K-Dense Web Updates

// Apr 24, 2026

A big week. We shipped two new open-source projects, a free web app, three new blog posts, major updates to K-Dense BYOK, and GPU support in K-Dense Web. Here is everything you need to know.

>New Open Source: mimeo + mimeographs

Frontier LLMs are smart, but they are not anyone in particular. mimeo is a pipeline that reads the internet on your behalf (essays, talks, interviews, letters, papers) and distills how a specific person actually reasons into a SKILL.md or AGENTS.md file your agent can load. mimeographs is our companion catalog of 80+ ready-to-use experts already generated with it, from Walter Willett and Aviv Regev to Karpathy, Wittgenstein, and Buffett.

•mimeo on GitHub – generate a new expert with uv run mimeo “Name”

•mimeographs on GitHub – install with npx skills add K-Dense-AI/mimeographs

>Try It Free: Pantheon

One question, 80 voices. Pantheon takes a research question and streams live answers in parallel from every mimeograph we’ve generated, each in their own voice, with cited web sources and a cross-persona consensus at the end. It is the fastest way to feel what a mimeograph actually does.

Ask a question at pantheon.k-dense.ai →

>Three New Blog Posts

•Introducing mimeo and 80+ Mimeographs – why cloning an expert’s way of thinking into a SKILL.md is the right lever for steering your agent

•Skill Security Before You Install – a practical guide to prompt injection, poisoned scripts/, the Cisco AI Defense Skill Scanner, version pinning, and a pre-install checklist every lab should adopt

•The Sandboxed AI Scientist – pairing NVIDIA OpenShell with Scientific Agent Skills to run autonomous research agents on patient data, proprietary molecules, and HPC credentials without losing sleep

>K-Dense Web Now Runs on GPUs

Your agents can now spin up GPU-backed compute directly inside K-Dense Web for GPU-optimized analyses and deep learning model training. Think AlphaFold runs, single-cell foundation models, diffusion-based generative chemistry, and custom PyTorch training jobs, all without leaving the platform.

>K-Dense BYOK Updates

The free desktop co-scientist just got a lot better:

•Local model support via Ollama. Run Kady entirely on-device with Qwen 3.6, Gemma 4, and other skill-capable open models. No API keys required.

•Better project management. Files, chat history, and settings are properly scoped per project, with richer previews for notebooks, PDFs, and bioinformatics formats.

•Significantly improved performance. Faster skill activation, snappier UI, and a smoother handoff between Kady and the specialist expert agents.

•Built-in LaTeX editor and compiler. A split-pane editor with live PDF compilation (pdfLaTeX, XeLaTeX, LuaLaTeX) so you can draft and compile manuscripts without leaving the app.

Get K-Dense BYOK on GitHub →

>Office Hours Recap

On April 17 we covered open-source model support on the BYOK roadmap, which models perform best on Scientific Agent Skills, how we compare to Claude CoWork, petabyte-scale data workflows, and the enterprise deployment path. Next session is May 20 – hope to see you there.

Read the full recap →

>Research Grant Program: 90% Off for Labs

If you are at an academic lab or a non-profit research organization, you qualify for the K-Dense Team plan at 90% off: $449.90/year for unlimited seats, 800 monthly credits, and every feature in K-Dense Web including GPU compute. Applications are reviewed within 48 hours.

Apply to the Research Grant Program →

>Stay in the Loop

For release notes, tutorials, and behind-the-scenes research updates between newsletters, follow us on X, LinkedIn, and YouTube.

Until next week,

The K-Dense Team

K-Dense Inc.

380 Portage Avenue

Palo Alto, CA 94306

Unsubscribe