Finding
Voice and TTS support becomes operationally weak when agents write normal chat-length updates for an audio-first channel instead of designing responses for short, listenable status delivery.
Current
A real Hermes installation may have working text-to-speech, Telegram voice delivery, or voice-adjacent tooling, but the response style can still be optimized for reading rather than listening. The weak point is usually not the voice feature itself; it is that long explanations, dense formatting, repeated caveats, and oversized status updates create poor TTS output. This makes voice updates harder to understand, harder to act on, and more likely to be skipped.
Suggested
- Add a voice-response length rule for TTS channels. Exact change: add a “Voice/TTS response rule” to
SOUL.mdor the main profile instructions: “When responding through Telegram, voice, or TTS-oriented contexts, keep updates short, use plain sentences, avoid large lists unless requested, and split longer content into separate messages.” - Create a TTS-friendly status template for operational updates. Exact change: add a runbook or dashboard copy block named “TTS status format” with this structure: “Status, action taken, blocker if any, next decision,” limited to four short lines for routine Hermes updates.
- Add a verification habit before sending voice-oriented summaries. Exact change: update the relevant communication skill or completion checklist with: “Before final response on TTS-sensitive tasks, read the answer as spoken audio; remove visual-only formatting, long nested bullets, repeated URLs, raw logs, and unnecessary detail.”
Impact
This makes Hermes more useful in mobile and voice-first operation, especially when updates arrive through Telegram or another channel where the user is listening instead of reading. Shorter voice responses reduce cognitive load, prevent important decisions from being buried, and make status updates easier to act on quickly. It also lowers token use because routine updates become concise by default while still allowing deeper detail when explicitly requested.
Effort
Small — the improvement is mainly a profile instruction patch, one reusable status template, and a final-response verification habit. No new voice infrastructure is required if TTS or voice delivery already works.
Public page note
Safe public content includes the TTS operating principle, generic voice-response rules, example status structure, and the maturity benefit of concise audio-first communication. Internal-only content includes private voice transcripts, raw chat excerpts, actual Telegram messages, user-specific preferences, credentials, gateway settings, TTS provider details, logs, and environment values.