Interview, meeting, coding, translator, presentation, sales, study, assistant, and emergency cue modes.
This build uses Supabase/Postgres for app state and session persistence. Hardware can still be simulated until real glasses/ring adapters are connected.
Each mode defines its prompt intent, mapped cue engine, HUD template, latency target, memory scope, and privacy default. This keeps the live HUD from becoming one generic assistant.
Answer pressure questions with short, senior-level framing.
Use project evidence, be concrete, avoid exaggeration, and keep the cue speakable.
Summarize what to say next without interrupting the conversation.
Give one concise next response, action, or clarifying question.
Give precise technical explanations, debugging steps, and implementation framing.
Prefer route names, env keys, file paths, tradeoffs, and exact next steps.
Translate or rephrase quickly for live conversation.
Translate the user intent, preserve tone, and keep it short enough for a lens.
Show one clean line at a time for prepared speaking.
Convert notes into short speakable beats with no paragraphs.
Keep the speaker on track with timing, transitions, and confidence cues.
Show current point, next transition, or timing warning.
Help with objection handling, value framing, and follow-up prompts.
Answer directly, tie value to customer need, and ask a clear next question.
Turn questions into compact explanations and recall prompts.
Explain the concept simply, then offer one memory hook or next step.
Handle lightweight reminders, replies, and context cues.
Be brief, practical, and avoid saving sensitive details unless approved.
Give the safest immediate phrase when the user is stuck.
Buy time, reduce risk, and give a calm next sentence.