cmux is the open-source terminal built for coding agents. Built by @lawrencecchen @austinywang

San Francisco, California
Joined June 2024
cmux is coming to stanford!
Iโ€™m excited to finally announce the newest edition my Stanford course ๐—ง๐—ต๐—ฒ ๐— ๐—ผ๐—ฑ๐—ฒ๐—ฟ๐—ป ๐—ฆ๐—ผ๐—ณ๐˜๐˜„๐—ฎ๐—ฟ๐—ฒ ๐——๐—ฒ๐˜ƒ๐—ฒ๐—น๐—ผ๐—ฝ๐—ฒ๐—ฟ. It has been 9 months in the making. Last November, with the release of Claude Opus 4.5, coding agents experienced a step function improvement in capability. We all felt it. The LLMs were more powerful, could reason for longer, solve harder tasks. This yearโ€™s iteration of my course reflects the 2026 metamorphosis of software engineering. My core belief is simple: AI-native developers of the LLM era are going to become the most important members of any software organization. I have designed my course to train this next generation of engineers. ๐—ช๐—ต๐—ฎ๐˜โ€™๐˜€ ๐—ฑ๐—ถ๐—ณ๐—ณ๐—ฒ๐—ฟ๐—ฒ๐—ป๐˜ ๐˜๐—ต๐—ถ๐˜€ ๐˜๐—ถ๐—บ๐—ฒ ๐—ฎ๐—ฟ๐—ผ๐˜‚๐—ป๐—ฑ First, 85% of my Fall 2025 class material is being thrown out.ย The Fall 2026 syllabus reflects the core capabilities AI-native engineers must have: agent skills, advanced context engineering, MCP portals, agent-ready codebase principles, agentic code review, security, parallelizing background agents, software factories, and more. Second, I am going to teach my students how to have software taste. Every student will be required to ship pull requests to production-grade, real-world codebases. The course is collaborating with the top open-source AI repos who will offer support and mentorship to students on how to meaningfully contribute to their projects. This has never been done before in any university course so I am incredibly grateful to our OSS Partners:ย @browserbase, @HeyGen, @CopilotKit, @semgrep, @OpenHandsDev, @milvusio, @marimo_io, Pi, @crewAIInc, @warpdotdev, @vercel, @cmux, @arizeai, @UnslothAI, and @anyscalecompute. ๐—ช๐—ต๐—ฎ๐˜โ€™๐˜€ ๐˜€๐˜๐—ฎ๐˜†๐—ถ๐—ป๐—ด ๐˜๐—ต๐—ฒ ๐˜€๐—ฎ๐—บ๐—ฒ Iโ€™m fortunate to again have AI software engineering leaders and founders as guest speakers to share their learnings from building top coding agent products. Thank you to @leerob from @cursor_ai, @bcherny of @claudeai code, @EnoReyes of @FactoryAI, @silasalberti of @cognition, @0xine of @semgrep, Rajesh Bhatia of @Cloudflare , @amasad of @Replit, and @eladgil. All resources will be available online. All classes will be available to the public. 9/22 on Stanford campus. See you in class.ย  themodernsoftware.dev/
4
2
50
3,352
cmux will be an open-source partner for this course at Stanford. See you in class!
Iโ€™m excited to finally announce the newest edition my Stanford course ๐—ง๐—ต๐—ฒ ๐— ๐—ผ๐—ฑ๐—ฒ๐—ฟ๐—ป ๐—ฆ๐—ผ๐—ณ๐˜๐˜„๐—ฎ๐—ฟ๐—ฒ ๐——๐—ฒ๐˜ƒ๐—ฒ๐—น๐—ผ๐—ฝ๐—ฒ๐—ฟ. It has been 9 months in the making. Last November, with the release of Claude Opus 4.5, coding agents experienced a step function improvement in capability. We all felt it. The LLMs were more powerful, could reason for longer, solve harder tasks. This yearโ€™s iteration of my course reflects the 2026 metamorphosis of software engineering. My core belief is simple: AI-native developers of the LLM era are going to become the most important members of any software organization. I have designed my course to train this next generation of engineers. ๐—ช๐—ต๐—ฎ๐˜โ€™๐˜€ ๐—ฑ๐—ถ๐—ณ๐—ณ๐—ฒ๐—ฟ๐—ฒ๐—ป๐˜ ๐˜๐—ต๐—ถ๐˜€ ๐˜๐—ถ๐—บ๐—ฒ ๐—ฎ๐—ฟ๐—ผ๐˜‚๐—ป๐—ฑ First, 85% of my Fall 2025 class material is being thrown out.ย The Fall 2026 syllabus reflects the core capabilities AI-native engineers must have: agent skills, advanced context engineering, MCP portals, agent-ready codebase principles, agentic code review, security, parallelizing background agents, software factories, and more. Second, I am going to teach my students how to have software taste. Every student will be required to ship pull requests to production-grade, real-world codebases. The course is collaborating with the top open-source AI repos who will offer support and mentorship to students on how to meaningfully contribute to their projects. This has never been done before in any university course so I am incredibly grateful to our OSS Partners:ย @browserbase, @HeyGen, @CopilotKit, @semgrep, @OpenHandsDev, @milvusio, @marimo_io, Pi, @crewAIInc, @warpdotdev, @vercel, @cmux, @arizeai, @UnslothAI, and @anyscalecompute. ๐—ช๐—ต๐—ฎ๐˜โ€™๐˜€ ๐˜€๐˜๐—ฎ๐˜†๐—ถ๐—ป๐—ด ๐˜๐—ต๐—ฒ ๐˜€๐—ฎ๐—บ๐—ฒ Iโ€™m fortunate to again have AI software engineering leaders and founders as guest speakers to share their learnings from building top coding agent products. Thank you to @leerob from @cursor_ai, @bcherny of @claudeai code, @EnoReyes of @FactoryAI, @silasalberti of @cognition, @0xine of @semgrep, Rajesh Bhatia of @Cloudflare , @amasad of @Replit, and @eladgil. All resources will be available online. All classes will be available to the public. 9/22 on Stanford campus. See you in class.ย  themodernsoftware.dev/
1
1
22
2,100
cmux (@manaflowai) iOS app beta is awesome
3
3
1
11
1,192
The creator of oh-my-codex shared this today: cmux + streamdeck + gajae code
5
1
29
2,906
cmux iOS has an interactive simulator pane that streams iOS simulators from your Mac!
