K-Dense BYOK: AI research assistant on your computer
K-Dense BYOK
A free open-source app where the AI assistant Kady handles research work with data and literature on your computer
- 1,412 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Kady writes and runs code in the project and works with files, including a shell and Python
- Connected keys and subscriptions are spent on your behalf, and request material goes to the model provider
Install
Manual install
git clone https://github.com/K-Dense-AI/k-dense-byok.git && cd k-dense-byok && cp .env.example .env && ./start.shFor macOS and Linux.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Kady writes and runs code in the project and works with files, including a shell and Python
- Connected keys and subscriptions are spent on your behalf, and request material goes to the model provider
This is third-party code. Review the repository files before installing.
What it does
K-Dense BYOK is a free open-source app that runs an assistant called Kady. You describe a task in plain words: analyze a dataset, check a manuscript, find literature, build a figure. Kady writes and runs the code itself, shows progress and saves results into ordinary project folders on your computer. The assistant reads web pages, PDFs, GitHub repositories and YouTube videos, checks claims against sources with exact quotes and can create Word, Excel, PowerPoint and PDF files. Schematics and graphical abstracts can be generated with an OpenRouter image model, while data plots are built with code so they can be checked and regenerated. A Paperclip key adds search across papers, preprints, clinical trials, FDA documents and patents. The app keeps a lab notebook of reasoning and a separate automatic log of the code that actually ran and the files it created. The notebook is linked by a hash chain. Spending on paid models is tracked against project limits, and MCP servers are connected in settings for one project or all of them.
Who it is for. For scientists and graduate students who want to hand the routine of data analysis and literature work to AI without sending project files to someone else storage.
Good fit when
- Clean a table, pick a statistical method, compare groups and get a publication-ready figure
- Find and compare papers, extract methods and check a claim against quotes from sources
- Review a manuscript and draft a reply to reviewers with a point-by-point breakdown
- Put together a Word, Excel or PowerPoint report after the analysis
- Work with sensitive data on a local model through Ollama with nothing sent out
Not a fit when
- You want a small helper inside Claude Code or Cursor: K-Dense BYOK runs as a separate app in the browser
- You need an exact result with no human check: Kady output must be verified and the app is in beta
- You have no computer with Node.js and Git: on Windows you need Node.js 22 and Git for Windows
Example request
Run a differential expression analysis on counts.csv comparing treated and control and plot a volcano plot.Limitations
The app is in beta. On first start the app spends a few minutes installing dependencies. On Windows Kady runs commands through Git Bash from Git for Windows.
The app is free, and model costs remain yours: you need an OpenRouter key, a key from another provider, a connected subscription or local Ollama models. A subscription does not mean unlimited use.
Comparison tables with other systems were compiled by the authors themselves and are not an independent evaluation.
How to disable. Stop the app with Ctrl+C in the terminal and delete the repository folder. Project data sits in ordinary folders and is deleted separately; revoke provider keys in their dashboards.
FAQ
How do I start K-Dense BYOK?
Run `git clone https://github.com/K-Dense-AI/k-dense-byok.git`, enter the folder, optionally copy .env.example to .env and run ./start.sh. After a few minutes the browser opens http://localhost:3000. On Windows use .\start.cmd.
Which models can I connect?
An OpenRouter key, a key from a Pi-supported provider (Anthropic, OpenAI, Google, Groq, Mistral, DeepSeek, NVIDIA NIM, Azure, Bedrock, Vertex), a ChatGPT, Claude Pro or Max, GitHub Copilot, xAI, Kimi Code or Meta Muse subscription, or free local models through Ollama.
Where is my data stored?
Projects, conversations, notebooks and results live in ordinary folders on your computer, and K-Dense does not store them. With a cloud model the material needed for a request goes to the chosen provider. Use a local Ollama model to keep data on the machine.
Can I connect MCP servers?
Yes. In settings you add reference managers, GitHub, databases and other services for one project or all, test the connection and check status. Large servers are called through short scripts by default so every tool is not loaded into the conversation.
How is K-Dense BYOK different from a skill set for Claude Code?
It is a separate app with a project interface, a step log, spending limits and model choice. Scientific skills for Claude Code from the same K-Dense catalog install directly into the agent, while here Kady uses them inside its own workspace.
Discussion
No questions or reviews yet.
Related
CLI toolsA Claude Code plugin and MCP server: over 1000 scientific tools for biology, chemistry and medicine, data-check commands and routing skills
A CLI and MCP server that extracts structured knowledge from documents: graphs, hypergraphs, and fact search with sources
A Google DeepMind skill collection for science tasks: access to biology and chemistry databases, literature search and structural biology
A local MCP server that turns a folder of documents into a knowledge base with hybrid search for Claude Code and Cursor
MCP server that connects Claude Code and Codex to an open R or Python session in RStudio and Positron, where your data lives
A lightweight database workspace with an MCP server: the agent reads the schema, runs queries and explores data in PostgreSQL, MySQL and more
README in short
Kady is meant as an AI co-worker for research that runs on your computer. On the public CompBioBench test from Genentech, 100 computational biology tasks, its best run scored 98% and held second place as of September 11, 2026. There are video walkthroughs and an arXiv paper.