Lynx: The open source AI workspace you can fully customize

Docs for humans - Docs for agents - Community - Roadmap - Download for macOS

About

Lynx is a fully-customizable AI workspace designed to keep you organized and fit your work patterns.

Most of our interactions with AI today happen through walls of text that are tiring to parse through and inefficient for working. Active AI users also quickly find themselves with more threads open than they can manage.

Lynx solves this by putting interactive interfaces at the center of our interaction with agents. Workspaces in Lynx are fully-customizable UIs, connected to the services you use, and able to trigger agents using your local Claude Code or Codex.

Features

• 100% local: Lynx runs on your device only and does not send data anywhere.
• Use your own subscriptions: Lynx makes use of Claude Code and Codex running on your machine. Switch between them as you like.
• Connect any MCP or API: Connect any service you use via MCP or API. You can also ask Lynx to help you connect anything else.
• Templates: Official and community workspace templates for you to start from and modify to your needs.
• Fully customizable workspaces: Create as many fully custom workspaces as you want and quickly switch between them.
• Point to edit: The point tool makes it easy to point out exactly what you want changed in a UI. Changes go live as soon as the agent is done editing.
• Scheduled tasks: Schedule tools and agents to run on a cron.

Watch demo

https://github.com/user-attachments/assets/246e20b2-1a1c-4563-94e7-08c9f9db1f2f

Why Lynx?

As of today, Lynx is an experiment about a better way to work with AI. It's so far proven to be very useful to me and some early testers, which is why I'm releasing it publicly, but it remains an experiment. For what it's worth though, I've done over 80% of my daily work from Lynx for a couple weeks now.

There were a few reasons for building it:

1. Interacting with agents through walls of text was getting tiring and inefficient
2. It was hard to keep track of everything my agents were doing
3. Harnesses for coding felt like the wrong shape for other types of work
4. Interactive tools and artifacts proved very useful, but ended up fragmenting work even more
5. It was hard to stay in the flow jumping through various tools / programs / websites

From this, I decided to explore a way of working that was more visual and interactive, while still feeling cohesive rather than fragmented.

A lot of people and companies seem to have come to the conclusion that UIs are still very relevant for humans. Many of us still prefer to click a button that says "Do X" rather than tell an agent "Do X for me". We also like graphs, tables, and colors.

For this reason we've seen artifacts take off, and also MCP-UI starting to become more relevant (basically the little visualizations inside ChatGPT chats).

But these still feel additive to a chat experience, rather than central to our interaction with AI. Most times we're still driving everything from a chat, just using these visual tools as a resource along the way.

So with Lynx, that concept was flipped on its head. Interactive UIs are at the core of the experience, and the chat is there to augment it.

To-dos / task tracking example

A typical workflow for a lot of people includes waking up and asking Claude: "Check Jira for tasks assigned to me and tell me the important ones to focus on today". They then receive a bunch of text back, and go back and forth on it via chat until they've gathered the necessary context to work

With Lynx, you'd create a workspace called "Task Tracking" that connects to Jira and displays your list of tasks in a nice UI. Every morning an agent would look through these and prioritize for you, so that when you open the workspace you already have a clear overview of what's going on. Then, instead of going to a chat to ask about a specific task, you can add a button to trigger an agent to triage it, or even start working on it. To find the output of this agent you wouldn't need to search through dozens of threads, you just go to your Task Tracking workspace and see it from the UI.

Other example workflows

Below are real examples of things people are doing with Lynx, with actual screenshots of Lynx workspaces.

Document editing

Why reinvent something like Google Docs inside Lynx? While it's likely not relevant for most people, being able to make small tweaks to the UI and using your own AI subscription connected to your data can be really powerful.

Coding

Because Lynx hooks into your coding agents and runs on your own machine, you can build your own agent orchestrator for coding the way you want it to be.

Analytics dashboards

This very README

This README was completed much faster due to two Lynx workspaces that were super helpful:

GIF Maker

A workspace that turns videos into high-quality GIFs using local ffmpeg. Could have been a script, but works much better with a UI.

README Editor

Better visual editing for GitHub READMEs with native support for media. Took 10min and one prompt:

"build me a github readme editor where I have the full markdown experience but can also put in images gifs etc and have them go to a specific dir that makes it work when I push. should be able to push directly too."

Games :)

License

MIT.