TokenSpeed is a speed-of-light LLM inference engine.

CPU-first, eager-execution tensor/autograd runtime and LLM inference engine written in pure Zig

Samples and documentation for using VP Link with Microsoft Bonsai

Faster attention kernels for serving TML's Inkling model on vLLM. 2.7x over the shipping path on H100, and the only implementation that runs on A100.

Tools for converting vector graphics from the Wacom Inkling device

Run Thinking Machines Lab's Inkling (975B-A41B MoE, multimodal) on Apple Silicon with MLX — 4/6/8-bit.

Self-hosted AI résumé screening with a built-in bias audit. Fine-tuned Inkling model, calibrated scores, blind review, human-in-the-loop. Your data never leaves.

Is Inkling AI the Ultimate Open Source Model? Full Test - A 975B Mixture-of-Experts (MoE) multimodal foundation model by Thinking Machines, with local setups for vLLM, SGLang, Hugging Face, TokenSpeed, and Unsloth, plus three advanced agentic/epistemic benchmarks.

A set of documents, fonts, and other utilities to aid in support of the Inkling fan-language.

the image files for the boo-yah doc

Agentic terminal chatbot for thinkingmachines/Inkling — tool use, permission policy engine, and a glass TUI