baabaa

Your own assistant and coding agent, on your own GPU

baabaa gives the models you run with Ollama the things a model alone lacks: conversations that are kept, projects, memory, artifacts, and an agent that reads, edits and runs code in your folders. It runs on your computer, for you or for everyone on your network, and nothing leaves it.

While baabaa works, a strand of wool curls as the model's words arrive, so you can see that it is alive. The preview is the real interface with recorded replies; it runs no model.

Install

curl -fsSL https://github.com/kristiandroste/baabaa/releases/latest/download/install.sh | sh

No sudo and no pip. The installer checks what you have, downloads the release and verifies it, and asks whether to start baabaa. The tutorial walks through every step, from an empty machine to the first reply.

What it does

Chat and code in one place

A conversation is a chat. Give it a working folder and it reads, edits and runs things there. The browser and the terminal show the same history.

You decide what runs

Four modes, from asking before every edit and command to Auto, where safe actions run and anything risky waits for you.

Tools in a sandbox

Every command the model runs can change only the working folder and has no network unless you allow it.

Models that fit your GPU

A model is offered only after a test shows it loads entirely into GPU memory. Nothing falls back to the CPU.

What a model alone forgets

Projects with their own instructions and documents, memory across conversations, search of earlier chats, artifacts beside the chat.

Yours

Accounts for the people in your home or office, each with their own conversations. No cloud, no telemetry: usage statistics stay on your machine.

Read on