Home / Projects

๐Ÿง LLM

A local AI server and chat web app that runs on my own computer

MODELSGemma 4 E2B ยท EXAONE 3.0 7.8B ยท Qwen 2.5 7B
TECHOllama ยท Python API server ยท Cloudflare Tunnel
ACCESSAPI key holders only

Overview

If you would rather not send certain writing, work documents or personal conversations to an outside AI service, a self-hosted AI is the answer. This project runs several LLMs on a single Mac and exposes them through a simple API and a chat web page you can use from anywhere.

Key features

๐Ÿ’ฌChat and writing

Questions, code and writing are split into areas, each with its own saved conversation.

๐Ÿ”Web search answers

Reads search results and answers with source links.

โœ‰๏ธMessage drafting

Pick the type, tone, length and recipient to get a message split into greeting, body and closing.

๐Ÿ—ฃ๏ธLearns my writing style

Filters only my messages out of chat exports inside the browser to learn my style. Personal data never leaves the device.

๐Ÿ“ŽFile attachments

Converts PDF, Word, Excel, PowerPoint, Hangul (hwpx) and photos (OCR) to text so the AI can read them.

๐Ÿ”ŒSimple API

/ask, /chat, /message and /search-ask plug straight into your own service, with key auth and rate limits.

What I can build for you

Work I can take on, based on what I learned building this project.

  • Model selection and benchmarking, tuned to run on low-RAM machines
  • Local AI servers for personal or team use, with auth and secure external access
  • Connecting document conversion, search and writing tools to real workflows

Each model has its own license (for example, EXAONE is non-commercial). For commercial use I help you choose the right model.