communicator works from the same Voice Canvas topology. Do not reorganise learned positions to make a lesson, routine or partner prompt easier.
Step 2
Identify a few recurring moments across home, school and therapy where partners can model comments, questions, refusal, help, social language and requests without requiring imitation.
Step 3
Treat STOP, NO, HELP, BREAK and other self-advocacy language as communication. Do not make access conditional on task completion, compliance or a correct response.
Step 4
When communication breaks down, document the observable access or vocabulary gap. Avoid converting a search, pause or abandoned path into a claim about competence, emotion or intended meaning.
EyeSay is designed to support aided-language modelling as a communication-partner practice. It should not be used as a compliance task, a quiz or a demand for imitation. We do not claim that modelling guarantees a particular language outcome; EyeSay-specific effectiveness claims require EyeSay-specific evidence.
EyeSay is not asking clinicians to discard established AAC practice. The quickest orientation is architectural: a stable motor-language plane remains separate from a changing context plane; a permanent agency layer remains reachable; and partner modelling uses the same topology while remaining outside communicator authorship. Those are EyeSay design choices, not claims that another AAC system is unsafe or ineffective.
Take it to the team
Download the Model with EyeSay quick-start and the same-topology low-tech board. Both are public, versioned, screen-reader-readable HTML resources with A4 print layouts and no account requirement.