Open-source tools for local AI

Run AI on your own machine

The same local-AI toolchain that ships in Shadowfetch Linux installs on any Debian Testing box. Agents, local LLMs and a private chatbot, all running on your own hardware -- no cloud, no API keys, and no need to switch your OS.

One command, any Debian

The toolchain lives in shadowfetch-defaults. Add the signed repository and install it, and the assistant, model and GPU helpers are set up for you -- without reinstalling your system.

sudo apt install shadowfetch-defaults

Add the APT repository →

The toolchain

Each is one command. Start with shadowfetch-agents, or reach for whichever tool fits the job.

  • shadowfetch-agents

    The one command that shows your choices, confirms prerequisites, and launches whichever agent runtime you pick. Start here.

  • shadowfetch-ai

    A private local chatbot, no agent framework -- Ollama with an Open-WebUI front end, running entirely on your hardware.

  • shadowfetch-llm

    Pick and compare local models matched to your GPU and memory, so you run the largest model your machine can actually hold.

  • shadowfetch-hermes

    The Nous Research self-improving agent. `hermes model` picks a provider or your local Ollama, so it can run fully on-device.

  • shadowfetch-openclaw

    A friendly personal agent installed through npm -- a quick start that drives your local Ollama models out of the box.

  • shadowfetch-agent-workspace

    Agent Studio: create a separate, sandboxed workspace per job and enable only the browser, terminal and tool capabilities that job needs.

  • shadowfetch-agent-tools

    Browser automation, tools and isolation for agents. Run shadowfetch-agent-doctor any time to check versions and health.

What runs on your hardware

We benchmark local models on our own card — a single 16 GB GPU — and publish the numbers: tokens per second, memory used, and whether a model fits entirely on the GPU. See what your hardware will actually do before you install anything.

The 16 GB Local AI Playbook →

It stays on your machine

Both runtimes can run entirely on your hardware via Ollama. Ollama listens only on localhost and the firewall blocks port 11434 -- your models never leave the machine. There is no account, no telemetry, and nothing to sign in to.

Go deeper