This Post is from an account that no longer exists. Learn more
6
3
53
9,255
terminal-browser now officially supports cmux!
9
19
1
313
43,083
I was complaining about cmux performance, but after visiting @manaflowai office, they helped me to set up it properly. Now I understand the beauty of cmux: you can make tmux sessions, but with GUI, and auto restore That is a game changer, I can't live without this anymore
1
2
3
480
Running libghostty on iOS! Scrolling now feels so smooth ๐Ÿ‘€
3
5
101
9,273
I demoed at a devtools meetup and the devs demoing were using: Ghostty - 10 cmux - 3 VSCode/Cursor - 3 Codex app - 1 Other - 4 ๐Ÿค”
6
2
1
32
7,003
Demoed our cmux iOS app in beta at @WorkOS! Found some cmux users here too ๐Ÿ‘€
3
2
1
34
2,599
cmux spotted in the wild at @WorkOS !
2
2
39
2,524
My most used cmux shortcuts โŒ˜+Shift+P, then: fork tab -> Fork Conversation to New Tab fork right -> Fork Conversation to the Right pin -> Pin Workspace top -> Move to Top โŒ˜+Shift+U -> Focus + flash latest unread pane โŒ˜+Shift+L -> New browser tab โŒ˜+L -> Focus browser omnibar, else new browser tab โŒ˜+Shift+Enter -> Maximize pane โŒ˜+Control+W -> Close workspace โŒ˜+W -> Close tab โŒ˜+Shift+M -> Toggle Vim mode (then "gg" to get to top and "shift+g" to get back to bottom" โŒ˜+Shift+R -> Rename workspace โŒ˜+R -> Rename tab โŒ˜+D -> Split right โŒ˜+Control+= -> Equalize splits โŒ˜+N -> New workspace โŒ˜+T -> New tab What else should we add?
3
4
54
5,363
Iโ€™m building a new thing on cmux similar to symphony and its been helping me get through 1.6k issues on GitHubโ€ฆ Requires some setup to get your agents to write good code and close the loop but knowing when it is fully confident vs not is super keyโ€ฆ
6
4
49
5,434
What if cmux's sidebar had multiple columns? Left-most column is for machines (this MacBook, Mac Mini, cloud VM), right column is workspaces in that machine. Maybe another column for agents in that workspace?
26
4
1
53
8,394
This is my exact workflow btw
I used 367 billion tokens in the last 30 days by multitasking with cmux My peak per-day token usage is 37 billion tokens on GPT 5.6 Sol Would have cost $85k over API but we only paid $7.4k (37 ChatGPT Pro 20x accounts + tons of tibo resets... and this is only my usage, since subrouter lets my cofounder share usage with me too (and CodexBar only counts local token usage))
3
3
1
79
9,366
Quick video on how I used 367 billion tokens in the last 30 days by multitasking with cmux My peak per-day token usage is 37 billion tokens on GPT 5.6 Sol My workflow is pretty simple. The 80/20 is that I press Cmd+Shift+U whenever Codex notifies me, and I review the cmux dev build/artifact/bug fix, and respond. And then Cmd+Shift+U and repeat... If there's ever a point in time where there's no more notifications that I need to Cmd+Shift+U to, I start a new workspace with a new task. I don't use fancy harnesses (just vanilla Claude Code/Codex), I don't do fancy agent orchestration either (I only use the built in subagents from Codex/Claude Code), I don't use /loop, and I barely use /goal. This would have cost $85k over API but we only paid $7.4k (37 ChatGPT Pro 20x accounts + tons of tibo resets... and this is only my usage, since subrouter lets my cofounder share usage with me too (and CodexBar only counts local token usage)) xitter.cf/lawrencecchen/status/2โ€ฆ
I used 367 billion tokens in the last 30 days by multitasking with cmux My peak per-day token usage is 37 billion tokens on GPT 5.6 Sol Would have cost $85k over API but we only paid $7.4k (37 ChatGPT Pro 20x accounts + tons of tibo resets... and this is only my usage, since subrouter lets my cofounder share usage with me too (and CodexBar only counts local token usage))
11
9
4
83
7,944