Skip to content
@sybil-solutions

Sybil Solutions

A place for all things AI
README.md
Sybil Solutions

Sybil Solutions

A place for all things AI.

Website X Email Repos

Sybil Solutions builds local-first tooling for running, steering, and using self-hosted LLM backends. Everything here assumes you own the box it runs on.
Run a model on your machine omarchy-local-ai: one button, the model tested on your GPU, your coding agent on it
Find the model for your card local.sybilsolutions.ai, built from local-ai-registry
Add a recipe, earn $50 Open bounties, run with the recipe kit · how bounties work
See who uses it usage.sybilsolutions.ai

New here? Read CONTRIBUTING.

Applications and plugins

Repository Language Stars Description
local-studio TypeScript stars Control panel for vLLM, SGLang, llama.cpp, ExLlamaV3 — localstudio.ai; website folded into site/ on the integration branch
omarchy-local-ai Shell stars Omarchy plugin: one bar button serves the model validated for your GPU, opens a coding agent on it, shares it on your tailnet; install counter folded into usage/
codex-shim Python stars Local Responses-API shim exposing Factory BYOK models (and optional ChatGPT GPT-5.5 passthrough) to Codex Desktop
ai-data-extraction Python stars Extract all your personal data history from Cursor, Codex, Claude Code, Windsurf, and Trae

Local AI program

Repository Language Stars Description
local-ai-registry HTML stars The model to run on your GPU and exactly how to run it — one validated recipe per machine, with the evidence attached (local.sybilsolutions.ai)
local-ai-recipe-kit Python stars Run a big model on your own GPU, measure it with one protocol, submit the result — intake for registry recipes
local-ai-images Python stars Attested container images pinned by the local-ai registry

Serving engines and kernels

Repository Language Stars Description
trellis-serve Python stars EXL3 trellis quantization in stock SGLang and vLLM on RTX 3090 and Intel Arc B70, bit-exact with ExLlamaV3
moetier Python stars Data-first standard for MoE expert placement and scheduling across VRAM, RAM and NVMe (records + core + simulator)

Hardware recipes

Repository Language Stars Description
glm-flash-lite Python stars GLM-5.3-Flash EXL3 on one 24 GB RTX 3090 + DDR4: elastic GPU expert cache, zero-copy experts, AVX2 CPU tier
dsv41-flash-offload Python stars DeepSeek-V4.1-Flash EXL3 on one 24 GB RTX 3090 + DDR4 + NVMe: staged-DMA prefill, AVX2 CPU expert tier, Engram on disk
deepseek-v4.1-flash-4x-rtx-pro-6000 Python stars DeepSeek V4.1 Flash on 4× RTX PRO 6000 Blackwell: native weights, DSpark, vision, NVMe or locked-RAM Engram offload
qwen-serve Python stars Renamed from qwen38-3090-sglang; Qwen serving recipes organized with one folder per model and instance, including RTX 3090 SGLang and Intel Arc B70 configurations
qwen38-flash-next-3090-usb-offload Python stars Qwen3.8-Flash-Next EXL3 3bpw on one RTX 3090 over USB plus a Mac mini M4: patches, shim, kernels, results

Archived repositories

Historical repositories retained for reference; use the active projects above. local-studio-site is superseded by local-studio/site/ on the integration branch, but its GitHub archive flag is not yet set.

Repository Language Stars Description
exl3xpu Python stars Archived EXL3 inference on Intel Arc Battlemage: bit-exact ESIMD kernels as a vLLM XPU plugin
framework-research Python stars Archived research artifacts: Strix Halo (Framework Desktop) + RTX 3090 via llama.cpp RPC — 34 missions
sglang-moet Sass stars Archived SGLang-native serving for the Moet sign-symmetric W2 expert format with SM120 W2/W4 kernels
local-ai-usage JavaScript stars Install counter folded into omarchy-local-ai/usage/
local-studio-site TypeScript stars Website folded into local-studio/site/ on the integration branch; superseded standalone repository
qwen36-b70 Python stars Qwen3.6-35B-A3B on Intel Arc B70 + RTX 3090; consolidated into qwen-serve/qwen3.6/b70/
qwen38-b70 Python stars Qwen3.8-27B on Intel Arc Pro B70; consolidated into qwen-serve/qwen3.8/b70/
qwen38-flash-next-b70-offload Python stars Qwen3.8-Flash-Next with NVMe expert offload on Intel Arc Pro B70; consolidated into qwen-serve/qwen3.8-flash-next/b70-offload/

Focus areas

  • Local model lifecycle — install, launch, and supervise inference engines on your own hardware.
  • Validated recipes — one measured, reproducible configuration per GPU, evidence attached.
  • Offload tiers — expert placement across VRAM, RAM, and NVMe for models bigger than the card.
  • Serving engines and kernels — EXL3 trellis, Intel XPU, and SGLang expert-format work.
  • OpenAI-compatible proxies — point existing tools at local endpoints without changing call sites.

