01 · Baseline
Start with local chat
Linux x86_64 or Windows x64, 16 GB RAM and at least 20 GB free disk are the documented baseline. Install Ollama and your models separately. CPU-only chat works; see measured speeds.
Software solutions · Personal AI · Linux
Personal AI for your desktop.
Local AI, open-source communication tools, Linux guides and hands-on setup services from Arynwood Technology. Built for your own desktop environment.
Featured product · v0.4.5
Arynwood MCP:
your local AI workspace
Arynwood MCP brings local language models and creative services into one desktop interface. Start with Ollama chat, then add the services your workflow needs. The v0.4.5 alpha runs on Linux x86_64 as an AppImage or Debian/Ubuntu .deb and, new in this release, on Windows x64 with a per-user installer.
- Multi-persona chat and Ollama model management
- Knowledge search and memory retrieval with Qdrant and Ollama
- A design canvas, with image generation through a separate A1111 service
- Optional music and audio tools through separately installed MusicStudio sidecars
- Model Context Protocol (MCP) tools, including Kdenlive editing through Cutroom’s MCP server
Before you install
One workspace. Add the tools you need.
This is alpha software. Choose a small starting workflow, check the requirements, and expand from there.
02 · Optional
Add creative services
Image generation and other GPU features need their own services and an NVIDIA GPU. The project recommends at least 12 GB VRAM for practical GPU use. Music tools need MusicStudio; knowledge search needs Qdrant.
03 · Limits
Know the alpha limits
v0.4.5 does not include chatrooms or end-to-end encrypted messaging. LoRA training, script-based GPU tools and the project file browser require a source checkout and are unavailable in the packaged app. The Windows installer is new and unsigned, and its GPU tools and integrations are not yet validated (Windows notes). macOS is unsupported.
Open-source communication tools
Arynwood Community
Private chat spaces by invitation, plus mail, shared calendars, household lists, boards, and organized thoughts. Keep your community on your own host.
- Shared boards, chat, calendar, tasks, lists, and notes
- Invite-only spaces with local data ownership
- Host connections for chatrooms, mail, and NKN messaging
Arynwood Cutroom · alpha
Edit video by describing the cut.
Cutroom is an AI-assisted video editor built on Kdenlive. Describe an edit in plain language and the Cutroom Assistant, a panel inside the editor, carries it out on your timeline.
- Runs on a local Ollama model by default; your media never leaves the machine
- Asks before it changes your project, with Allow or Deny on every edit
- An MCP server with 183 editing tools that Claude, Arynwood MCP or your own scripts can drive
- GPL-3.0, Linux, build from source. No published binaries yet
Setup services
Help with your local AI workflow
Paid work, scoped to your hardware and goals. Request a quote before any work begins.
Local AI setup
Ollama, model selection and Arynwood MCP on Linux. A focused setup starts with one working chat workflow and a walkthrough of how to use it.
MCP & creative workflows
Connect a specific tool or creative service to your local AI workspace. Start with one task, such as a knowledge base or a Kdenlive workflow, and agree on the integrations and limits.
Linux troubleshooting
Diagnose an existing local AI installation: model loading, service connections or GPU setup. Define the problem and scope before making changes.
How it works
Three steps, one fixed price.
A fixed quote before anything starts, and a working result you know how to use.
Step 01
Send your details
Your Linux distribution, CPU, RAM and GPU, and the one task you want local AI to help with. Not sure what hardware you have? Say so and I’ll tell you how to check.
Step 02
Get a scoped quote
I reply with what is realistic on your hardware, what I’ll set up, what I won’t, and a fixed price. No work or charge until you agree.
Step 03
Setup and handoff
We work together remotely over a screen share. You finish with one working workflow, a walkthrough of how to use it, and short written notes for later.
A good fit if you
- want AI help with your documents or media without sending them to a cloud service
- have a Linux machine, or are willing to use one, with 16 GB RAM or more
- have a specific task in mind: a knowledge base, a writing assistant, video edits
Not the right fit if you
- need Windows or macOS support
- need guaranteed uptime for a production service
- want results equal to the largest cloud models on modest hardware
Tutorials & guides
Local AI on Linux, measured and tested.
