LLMChat

LLMChat is a chat interface between a local LLM (running under Ollama) and the cloud-based OwnSona MCP server. It is built on the Kiss web development framework.

You talk to a model running entirely on your own machine, but its answers are grounded in your personal memory held by OwnSona — and anything the model learns about you during the conversation is written back to OwnSona.

What it does

On each message, LLMChat:

1. Recalls relevant facts about you from the OwnSona MCP server.
2. Builds a prompt from those facts, the conversation so far, and your message.
3. Generates an answer with your local Ollama model.
4. Displays the answer.
5. Learns — durable new facts from the exchange are written back to OwnSona.

The local model never leaves your machine, and OwnSona holds the memory.

Your local LLM can learn — and what it learns is shared everywhere

OwnSona is your own personal, user-controlled memory server. Because LLMChat writes everything the local model learns back into OwnSona, and because OwnSona is also accessible to the major cloud assistants — ChatGPT, Claude, Grok, and Gemini — anything your local LLM learns about you through LLMChat becomes available to all of them too.

In other words, a fact you teach your local model here is remembered once and shared everywhere: your local LLM and every connected cloud LLM draw on the same evolving personal memory.

Two interfaces

LLMChat offers both:

• Command line — a terminal REPL (in the spirit of ollama run), with line editing, history, model selection, and streamed (token-by-token) replies.
• Web — a browser chat UI served by the embedded Kiss/Tomcat server.

Both interfaces share the same underlying engine, so they behave identically.

Related projects

• OwnSona (the cloud memory MCP server): https://github.com/blakemcbride/Ownsona
• Kiss (the web development framework this is built on): https://github.com/blakemcbride/Kiss · kissweb.org

Requirements

• Java JDK 17+ (tested with 17/21/25)
• Ollama running locally with at least one model pulled (e.g. ollama pull llama3.2)
• Access to a running OwnSona MCP server (see the OwnSona project)
• An internet connection for the first build (to download dependencies) and to reach OwnSona

Build

./bld build        # Linux, macOS, BSD   (use  bld build  on Windows)

This downloads dependencies and compiles the application.

Configure

Settings live in src/main/backend/application.ini. The key values:

OllamaUrl     = http://localhost:11434      ; local Ollama (no trailing /api/)
OllamaModel   = llama3.2:latest             ; default model
OwnSonaMcpUrl = https://ownsona.com/mcp     ; your OwnSona MCP endpoint

[OAuthClient ownsona]
Url = https://ownsona.com                   ; OwnSona origin (OAuth)

The first time you connect, LLMChat opens your browser once to log in to OwnSona; the token is then cached and refreshed automatically.

Run

Command line (default)

./llmchat                 # start the chat REPL
./llmchat -m gemma3        # choose a model for this session
./llmchat --check          # print Ollama/OwnSona status and exit
./llmchat --help           # full usage

In the REPL, type a message to chat. Useful commands: /model <name>, /models, /memories (show what was recalled), /learn (toggle auto-learning), /agent (toggle agentic tool use), /stream (toggle streamed output), /new, /quit. Say "remember that …" to store a fact explicitly.

Web

./llmchat --web

Then open http://localhost:8000 in your browser.

Documentation

Design and implementation notes are in LLMChat-Plan.md.

Credits

LLMChat was written by Blake McBride. It is built on the Kiss framework and connects to the OwnSona MCP server, both also by Blake McBride.