Links

Profile README lives in sybil-solutions/.github. Edit profile/README.md to update this page.

Popular repositories Loading

  1. local-studio local-studio Public

    Control panel for VLLM, Sglang, llama.cpp, exllamav3

    TypeScript 1.8k 168

  2. ai-data-extraction ai-data-extraction Public

    extract all your personal data history from cursor, codex, claude-code, windsurf, and trae

    Python 1.4k 115

  3. codex-shim codex-shim Public

    Local Responses-API shim that exposes Factory BYOK models (and optional ChatGPT GPT-5.5 passthrough) to Codex Desktop.

    Python 1.1k 102

  4. glm-flash-lite glm-flash-lite Public

    GLM-5.3-Flash EXL3 on one 24 GB RTX 3090 + DDR4: elastic GPU expert cache, zero-copy experts, AVX2 CPU tier, OpenAI API

    Python 321 35

  5. local-ai-registry local-ai-registry Public

    Local AI registry: one validated recipe per machine, with the evidence attached

    HTML 260 92

  6. dsv41-flash-offload dsv41-flash-offload Public

    DeepSeek-V4.1-Flash EXL3 on one 24 GB RTX 3090 + DDR4 + NVMe: staged-DMA prefill, AVX2 CPU expert tier, elastic VRAM expert cache, Engram on disk, OpenAI API

    Python 151 22

Repositories

Showing 10 of 25 repositories
  • local-ai-registry Public

    Local AI registry: one validated recipe per machine, with the evidence attached

    sybil-solutions/local-ai-registry's past year of commit activity
    HTML 260 MIT 92 15 31 Updated Oct 23, 2026
  • spark-stack Public

    One container that prepares, serves, frees, and reports on three models across two DGX Sparks, wired to the DeepSeek harness

    sybil-solutions/spark-stack's past year of commit activity
    Python 0 0 0 0 Updated Oct 22, 2026
  • local-ai-recipe-kit Public

    Run a big model on your own GPU, measure it with one protocol, and submit the result. Intake for local-ai-registry recipes.

    sybil-solutions/local-ai-recipe-kit's past year of commit activity
    Python 22 MIT 9 1 24 Updated Oct 22, 2026
  • qwen38-flash-next-3090-usb-offload Public

    Qwen3.8-Flash-Next EXL3 3bpw on one RTX 3090 over USB plus a Mac mini M4: patches, shim, kernels, results

    sybil-solutions/qwen38-flash-next-3090-usb-offload's past year of commit activity
    Python 3 MIT 0 0 0 Updated Oct 19, 2026
  • local-ai-images Public

    Attested container images pinned by the local-ai registry

    sybil-solutions/local-ai-images's past year of commit activity
    Python 8 7 2 6 Updated Oct 19, 2026
  • .github Public

    Organization profile and dashboard for Sybil Solutions

    sybil-solutions/.github's past year of commit activity
    Shell 0 0 1 1 Updated Oct 18, 2026
  • deepseek-v4.1-flash-1017-exl3-spark Public

    DeepSeek-V4.1-Flash-1017 EXL3 2.7 bpw for 2x DGX Spark: full recipe (sudoingX method), calibration builder, plan, logs, panel KLD

    sybil-solutions/deepseek-v4.1-flash-1017-exl3-spark's past year of commit activity
    Shell 4 Apache-2.0 0 0 0 Updated Oct 18, 2026
  • glm-flash-lite Public

    GLM-5.3-Flash EXL3 on one 24 GB RTX 3090 + DDR4: elastic GPU expert cache, zero-copy experts, AVX2 CPU tier, OpenAI API

    sybil-solutions/glm-flash-lite's past year of commit activity
    Python 321 MIT 35 4 3 Updated Oct 18, 2026
  • deepseek-v4.1-flash-4x-rtx-pro-6000 Public

    DeepSeek V4.1 Flash on 4x RTX PRO 6000 Blackwell: native weights, DSpark, vision, NVMe or locked RAM Engram offload.

    sybil-solutions/deepseek-v4.1-flash-4x-rtx-pro-6000's past year of commit activity
    Python 78 MIT 13 2 2 Updated Oct 17, 2026
  • qwen-serve Public

    Fastest validated sglang recipe for Qwen3.8-27B (AWQ-INT4 + DSpark) on RTX 3090s — 1×/2×/4× GPU presets, image+video, full benchmarks

    sybil-solutions/qwen-serve's past year of commit activity
    Python 64 6 4 5 Updated Oct 17, 2026
View all repositories

People

This organization has no public members. You must be a member to see who’s a part of this organization.

Most used topics

Loading…