Exxpert/Lovely-2.0

Lovely 2.0

Real, modular personal AI companion for Bhanu (Expert Gamer) — Windows PC, Telegram, Android/mobile, voice, vision, memory, automation. No fakes, no mock success: every status in the registry is earned by a passing test or stated as blocked.

Current milestone: M1 (core contracts) — UNTESTED

Built: kill switch, event bus, tiered permissions + dangerous-command PIN, SQLite memory with decay/pinning, personality engine (honest failure reporting, no-dead-end, absurd-spec roast detection), OpenAI-compatible model router with fallback chain, CLI with doctor. Code is written and reviewed but has not been executed yet — see TESTING.md for how to verify.

Quick start (owner's Windows PC)

git clone https://huggingface.co/Exxpert/Lovely-2.0
cd Lovely-2.0
python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt
python main.py doctor        # first thing, always
python main.py pin           # set dangerous-command PIN (stored hashed)
# set one provider key for chat, e.g.:
setx GROQ_API_KEY "gsk_..."
python main.py chat
python -m pytest tests/ -v   # core unit tests

Emergency stop from ANY surface: create kill.flag in the data dir, or python main.py kill on. Nothing in the system may bypass it.

Layout

• lovely/core/ — config + resource laws, event bus, kill switch, permissions, memory, personality
• lovely/router.py — the single LLM wrapper (spec §12 #1)
• lovely/agents/ — M3 agent engine (planned)
• registry/features.json — the 188-feature registry (never lose this)
• docs/ — architecture, testing, security, limitations
• main.py — CLI: status, doctor, kill, pin, chat, memory, prefs

Status reporting

Every work cycle appends to CHANGELOG.md with COMPLETED / VERIFIED / FAILED / BLOCKED / CHANGED / RESOURCE IMPACT / NEXT. Failures are never hidden.