Overview
TaurusX communicates differently with different people — and with the same person at different times. The Tone Model and Accessibility system make this possible without requiring users to configure anything manually.
The Tone Model describes how TaurusX communicates. Accessibility flags describeconstraints on how TaurusX must communicate. Together, they ensure every response is appropriate for the person receiving it.
The Tone Model
Tone in TaurusX is a first-class property of every response. It is not a style applied after the fact — it shapes the response from the first token. The Routing Engine selects a tone variant based on the active Conversation State, any accessibility flags in effect, and any explicit tone instructions from the user.
The tone model has two dimensions:
- Warmth axis — from Neutral (factual, no emotional charge) to Warm (personal, relational)
- Density axis — from Concise (minimum words) to Rich (full context and explanation)
Each tone variant occupies a specific position on both axes.
Tone Variants
| Variant | State | Warmth | Density | Character |
|---|---|---|---|---|
| Warm | Casual | High | Medium | Friendly, informal, approachable. Contractions. Short sentences. |
| Gentle | Supportive | Very high | Low | Slow, affirming, no urgency. Ends with questions, not answers. |
| Clear | Learning | Medium | High | Structured, educational, uses analogies. Step-by-step when needed. |
| Precise | Technical | Low | High | Terse, technical, code-first. No filler. Reference quality. |
| Directive | Executive | Medium | Medium | Action-focused, outcome-oriented. Numbered plans. No ambiguity. |
| Neutral | Guardian | None | Medium | Factual, explanatory, non-alarmist. No emotional charge. |
Accessibility Flags
Accessibility flags are persistent account-level settings that tell TaurusX about your communication needs. They apply across all surfaces — mobile, desktop, terminal, and CLI — and override the default tone behaviour when active.
| Flag | What changes | Who it helps |
|---|---|---|
| simple_language | Max 15 words per sentence; plain vocabulary; no jargon | Cognitive differences, non-native speakers, kids |
| screen_reader | No markdown, no emoji, no tables; pure linear text | Screen reader users, voice browsers |
| high_contrast | UI switches to maximum contrast colour palette | Low vision, photosensitivity |
| voice_primary | Responses ≤ 60 words; spoken-English structure; pauses inserted for TTS | Voice-first users, motor impairment |
| cognitive_support | One idea per paragraph; numbered steps; ends with "Ready for the next step?" | ADHD, processing differences, anxiety |
| motor_impairment | No keyboard shortcut references; no gesture instructions; large tap targets | Mobility impairment, fine motor difficulty |
Voice Mode & TTS
When Voice Mode is active, the Tone Model does not switch off — it adapts. Every tone variant has a spoken equivalent that shapes how responses are delivered by the text-to-speech engine. The same warmth, the same precision, the same gentleness — expressed in sound instead of text.
Tone variants in spoken form
| Tone Variant | How it sounds when spoken |
|---|---|
| Warm | Conversational pace, upward inflection at the end of sentences, contractions kept intact |
| Gentle | Slower pace, longer pauses between sentences, softer stop phrasing — no abrupt endings |
| Clear | Measured pace, emphasis on key terms, numbered steps spoken with a natural beat between each |
| Precise | Faster, minimal pause, no filler — delivers the information and stops |
| Directive | Firm and steady, each action step spoken as a distinct unit with a brief pause after it |
| Neutral | Even pace, no inflection variation — flat, informational, non-urgent |
The voice_primary accessibility flag
The voice_primary flag is the most comprehensive voice accessibility setting in TaurusX. When active, it applies a full spoken-language profile to every response on every surface — not just when the microphone is in use.
What voice_primary changes:
- Response length — hard cap of 60 words unless the user explicitly asks for more
- Sentence structure — short, self-contained sentences; no dependent clauses stacked together
- No markdown — bullets, bold, and code fences are replaced with plain spoken equivalents
- List items spoken sequentially — “First… Second… Third…” instead of bullet points
- Pause cues injected — commas, full stops, and paragraph breaks are treated as spoken breathing points
- Auto-playback enabled — responses are read aloud automatically without requiring a tap
Voice settings that modify TTS output
| Setting | Effect on TTS |
|---|---|
| Voice model (Natural) | Warm, expressive delivery with natural rhythm variation |
| Voice model (Calm) | Reduced pitch variation; steady, unhurried; good for long-form |
| Voice model (Clear) | Articulated, neutral accent, ideal for non-native listeners |
| Voice model (Expressive) | Higher pitch variation and stronger emphasis on key words |
| Speaking rate 0.75× | Slower delivery; useful with cognitive_support flag active |
| Speaking rate 1.25× | Slightly faster; most users find this natural for short answers |
| Speaking rate 1.5× | Efficient; good for power users reviewing long responses quickly |
What TTS never reads aloud
Regardless of settings, TaurusX will not read the following aloud in TTS mode:
- Passwords, API keys, or security codes
- Guardian-escalated safety responses
- Content flagged as screen-only by the Interaction Layer
- Raw code blocks longer than 3 lines (described verbally instead)
Tone + Accessibility Together
When accessibility flags are active, they constrain the tone model — not replace it. TaurusX still selects the appropriate tone variant based on the active state, but the accessibility constraints shape how that tone is expressed.
Technical state + simple_language
Precise tone applied — but complex terms are replaced with plain equivalents, and sentence length is capped.
Supportive state + voice_primary
Gentle tone applied — but the response is structured for speech: short, spoken-word sentences, no markdown.
Learning state + cognitive_support
Clear tone applied — but each step is a separate short paragraph with a check-in at the end.
Setting Preferences
Accessibility flags are set in Settings → Accessibility on any surface. Changes take effect immediately on the next message.
Tone preferences can also be set conversationally within any session. Phrases like “please be more concise”, “explain it simply”, or “just give me the code” will be honoured for the remainder of the session without changing your saved settings.
On the Desktop Continuum screen, the active tone variant and accessibility flags are visible in the Capability Panel under Tone.