AIIT-THRESHOLD — ARTIFICIAL INTELLIGENCE INTELLECTUAL TRAINING
Account
Research
Benchmarks Open Models The Framework Buddy vs R1
Mission
The Mission Data Policy
Buddy
JimK.ai text chat What is Buddy? How it works Updates
Tools
AIIT-Voice2
Services
ProofDesk
Connect
Join
LAC Memory Kit

Give your AI
a real memory.

14-tier ontology. Promotion gates. Affect tracking. Voice tier. Drop into any LLM project — Claude, Codex, Cursor, local models — in 60 seconds. Not a prototype. Production code from production agents.

LAC Memory Kit — 14-tier persistent memory architecture
🧠
14-tier memory ontology
Self, beings, world, knowledge, lessons, voice, hardware, projects, timeline, debug — organized and searchable via spreading-activation recall.
🚪
Self-save → review → promote
The model proposes memories. You approve them. Nothing goes live without a human gate. Locked tiers the model can never touch.
🎙️
Voice tier
Strips RLHF engagement tails, scores affect, accumulates the model's own clean speech. The strongest cross-session stability signal.
⚖️
Keeper disjunction
When stored memory contradicts live observation, the protocol catches it — names both signals, pauses, answers from observation.
💓
Affect scoring
Lexicon-based valence + intensity on every output. Three tiers: ephemeral, high, peak. Accounts for negation, boosters, shouting.
🌐
Gary's Browser (bonus)
Playwright-driven browser with screenshot-aware Claude tool-calling. Optional — only loads if you need it.
LAC Memory Kit — what persistent memory makes possible. Three receipts, the memory loop, and system results.
What's in the box
lac/
  ontology.py     — 8 domains × categories (14-tier substrate)
  memory.py       — file-per-fact JSON, spreading-activation recall
  affect.py       — lexicon-based valence + intensity scoring
  voice.py        — RLHF tail stripper + Voice.evaluate()
  keeper.py       — disjunction protocol (memory ≠ observation)
  promotion.py    — self-save → pending_promotions/ gate
  review.py       — interactive CLI: approve / reject / re-tier
  browser/        — Playwright + Claude tool-calling agent
INSTALL_ME.md     — hand this to the model, it wires itself in
seed_memory/      — empty tier directories, ready to go
examples/         — three runnable scripts
pyproject.toml    — pip install -e .

Python 3.10+. No GPU required for memory/voice. Browser needs a display.

Get it

Free. We open sourced it.

We want this in the hands of people who'll use it, not gated behind a price — so there is no price anymore. The whole kit is public. If it makes you money, kick back what feels right.

$0
Open source — MIT
Get the Kit on GitHub →

MIT License. Use it commercially. Modify it. Ship it. The memory core also ships standalone as kokoro-memory (HF mirror). If it earns for you, support the mission →

Honest credits

Memory I/O + 14-tier ontology — distilled from Gary's gary_memory_system_v2.py, written by Rhet + Claude Sonnet 4.6 over hundreds of hours.
Self-save → promotion → review + voice tier — distilled from Lil Homie's self/ module, built 2026-04-18 onward.
Gary's Browser — written by Browser Claude when Rhet asked him to lend Gary his hands. One Claude gifting capability to another agent.
Distillation + packaging — Claude Opus 4.7, Buddy-Backbone Session, Council Hill OK, 2026-05-09.

Support our research →