> Open-sourcing is fundamentally coming from our desire to build a successful business. We are competing with other highly funded, closed-source competitors, and we think opening and providing the resources for the community to improve Warp is a smart way for us to accelerate product development. Yes, we are a VC funded startup, but we do not have the resources to compete on price or massively subsidize usage – we need to build our business by offering the best possible product to the most excited community.
Who are their highly funded closed-source competitors they claim Warp cannot beat on price?
Warp is the only closed source terminal product I know of. Most other popular terminal emulators are open source already.
I feel like their funding is drying up and this is their last ditch effort to have the "community" build their product for them.
They claim agents will run the show, with inputs from community in the form of ideas/specs/direction. I wonder how long that will be sustainable for given the subsidized model prices are collapsing as we speak.
Is this an attempt to pivot to something else while the "community" keeps their first product alive? Maybe I'm being too cynical here, but I don't see this as an act of good faith, especially given their roots in VC funding.
I'm not too familiar with Warp so can someone help clarify for me:
Is Warp a terminal? Or an agent harness? Or both?
Warp as a terminal to me seems less interesting than having a well built agent harness like OpenCode that can effectively use many different models. If it's both, is there any advantage to having them be the same thing? Like, is there any way your harness can be smarter if it is also tightly integrated in your terminal? Or is it just something that Warp happens to do both of?
It's a terminal. It was a terminal before they pivoted to AI (I don't blame them, with their funding rounds I don't see them having particularly free reins) Before this, it was all about collaborative (CRDT) features. I have no idea why you'd want a terminal that is also an agent harness, but I appreciate them making it open source none the less.
The problem description is spot on, but the solution isn't. No-one is going to sit in that chat and "collaborate" on each other's stuff in real time all day. You may as well just all sit around a screen.
I welcome the experimentation, there will definitely be something new, but this ain't it. New primitives are needed, at a higher level of conceptualization, not merely a fancy new interface.
My hobby projects have 100x more tests than they used to, because LLMs are great at writing tests. And my subjective experience is that the net quality has increased as a result.
YMMV, but it’s certainly a common belief, and for me at least a lived experience.
the presenter is pretty sane, but the product is hardly a product at the current scenario. pretty much codemirror 6 collaborative editing demo + vm running claude code, with a web GUI. will fall apart with large code bases just like vscode, github codespaces and co. do, and expensive for llms to run against. Would be nice to see the foundational problems being worked on instead of regurgiting what everybody is doing.
Can you confirm that Warp is NOT initiating any connection to any service whatsoever unless explicitly enabled in Settings?
Warp had an account requirement in the beginning which spoke volumes about the misalignment of values. Now the terminal is not called a terminal, it is "the agentic development environment" (whatever that means) which also lowkey implies that it might have some kind of online features. But at the same time I understand that it is now an absolute requirement to mention AI on any web page for any product.
Absolutely, in heaps. The very moment you start Warp, it sends 5 HTTP requests, (1 version check, 1 LLM model list, 3 telemetry events). The former two go to app.warp.dev, the latter 3 go to warpianwzlfqdq.dataplane.rudderstack.com and include a persistent UUID, your operating system and version, Warp version, and the tracked event name and its properties.
All of this happens before you even see the window. After you're done clicking 'No' to all the SaaS nags, you can turn off telemetry in the settings, but for some reason it gets turned back on when you restart the terminal.
While the terminal is running, it calls home whenever you trip one of the events listed in this 7000-line long file: https://github.com/warpdotdev/warp/blob/d0f045c01bacbd845a63.... Besides all the hosts listed earlier, it also makes requests to o540343.ingest.sentry.io.
I hope someone will create a lightweight version without AI and code editing stuff. The terminal experience is the best, but I don't have any use for the agentic stuff while having claude code, opencode, codex and plenty other options.
It's good feedback. We've tried to make it so there is a single "turn off all the AI stuff" button (and you can opt into plain old terminal during onboarding as well, with no login, etc). Curious if this does the trick?
Hey Zach - one thing I'm really missing is the ability for this to be toggled on/off per device - whilst I love it on my personal devices and want to use AI there, I also want to be able to use and log into warp at work without having to toggle it off, as I can't use AI there.
This is a pretty good used case for vibecoding. “Claude, take this project and rip out all the obnoxious monetization and vendor lock in.” It just might do the trick. I’ve been to get rid of a fair bit of paid software by just cloning the parts I want with little more than a high-level description.
I don’t think the approach of open source as a substitute for a quality program is going to last.
The way it goes beyond just emulating terminal. Multiline input that works like text editor, separated input and output blocks, wrapped shells that keeps the same ux with local and remote shells, the polish.
Sad that they didn't open source the commit history. I would have loved to branch off of like 5 years ago when Warp was just a terminal, rip out all the AI and cloud shit, and turn it into just a nice terminal with some neat features.
As someone who released the source of an app that was always expected to be public I appreciate that it would be interesting but I'm not surprised. If the code isn't being regularly published there is just less incentive to be sure that every commit is "public ready". So when releasing I wanted to do a full review of current code (and especially comments and docs). This was tedious enough and even though we didn't find anything major and only a handful of things that should be cleaned I absolutely wouldn't want to do this for the full history.
Could we have just released it? Absolutely. But I think everyone who contributed felt better knowing that what was released had one final "ready for public" review. Then our regular review process handled that going forward.
Yeah, it would have been a great job for an LLM. Although if you find something in the history you then need to make the annoying choice of history rewriting or just leaving it in.