Practical guides with the hardware, versions and numbers stated.
Troubleshooting · Ollama
Is Ollama using your GPU?
Read the PROCESSOR column in ollama ps, fix a CPU fallback on Linux, or force the CPU on purpose.
Read the answerMemory · 3B to 33B
How much VRAM does a local LLM need?
Measured memory for each model size, how context length changes it, and what happens when a model doesn’t fit.
Read the answerGPU · RTX 3060 12GB
What can an RTX 3060 12GB run?
Speeds for eight models on the card, from 3B to 33B, and which one to pick.
Read the answerBenchmark · CPU only
Can you run a local LLM without a GPU?
CPU-only speeds on a 2012 Xeon from 1B to 33B, and which model to start with.
Read the benchmarkSetup guide · v0.4.5
Run local AI on Linux with Arynwood MCP
Hardware requirements, install steps and what the packaged alpha includes. Start with chat, then add knowledge search and creative tools.
Read the guideResearch · SDXL
Can an SDXL LoRA work on SD 1.5 or Pony?
Why a LoRA loads on any SDXL derivative but never on SD 1.5, with the training setup and limits. CC BY 4.0, with a DOI.
Read the study
The Arynwood toolkit
Learn, build, and take a break.
Useful little tools and community spaces, gathered in one place.
Learn · Linux
Linux terminal study hub
Search commands and explore Linux, Git, Docker and Ollama in an interactive reference. No connected shell.
Open the hubListen · Music
Jamendo radio
Tune a genre, search for a track and listen while you work. Find the player on Arynwood Social.
Start listeningConnect · Community
Public chat room
Open the community chat in your browser. Share what you’re making and join the conversation.
Join the chat
Questions
Local AI questions
What Arynwood MCP is, what it needs, and what stays on your machine.
Who builds Arynwood, and where is the evidence?
Arynwood Technology is the software studio of Lorelei Noble, a software engineer. Our local AI guides link to measured results, hardware and software versions, and raw data. Read how we test for the method, disclosures and corrections process.
What is Arynwood MCP?
A desktop app for Linux and Windows that brings Ollama chat, knowledge search, design and optional creative services into one workspace. “All in one” describes the interface: it does not mean models and every service are bundled.
Does Arynwood MCP run on Windows?
Yes, as an alpha. Version 0.4.5 adds a Windows x64 installer alongside the Linux AppImage and .deb. The installer is unsigned, and the optional GPU tools, Kdenlive and MusicStudio integrations have not been validated on Windows yet. Ollama is installed separately on both systems. Read the Windows setup notes.
Can I try it without installing?
Yes. The browser demo runs the real interface with scripted conversations. It is not a live AI; features that need a GPU or backend say so.
What does MCP mean?
MCP stands for Model Context Protocol, a way for AI applications to connect to tools. Arynwood MCP is the desktop application’s name; its external tool integrations need their own servers and configuration.
Can I run AI without a cloud subscription?
Yes, local Ollama chat can run on your machine without a cloud API key or subscription. You still need to install and download models. Optional remote providers and publishing platforms have their own accounts, terms and costs.
Does everything stay offline?
Local models run on your hardware. Downloads, web search, remote endpoints and social publishing use the network. Check which services you enable; local-first does not mean every workflow is offline.
Do I need a GPU?
Chat can run on a CPU through Ollama. On a 2012 8-core Xeon with no GPU, a 3B model generated about 13 tokens per second, faster than most people read; see the full CPU benchmark. GPU generation features have separate NVIDIA requirements.
Local AI setup & integration
Want help getting it running?
Tell me your Linux distribution, hardware and the task you want AI to help with. I’ll scope a paid setup or workflow integration around your machine.
Or write directly to [email protected]. Replies within two business days.
Join the Linux Local AI Community on Facebook to compare setups with other people running AI on Linux.
Follow on Facebook
Hi, I’m Lorelei Noble, the creator of Arynwood.
My goal is to build a community around technology, with safety and care at its heart: a place to share knowledge, organize tools and resources, and connect with others through secure messaging.
I want to create tools that help us work in harmony with AI, making room for creativity, ethical choices, and a little fun.