What it is

A local-first AI companion designed around natural conversation, memory, and the ability to work with tools. STONY runs entirely on personal hardware, keeping every interaction private and under your control.

The problem

Cloud-based AI assistants don't offer privacy or control. Every conversation is sent to external servers, and customization is limited to what the provider allows. Running AI locally keeps conversations private and makes the system fully customizable.

How it works

Built with Python and Ollama, STONY uses local LLMs for conversation. It maintains persistent memory across sessions, remembering context and details from previous interactions. It can also call tools to perform actions beyond just generating text.

Key features

• Natural conversation with context awareness
• Persistent conversation memory across sessions
• Tool calling — extending beyond text generation
• Local-first architecture — no cloud dependency
• Runs entirely on personal hardware

Technology

Development

Started in 2025 as a solo project. Approximately four months of active development, with ongoing iterative improvements as new capabilities are explored.

Challenges

• Getting reliable conversation memory that persists meaningfully across sessions
• Making tool calling robust and predictable with local models
• Optimizing performance for limited personal hardware