One line to the fleet

Swipe ↓ from the Deck (admin only): a direct, one-to-one conversation with Gia — the Jarvis of the fleet — with Mia selectable. Everything that flows to Telegram mirrors here and vice-versa: one conversation, two doors. The noise problem is solved by per-topic response levels, not by muting the whole line.

Gia● listening · mirrored ↔ Telegram
Gia ▾ | Mia
⚙ python shell · 3 events summarized · tap to expand
Morning brief is out. LinkedIn token renews itself tonight — no action. One thing needs you: the Replit export bundle triage is still waiting.
09:12 · also sent to Telegram ✓
what did the fleet audit find overnight?
All 23 domains green. One screenshot went stale on city.arnao.ai — re-shot, fixed, no cert issues. Cost ledger: $4.10 yesterday, within budget.
09:14 · summary level: costs=verbose
💸 cost controls · verbose — you asked to see every dollar
python shell SUM coding SUM deploys VERB costs VERB cron chatter IGN

Per-topic three-state filters — VERBOSE / SUMMARY / IGNORE. Ignored topics still log; summaries collapse into expandable one-liners; nothing is lost, only quieted.

Gia ● speaking — cloned voice "Three things this morning, Byron. First, the deck…"
🎙

Voice mode: the same line, hands-free. The pod is the biggest control; the avatar carries presence. Disclosure line applies whenever the cloned voice speaks to anyone but you.

Avatar — staged honestly

STAGE A · NOW
Presence orb

breathes when idle, pulses to speech amplitude, cloned Chatterbox voice. 100% reliable — ships with the Jarvis line.

weeks · reliable
STAGE B · NEXT
Viseme portrait

stylized portrait (your existing avatar art), amplitude-driven mouth + blink. Feels alive, still deterministic.

1–2 sprints
STAGE C · EARNED
🎬
Rendered talking head

photoreal lip-sync (LivePortrait-class), pre-rendered for briefs first, real-time only when local GPU proves it. Not promised until demoed.

research gate

Plumbing (all existing parts)

PieceSource
Message bus + mirrorthe Telegram bot Gia already runs — the Jarvis line is a second client on the same conversation, so Telegram ↔ harness mirroring is by construction, not sync code
Noise policythe shipped Telegram noise contract (guards silent-on-success) generalized into per-topic VERB/SUM/IGN levels stored per client
Voicelocal Chatterbox clone + pronunciation lexicon + whisper-back QA (the standing public-voice pipeline)
Agent switchGia ▾ / Mia — same pane, different endpoint; Mia stays local-LLM
Noise stance: the Jarvis line is Byron-only and calm by default — everything starts at SUMMARY except costs and deploys (VERBOSE), cron chatter starts IGNORED. Changing a level is one tap on the filter bar, and takes effect for both doors (chat + Telegram) at once.

← VCM · Back to proposal ↩