no-gpu

CPU inference for Kimi K3, a 2.78T-parameter MoE LLM, in pure Rust. No GPU, no BLAS, no PyTorch. Streams the checkpoint from disk. Byte-identical port of kimi-k3-in-c.

🎙️ AI-powered Telegram bot for voice-to-text transcription using OpenAI Whisper. CPU-only, no GPU required, privacy-focused with local processing.

Collama - Run Ollama Models on Google Colab

Pixel art without GPU. Any text-only LLM can draw. Size + prompt → self-contained HTML. Claude Code skill / MCP compatible.

A biomorphic neuromorphic inference engine inspired by cricket auditory neuroscience — performing real-time temporal pattern recognition via delay-line coincidence detection, without matrix multiplication.

DeepSeek V4 Flash prompt engineering + Flux free image generation. Zero cost, no GPU, China-friendly AI image workflow.

No-GPU 3D scanning from phone rotation videos — pure geometry, no neural networks, no cloud

A renderer that doesn't sample. It solves. Noise-free CPU global illumination — no Monte Carlo, no denoiser, no GPU.

Zero-memory, multi-core-CPU local AI training for ordinary laptops.

A 3D software rendering engine built from scratch in Rust. Implements a custom graphics pipeline, linear algebra library, and scene graph without hardware acceleration APIs.

Cascade MoE: a CPU-only expert system that turns plain-language sysadmin and networking questions into shell commands. No GPU, no cloud, no external model weights. Russian-first: 80% of held-out Russian queries answered with a working command, 38% in English. Patent pending. The benchmark that measures it ships with the source.

🎬 PhantomRec: The Ultimate Low-End Screen Recorder — Smooth 60 FPS on Weak PCs, Dual-Core CPUs, No GPU Required

微软免费线上 TTS 语音转音频（不用显卡）| Microsoft Edge TTS Workbench — free online text-to-speech, no GPU required

low-ends games web

Change text in scanned documents & photos using the document's own letters — typewriter, printed fonts, handwriting. Python + OpenCV, no GPU.

CPU-trained reasoning model pipeline. LoRA SFT + DPO on SmolLM2-360M, GSM8K math reasoning, single-laptop deployment.

Pixel-perfect click coordinates for AI agents. Vision models miss by 30–40 px; OCR + template matching land on the exact pixel. Pure-CPU screen automation — no GPU, no cloud API, no token cost.

Run real transformer/LLM architectures in your browser - loads actual Hugging Face weights, executes a live forward pass client-side (no backend, no GPU), and lets you inspect tensors and attention.