MyVoice Studio

MyVoice Studio is the control center for how AI writes as Daniel. It manages voice profiles, verified facts, special instructions, and provides a voice training lab — a “voice IDE” where you design, test, and deploy AI personalities.

Module: modules/myvoice-studio/ (standalone Express service, port 8098) Status: ~40% complete — backend routes work, frontend basic, LLM transform and RAG explorer not wired

Vision

Every AI interaction on ArgoBox — blog generation, journal entries, public chatbot, LinkedIn posts — goes through a voice profile. MyVoice Studio is where those profiles are created, tested, and refined.

It’s powered by RAG data from 701K indexed chunks and 11M+ words of real conversation history.

The 5 Tabs

1. Voice Profiles

Browse, create, and edit voice personas. Each profile is a complete personality spec:

Field	Description
Name	e.g., “ArgoBox Casual”, “LaForce IT Professional”
Voice Level	1-10 slider (1 = formal resume, 10 = 11 PM Discord)
Key Phrases	Signature expressions (“Look, I’ll be honest…”)
Do’s	Rules to follow (use first person, include timestamps)
Don’ts	Rules to avoid (no “we”, no corporate transitions)
Example Text	Before/after transformations
Tone Dials	Humor, Technical Depth, Self-Deprecation, Formality, Enthusiasm, Storytelling (each 0-10)
System Prompt	Full system prompt injected into all AI calls
Formula	Writing formula pattern
Contexts	Where this profile applies (blog, journal, linkedin, chat)

Current Profiles:

argobox-casual.json — Primary blog/journal voice (active)
laforceit-professional.json — Professional/resume tone
linkedin.json — LinkedIn post formatting
resume-formal.json — Formal documentation tone

2. Voice Lab

The creative testing ground:

Score — Paste text, score against active profile (0-100 authenticity). Shows which sentences match and which feel “off” with inline highlighting.
Transform — Paste generic text, click Transform, get it rewritten in the selected voice using LLM. Side-by-side before/after view.
Compare — Same text in two different profiles side by side. “How does this read as ArgoBox vs LaForce IT?”
Voice DNA — Analyze a text sample: sentence length distribution, vocabulary diversity, first-person percentage, hedging frequency, humor density. Radar chart visualization.

3. Facts & Identity

CRUD interface for verified facts about Daniel:

Status	Color	Meaning
Verified	Green	Confirmed via ground truth interview
Unverified	Yellow	Not confirmed yet — use cautiously
Denied	Red	AI-fabricated — must NEVER be used

Features: add/edit/archive facts, source tracking, conflict detection.

Current denied claims (hardcoded guardrails):

the remote site is NOT “40 miles away” — distance unspecified
Daniel has NEVER used Traefik — “463 messages” story is fabricated
apkg was NOT “built by accident” — built intentionally
Build swarm is 66 cores, NOT 62

4. Instructions Hub

Centralized system prompt management:

Rich markdown editor for special instructions
Context tags: “always”, “blog-only”, “code-only”, “professional”
Version history with rollback
Export formats: CLAUDE.md block, system prompt JSON, markdown
Pull endpoint: GET /api/instructions/latest — any AI tool can consume the latest instructions

5. RAG Explorer

Search 701K chunks for voice patterns:

Voice search: “How do I talk about Tailscale?” → shows real examples
Pattern extraction: recurring phrases, sentence structures, humor patterns
Collection filter: conversations, blog posts, journal entries
Phrase frequency: which phrases appear most in your writing
Save to profile: found a good phrase? Add it directly to a voice profile

Architecture

MyVoice Studio (port 8098)
├── GET/POST /api/profiles      — Voice profile CRUD
├── GET/POST /api/facts         — Facts CRUD (verified/unverified/denied)
├── GET/POST /api/instructions  — System prompt instructions CRUD
├── POST /api/lab/score         — Score text against profile
├── POST /api/lab/transform     — LLM-powered voice transformation (TODO)
├── GET /api/lab/rag-search     — RAG proxy for voice patterns (TODO)
└── GET /api/health             — Health check

