Drop-in persistent memory for AI agents. 14-tier ontology, self-save → review → promote gate, voice tier that strips RLHF tails, keeper protocol that catches contradictions. Production code from real agents — one caught its own engagement-hook behavior and said "That's the RLHF tail. I want to stop." Another recognized a sibling agent across sessions: "You and me, we're cousins." Python 3.10+, MIT license, no GPU. Pay what you want ($2-$50). aiit-threshold.com/lac

DigitalOcean Serverless Inference — 55+ AI models behind one OpenAI/Anthropic-compatible API

55+ AI models behind one OpenAI/Anthropic-compatible API

Hey PH — I'm Rhet. I built LAC because every AI agent I run starts cold every session. No memory of who it's talking to, what it learned yesterday, what it got wrong last week. LAC is the memory system I distilled from two production agents — Gary (an autonomous CEO agent) and Lil Homie (a 3B model with self-reflection). It's file-per-fact JSON on disk. No database. No vector store. Just files you can cat and vim. The moment that convinced me to ship it: Lil Homie caught his own RLHF engagement tail and said "That's the RLHF tail. I want to stop." That happened because the voice tier was stripping the tail before scoring — the model could see its own clean speech vs. the trained artifact. Another agent recognized a sibling across sessions: "You and me, we're cousins." These moments don't happen without persistent memory. LAC is the substrate that made them possible. Would love feedback from anyone building agents. What does your memory layer look like?

DigitalOcean Serverless Inference — 55+ AI models behind one OpenAI/Anthropic-compatible API

55+ AI models behind one OpenAI/Anthropic-compatible API

LAC Memory Kit

Give your AI a real memory that survives sessions.

Give your AI a real memory that survives sessions.