Skip to content
View SSHdotCodes's full-sized avatar
16 followers · 4 following

Achievements

Achievement: Pull Sharkx2Achievement: Pair ExtraordinaireAchievement: Starstruck

Achievements

Achievement: Pull Sharkx2Achievement: Pair ExtraordinaireAchievement: Starstruck

Block or report SSHdotCodes

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
SSHdotCodes/README.md
SSH.codes — Build. Measure. Ship. A detailed Blender render of my real workstation at a fixed angle, with blue-lit spinning fans. Click to explore it on ssh.codes.

Website  ·  Models & datasets  ·  Build log  ·  X  ·  Get in touch

I'm SSH. I build open-weight models, fast inference stacks, and tools that run on your machine. I like measured results, source you can inspect, and projects you can actually use.

The workstation above is a detailed Blender reconstruction of my real machine, rendered with blue lighting and spinning fans. Explore the interactive version on ssh.codes to rotate it and change the lighting.

Models, measured

The Grug Series — Open-weight models trained to spend fewer tokens thinking, with GGUF builds for local use.

Auto — Local permission intelligence for coding agents. Checks proposed tool calls against your request and agent history, with support for Pi, OpenCode, and Hermes.

Black Hole Benchmark — One prompt, different models, actual compiled C++ renders. Inspect the submissions →

Selected builds

Project What I’m building
GLM on Blackwell Single-GPU GLM-5.3-Flash serving, RAM expert caching, DFlash2, and reproducible benchmarks.
Qwen on Blackwell Qwen3.8-Flash-Next inference tuning, full-context serving, and correctness checks on an RTX PRO 6000.
Second Opinion Let Claude bring other AI models into the work through MCP.
Videopaper Native macOS live wallpapers, including a real-time black hole ray tracer in Metal.
Codex RPC Discord Rich Presence for the Codex CLI, with animated activity states.

More projects on ssh.codes →

The workshop

Models & inference   PyTorch · LoRA · GGUF / llama.cpp · CUDA · MLX
Interfaces & systems   Python · C++ · TypeScript · Swift · Metal · WebGPU · Three.js · MCP
Hardware & hosting   RTX PRO 6000 Blackwell · Apple Silicon · Raspberry Pi · Docker

Contribution trail
Snake animation of my GitHub contribution graph

Have something interesting to build? me@ssh.codes

Pinned Loading

  1. BetterWright/betterwright BetterWright/betterwright Public

    A persistent, policy-guarded Playwright browser for AI agents — network policy, encrypted credential vault, proof screenshots, and CAPTCHA solving.

    TypeScript 304 20

  2. qwen-3.8-flash-next-pro6000 qwen-3.8-flash-next-pro6000 Public

    SGLang 0.5.20 for Qwen3.8-Flash-Next on one RTX PRO 6000: 262K context, exact sampled MTP, validated hot-token/QSA patches, reproducible 8K-token benchmarks and real-task checks.

    Python 9 1

  3. bhbench bhbench Public

    Original source submissions behind bhbench.ssh.codes

    Makefile 3

  4. glm-5.3-flash-pro6000 glm-5.3-flash-pro6000 Public

    GLM-5.3-Flash UD-IQ3_XXS on one RTX PRO 6000: 56.84 tok/s across 2K-token probes, 512K context, native vision, RAM expert caching and DFlash2. Pinned source and reproducible benchmarks.

    Python 1