The focus for the next two years
Learn, plan and bring people together.
The direction is a connected set of education and learning tools working alongside Grove. Open Knowledge is in development; Grove and the browser toolkit are available to explore today. The full project collection continues below.
Education & research · In development
Arynwood Open Knowledge
An open-source project for educational and special education research: finding studies, tracing sources and keeping a local knowledge bank with citation export. Part of my two-year learning software roadmap, intended to work in unison with Grove. In development, not yet released.
Explore Open KnowledgeLearning tools · Available in your browser
Education & learning toolkit
Explore the Linux terminal study hub, tutorials, HTML editor and flowchart designer. Practical tools for building skills, trying ideas and organizing what you learn.
Browse the toolkitCommunity & events · Self-hosted software
Arynwood Grove
Bring teacher planning groups, households, clubs and event teams together. Coordinate calendars, tasks, notes and discussion in shared spaces by invitation, on a server you control.
Explore Grove & installationData storage · Included in Grove
A shared home for your files
Keep group documents and uploads in Grove alongside your plans. Folders, file versions, exports and backups help you organize and move your data on your own server or computer.
See how Grove stores your data
The Arynwood project collection
Developed projects & creative work
Alongside Grove and the learning toolkit above, explore my desktop software, creative tools and research. Each project links to its demo, download, source or published work, with alpha status shown where applicable.
v0.5.4 · Linux and Windows desktop alpha
Arynwood MCP,
local AI workspace
Ollama chat, knowledge search, a design canvas and optional creative services in one desktop workspace for Linux and Windows. Models and supporting services are installed separately.
- Multi-model LLM chat with streaming via Ollama
- Knowledge search and memory retrieval with Qdrant and Ollama
- Design canvas and optional A1111 image generation
- Optional MCP tool servers and MusicStudio sidecars
Community needs a Grove on 0.7.1 or later, installed separately. LoRA training, script-based GPU tools and the project file browser require a source checkout. The Windows installer is unsigned, and its GPU tools and integrations are not yet validated. Read the requirements and package limits or the Windows notes.
Video · Kdenlive
Cutroom
AI-assisted video editor built on Kdenlive. Describe an edit in plain language and the assistant carries it out on your timeline with a local model, asking before each change. Alpha, build from source.
View repositoryResearch · SDXL
LoRA training study
An SDXL cross-checkpoint portability study, with the training source and documentation.
View repositoryWriting · Benchmark
Local AI without a GPU
Ollama models benchmarked CPU-only and on an RTX 3060, on a 2012 Xeon workstation. Which size is usable, and when a GPU matters.
Read the benchmarkArt · Textiles
Art & textile design
Digital art, templates and textile designs, from fabric prints to journals and apparel.
View the gallery
About
Hi, I’m Lorelei Noble.
I’m the software engineer behind Arynwood Technology. My focus for the next two years is education and learning software designed to work in unison with Grove. My developed projects, including community software, local AI, browser tools and creative work, are gathered here as that work grows.
I also make digital art and textile designs. The source code for Arynwood projects is on GitHub.
- Learning and research: Open Knowledge, education resources and practical browser tools
- Community: Grove for shared planning, events, communication and files
- Local AI and creative tools: Arynwood MCP, Cutroom and measured Linux guides
Writing
Tested answers and research
Local AI questions answered with measurements from my own workstation, plus a DOI-registered study.
Troubleshooting · Ollama
Is Ollama using your GPU?
How to read ollama ps and nvidia-smi, fix a CPU fallback on Linux, and force CPU or GPU.
Read itHardware · Memory
How much VRAM does a local LLM need?
Measured memory for 3B to 33B models, and what happens when a model doesn’t fit.
Read itHardware · GPU
What can an RTX 3060 12GB run?
Measured Ollama speeds for eight models, from 3B to 33B.
Read itBenchmark · CPU
Can you run a local LLM without a GPU?
CPU-only speeds on a 2012 Xeon, from 1B to 33B.
Read itResearch · SDXL
Cross-checkpoint portability of SDXL LoRA fine-tunes
Why a LoRA loads on any SDXL derivative but never on SD 1.5. CC BY 4.0, DOI 10.5281/zenodo.21282393.
Read the studyData · CC BY 4.0
Local AI benchmarks
The raw measurements and the script behind them, open for anyone to reproduce and extend.
View the data
Contact
Get in touch
Tell me what you’re teaching, planning or building. Get in touch about learning tools, Open Knowledge, community events or organizing your group’s data, and I’ll reply within two business days.
Follow on FacebookWho builds Arynwood?
Lorelei Noble is the software engineer behind Arynwood Technology, including Open Knowledge, Grove, Arynwood MCP and Cutroom. Explore the products above for educational research, learning tools, community planning and creative work.
Where can you verify the benchmark results?
Read how Arynwood tests local AI for the hardware, methods, disclosures and corrections process. The benchmark repository contains the raw measurements and testing script.
How do you contact Lorelei?
Email lorelei@arynwood.com for questions, corrections or project inquiries.
