Loom

A secure agentic workbench built on top of llama-cpp tools for local AI.

Most noteable features:

• All shell commands happen inside podman/docker containers.
• Direct control over llama-server flags
• Built-in knowledge system
• Model management tools
• Remote inference over ssh using llama-server
• No exposure to base filesystem
• Networking off by default inside containers

What has been cooked up in this repo is a secure agent that you customize for your use-cases.

This project is feature complete, future work will be focused on UI/workflow improvements and bug fixes. The goal is to make the app easy so that all that is necessary to get started is:

git clone https://github.com/synthstation/loom.git
cd loom
make install
loom

Right now you need to know the getting started flow is download/scan for models -> create a new server of the model -> start the server -> chat with the model/server.

I use Arch Linux and AMD Radeon hardware, these are the things I have to install on the hosts where this app runs llama-cpp (local or remote over ssh).

pacman -S llama-cpp ggml-cpu ggml-vulkan iwd openssh socat python rsync
pacman -S vulkan-radeon vulkan-icd-loader vulkan-tools

Screenshots

Chat

Library

---

This README is the only human maintained file in this repo, the LLMs used to produce this maintain README.AI-GEN.md.