I shipped another one: Personal Replicant, an open-source GUI for training a language model of you. It is live, it is public, and you can poke at it right now: replicant-rust.vercel.app.
Here is the itch I was scratching. Chatbots forget you the moment the conversation ends. A Personal Replicant is the opposite: a private, persistent model of one specific person. You feed it the raw material of a life, and it builds a searchable memory bank, then learns to think and speak in that voice.
The pipeline is four clean stages:
- Capture - write a memory in plain language: a fact, a story, a belief.
- Classify - AI files each memory under one of six identity pillars, extracts the people involved, and stores it as a vector embedding.
- Train - turn that memory bank into a fine-tuned LoRA adapter, a lightweight personality layer on top of a base model.
- Converse - chat with the result, and dial the persona intensity up or down to blend raw intelligence with your own grounded identity.
It runs local-first: a Python and ChromaDB backend with GPU-accelerated fine-tuning, your memories on your hardware. The hosted demo swaps the backend for a fully populated sample persona so the entire eight-tab workbench stays interactive with zero setup.
Why open-source, why public? Because I do not believe the future should run on one giant brain that decides what is true for everyone. I believe in millions of personal agents, each trained by a real person on their own values and memories. Your character is worth defending. Clone it, train your own.
Step inside → Launch the live demo