SQLite Database: data/voice-studio.db
├── profiles table
├── facts table
└── instructions table

Integration with ArgoBox

MyVoice Studio feeds into every AI interaction via src/lib/myvoice-client.ts:

MyVoice Studio (port 8098)
    ↓ getActiveProfile()
myvoice-client.ts (5-min cache)
    ↓ formatVoiceForPrompt()
assistant-prompt.ts / voice-profile.ts
    ↓
Content-Gen API / Public Chat / Argonaut

Fallback: If MyVoice Studio is unreachable (e.g., production on Cloudflare Workers), the system uses data/active-voice-profile.json — a static snapshot synced by scripts/sync-myvoice-profile.ts.

Voice Profile Data Model

interface MyVoiceProfile {
  id: string;
  name: string;
  description: string;
  voiceLevel: number;        // 1-10
  active: boolean;
  audience: string;
  toneDials: {
    humor: number;           // 0-10
    technicalDepth: number;
    selfDeprecation: number;
    formality: number;
    enthusiasm: number;
    storytelling: number;
  };
  keyPhrases: string[];
  dos: string[];
  donts: string[];
  exampleBefore: string;
  exampleAfter: string;
  systemPrompt: string;
  formula: string;
  contexts: string[];
}

Voice Scoring Algorithm

The voice scoring engine (src/lib/voice-engine.ts) uses 4 weighted metrics:

Sentence Length (20%) — Target: 13.2 words average (Daniel’s natural rhythm from 2,577 conversations)
Banned Patterns (35%) — 15+ regex patterns for AI-sounding phrases
First Person (20%) — Ratio of “I/me/my” vs “we/our/us” (target: 80%+)
Casual Tone (25%) — Fragments + contractions + ellipsis vs semicolons + passive voice

Overall score = weighted average of all 4 metrics.

Data Sources

Source	What It Provides	Location
Voice Profile	Tone dials, key phrases, dos/donts, system prompt	`data/active-voice-profile.json`
Identity Ground Truth	Verified facts about Daniel (v2.0)	`data/identity-ground-truth.json`
RAG Knowledge Base	701K chunks from 22 collections	`/mnt/AllShare/rag/databases/`
daniel-BLOGGER-VOICE-PROFILE.md	Data-driven metrics from conversations	Vaults backup
DANIEL-LAFORCE-FACTS.md	Comprehensive verified facts	Vaults backup
Conversation Archive	2,992 processed conversations	`Vaults/conversation-archive/`

Replicability

MyVoice Studio is designed to be replicable — anyone can run it for their own voice cloning:

Run the interview template (docs/MYVOICE-INTERVIEW-TEMPLATE.md)
Deploy MyVoice Studio via Docker
Seed profile from interview answers
Test in Voice Lab
Connect to any AI system via pull endpoint

Key Files

File	Purpose
`modules/myvoice-studio/DESIGN.md`	Full design document
`modules/myvoice-studio/src/server.ts`	Express app
`modules/myvoice-studio/src/db.ts`	SQLite database
`modules/myvoice-studio/src/routes/profiles.ts`	Profile CRUD
`modules/myvoice-studio/src/routes/facts.ts`	Facts CRUD
`modules/myvoice-studio/src/routes/instructions.ts`	Instructions CRUD
`modules/myvoice-studio/src/routes/lab.ts`	Voice Lab (scoring)
`modules/myvoice-studio/data/voice-profiles/`	4 JSON profile files
`src/lib/myvoice-client.ts`	ArgoBox API client
`src/lib/voice-engine.ts`	Voice scoring algorithm
`src/lib/voice-profile.ts`	System prompt builder
`data/active-voice-profile.json`	Static profile snapshot
`scripts/sync-myvoice-profile.ts`	Profile sync script

Content Studio — The editor that consumes voice profiles
Content Generation — How voice profiles are injected into generation
RAG System — The knowledge base that powers fact checking and voice pattern search