Asvaettir Labs

10/11/2026

Hail everyone tuning in,

First time on camera, we need improvement lol

Just explaining where we were at last night got few hours in and called it.

Be your VRAM is lacking we rely on CPU and hardware that can take much longer than we are used to with cloud, but accuracy far beyond what we anticipated.

Stay tuned for part 2.

10/11/2026

First hand exposure of the Labs

10/10/2026

10/10/2026

Parental Alienation in Moncton, NB, CA ,nb,camoment. I wouldn't say I have a " favorite". But I'd have to say if looking at life for what it is, we all experienced the same joke, here's to another era of children being harmed everyone knowing about it and not a single s*** was done about it.

Stay tuned for 2027's f******ery

10/04/2026

Ásvættir Labs – Workstation Update

Quick progress update on the Ásvættir Labs workstation build and our offline AI workflow.

The dual‑CPU Linux Mint setup (64GB ECC RAM, dual 500GB SSDs, Bionic runtime through LM Studio) has been performing extremely well. Even under heavy inference loads, both processors are only scratching around 20% utilization, which gives us a lot of headroom for scaling up.

Right now, the following model classes are running smoothly:

- 3‑bit
- 4‑bit
- 7‑bit
- 14‑bit
- 20B+ models in Q3/Q4 quantization

With the RAM upgrade and the stability we’re seeing, I’ll be pushing into the 30B, 70B, and possibly 128B ranges next. The goal is to find the exact cap of this workstation under Bionic’s optimized flow and our fine‑tuned skills.

The current workflow is clean, uses reduced context in chat, and automates well through the Bionic UI and our custom skill chain. More updates coming soon — including my first video walkthrough explaining how everything ties together.

Feel free to drop a comment or critique. Always open to feedback while I keep refining the setup.

10/04/2026

As you can see in our $Bionic we have been experimenting progressively into ensuring the model and newly added skills work with our system, testing full builds and handoff strategy, while this last run I have been hitting successful runs, only time will tell.

Developing on systems like this with low VRAM offered a slight adjustment and $extra hands on work but ultimately the costs are monetarily low comparison to $systems $operating with 32B - 70B models.

this is a huge disadvantage but for a small pop-up $Dev-studio, I am almost $satisfied. Currently the exchange with VRAM && CPU are almost dialed in 100%.

$Please share you're personal experiences working with LLM's on devices that otherwise are rated " under-rated "

09/28/2026

Ásvaettir Labs Update — Local AI, Infrastructure Shift & What’s Coming Next

Hail everyone — quick update.
We’re gearing up for software releases, but there’s no hard production date yet. I’ve been rebuilding our entire environment around local AI models using Bionic, and migrating everything off cloud dependencies for privacy and autonomy.

============================
Local AI Workstation (Current Build)
============================

Dell Precision Tower 7810

2× Intel Xeon E5‑2680 v3 (24 cores / 48 threads)

64 GB RAM

2× GTX 1060 (6 GB VRAM each — NOT shared)

1 TB storage

Ubuntu 24.04.4 LTS (GNOME 46, X11)

Kernel 7.0.0‑34

LM Studio — Bionic Runtime

======================
Models I’m Actively Running
======================

Qwen3-VL 4B — PROJECT_COORDINATOR
Qwen 14B — ARCHITECT
Qwen2.5-Coder 7B Instruct — SYSTEMS_BUILDER
DeepSeek-Coder-V2-Lite-Instruct — VISUAL_BUILDER
Qwen2.5-VL 3B Instruct — VISION_QA, with Qwen3-VL 4B as the fallback Vision QA model

============
The Challenge
============

The dual GTX 1060 setup is powerful but tricky:
Each card has its own 6 GB VRAM pool, so larger models need:

CPU/RAM offload

Careful layer distribution

Quantization tuning

Runtime adjustments
=======================
Why Progress Has Been Slow
=======================

Between work, legal schedules, and trying to maintain some normalcy, development has been slower -
but nothing has been abandoned.

I’m currently:

Re‑engineering routers

Rebuilding local model pipelines

Migrating Ásvættir Labs projects to a dedicated offline server

Finalizing cross‑platform builds (Forseti stays Android‑native; other apps are evolving)

This shift is intentional — privacy first, autonomy always.

================
Community Question
================

If you’re running local AI on similar specs:
Have you hit the same VRAM split issues?
What models or runtimes have worked best for you?

Drop your setup, your struggles, or your wins — I’m genuinely curious what others are doing with mid‑range dual‑GPU rigs.

09/24/2026

In Norse paganism, we don’t pretend the storm isn’t there. We face it, we brace ourselves, and we move with it. The old ways teach that every season has its weight, and every weight has its purpose.

Some days feel like walking through winter with no fire in sight — but even then, the ancestors remind us: “Keep moving. Even a single step is defiance.”

Life has been heavy lately, and I know I’m not the only one carrying more than I speak about. If you’re in your own storm, your own long winter, your own test — you’re not alone. Not in this hall. Not in this community.

We push forward.
We endure.
We rebuild.

And like the sagas say: “What is broken can be reforged.”

If you’re struggling, speak your truth below. The hall is open, and you’re welcome here.

The Daily Pagan Yes, Norse Mythology is my Religion.

09/21/2026

Forseti allows you to upload and quick view your States, rules, regulations, and laws and provides an automatic contents. If.gov cites do not provide.

Head on over to Google Play , Forseti check it out for free 3 days, no hidden subscriptions don't like it don't buy it simple made simple.