feat: Voice Persona Chat — live Venice character conversations

[hassan1731996]

What This Does

Adds voice-persona-chat — the first OpenHome ability to use Venice.ai's native character_slug API parameter. Users talk to historical and public figures by voice: Einstein explains relativity in his own style, Socrates asks probing questions, Alan Watts riffs on consciousness. The character engine is Venice's built-in persona system, not a generic LLM with a text system prompt.

Why It's Unique

Nothing in the catalog generates character-authentic responses at this level. dungeon-master-voice gets closest but injects personality via a text system prompt into the generic OpenHome LLM — the same model that self-censors and lacks real character knowledge depth. Venice's character_slug applies a trained persona that knows the character's actual biography, writings, speech patterns, and worldview.

How It Works

Three entry paths ordered fastest to slowest:

1. Name in trigger — "talk to Einstein" → matches inline → one confirm → chat starts
2. Returning user — remembers last character → "Want to talk to Einstein again?" → one confirm → chat starts
3. Full selection — live roster fetched from Venice API → user speaks name → LLM fuzzy match → confirm → chat starts

Character greets the user with a self-introduction on connect. Exit with "stop", "goodbye", or "that's all" — the character delivers an in-character farewell.

Technical Notes

• Venice API calls: GET /characters for live roster, POST /chat/completions with venice_parameters.character_slug
• Roster is live — no hardcoded character list, grows automatically as Venice adds new characters
• Storage: one key (voice_persona_chat_prefs) with create-first upsert pattern
• No background daemon required
• SDK compliance: no asyncio, no print(), no open(), resume_normal_flow() in finally

Patterns Applied

• get_single_key unwrap: .get("value", raw) ✅
• Storage upsert: create_key + success check → update_key ✅
• LLM guards: all text_to_text_response() calls wrapped in (... or "") ✅
• resume_normal_flow() in finally block ✅
• Two-tier exit detection: token match for short words, phrase match for multi-word ✅
• No asyncio, no print(), no hardcoded keys ✅

Setup Required

Users need a Venice API key added to OpenHome settings under venice_api_key.

Trigger Phrases

• "talk to Einstein" / "chat with Socrates"
• "let me talk to Alan Watts"
• "persona chat" / "character chat" / "voice persona"
• "connect me to Shakespeare"

Test Plan

• Trigger with name ("talk to Einstein") → skips roster, confirms, connects
• Trigger without name → roster spoken → name selected → confirmed → chat starts
• Return session → "Want to talk to X again?" → yes → instant connect
• Chat loop: multi-turn responses in character style
• Exit words ("stop", "goodbye") → in-character farewell → clean exit
• Empty Venice key → error spoken, clean exit, no crash
• validate_ability.py ✅ flake8 ✅

[github-actions]

🔀 Branch Merge Check

PR direction: feature/voice-persona-chat → dev

✅ Passed — feature/voice-persona-chat → dev is a valid merge direction

[github-actions]

✅ Community PR Path Check — Passed

All changed files are inside the community/ folder. Looks good!

[github-actions]

🔍 Lint Results

✅ __init__.py — Empty as expected

Files linted: community/voice-persona-chat/main.py

✅ Flake8 — Passed

✅ All checks passed!

[github-actions]

✅ Ability Validation Passed

📋 Validating: community/voice-persona-chat
  ✅ All checks passed!