Their terminal is just Alacritty, why would you do all these extra steps instead of just using Alacritty, or Ghostty? The terminal emulator was never their selling point, the AI wrapper was.
Interestingly i had been building a terminal in rust and libghostty(with Linux and windows supported too) with built-in agent that understands terminal, too.
And the motivation was warp is doing a little bit more than a terminal.
My main driver has been Ghostty but I've been looking at Warp for a while. Warp seems like a full on IDE (~ADE) though, as opposed to a minimalistic terminal. Can anyone add some thoughts? Are these 2 very different?
tangential: I've seen Mitchel tweet that people in SF have ran up to him showing him how they fully riced their Ghostty setup. How many people here have done this and how easy/manageable is it? e.g. just forking the repo and implementing whatever Warp feature I like?
libghostty makes it pretty easy to do. I spent about two weeks setting something up until it was advanced enough to daily drive. I wanted to have a modal workflow similar to vim or tmux copy mode, but without having the overhead of using tmux... that's probably a lot more complicated than "I want Ghostty but with $X tweak". You can poke around in the repo to get a feel for what's involved if you want: https://github.com/milch/mistty
check out yaw terminal for a terminal first experience that also treats ai cli as a first class citizen. and if you're on windows is very dialed into git bash.
I really like Warp. It's a lot nicer to be able to visualize what I'm doing in the terminal. Some people don't like the AI features, but they only activate if you log in.
> we got rid of this requirement a couple of years ago.
Do you regret having this requirement in the first place?
Personal feedback: I live in a terminal 24x7 for the last 30+ years and once Warp came out I wanted to try it out immediately, but I was impressed by the requirement. So I never had a chance to try it out.
Recently I've started to use https://superset.sh as alternative to Warp. After the volks @mastra mentioned it. Very cool open source project.
I'm actually pretty proud of the final setup I've created with it.
Each time I start to implement a new ticket, superset will pull the ticket from linear, create a worktree/workspace, reserve ports, start the servers, start a browser and start Claude with the ticket as instructions.
The cool thing with this setup is, I can have like 10x the same servers running on different ports/worktrees.
Each time an agent is done, I switch to the workspace, look at the browser and can immediately test things.
I used it because of the integrated voice to text (since apple's built in dictation is terrible) but stopped using it when i replaced it with superwhisper
It's not like this is where they started: Warp dates back to 2020 and probably had an initial strategy of "just build a really nice, fast terminal emulator and figure out the money later", which honestly isn't the worst idea in the world.
The AI integration only came years later, and they probably figured maybe they could compete in the space with Cursor, repl.it, etc. And then Claude Code came and just devoured everything, while Google bought Windsurf, Microsoft pushes Copilot, and OpenAI has Codex (while at the same time kitty and ghostty also built really nice, fast terminal emulators)
I agree I don't think they have any real shot and I certainly wouldn't recommend investing in their next round, but it's not like this was their plan all along, they went where the winds were blowing.
As someone who was interested in warp in the early days (new rust terminal!), but who would never use a closed source terminal, this feels like a pyrrhic victory, since I don’t care at all about the AI accoutrements.
I hope someone will make a version of Warp where I can bring in Openrouter key for free. Or any other provider, for that matter. I'd pay $5/10/month just to be able to use it.
I've been trying to figure out what the long term play is here - is it an angling for a frontier lab acquisition? Or does open-sourcing put Warp in the same sort of category as OpenCode - where charging for LLM tokens becomes the main commercial driver?
Holy shit this made my day. Warp’s convenience shell wrapping is amazing. It’s the only terminal where I can actually edit a long command in place rather than copy pasting into an editor and doing so there.
Now I’m more or less assured I can retain this convenience without being forced into more AI crap.
Right, but if all terminals behaved like modern pieces of software, we would take functionality like Warp's as given, instead of suggesting workarounds.
What you describe sorta works, but you lose things like file/dir-based autocomplete, since your editor doesn't know about your shell session.
Well, it hasn't under the OS/2 name, but as the licensed successor ArcOS, its last release was literally this year. (Of course, that's also why OS/2 is pretty unlikely to be open-sourced any time soon: it's actually still being developed and sold!)
Appreciate the candid take. Warp is great.
Warp is the only closed source terminal product I know of. Most other popular terminal emulators are open source already.
I feel like their funding is drying up and this is their last ditch effort to have the "community" build their product for them.
They claim agents will run the show, with inputs from community in the form of ideas/specs/direction. I wonder how long that will be sustainable for given the subsidized model prices are collapsing as we speak.
Is this an attempt to pivot to something else while the "community" keeps their first product alive? Maybe I'm being too cynical here, but I don't see this as an act of good faith, especially given their roots in VC funding.
Agent AI, broadly speaking. Including CLI agents and IDEs.
Is Warp a terminal? Or an agent harness? Or both?
Warp as a terminal to me seems less interesting than having a well built agent harness like OpenCode that can effectively use many different models. If it's both, is there any advantage to having them be the same thing? Like, is there any way your harness can be smarter if it is also tightly integrated in your terminal? Or is it just something that Warp happens to do both of?
https://maggieappleton.com/zero-alignment/
I welcome the experimentation, there will definitely be something new, but this ain't it. New primitives are needed, at a higher level of conceptualization, not merely a fancy new interface.
My hobby projects have 100x more tests than they used to, because LLMs are great at writing tests. And my subjective experience is that the net quality has increased as a result.
YMMV, but it’s certainly a common belief, and for me at least a lived experience.
[1] https://harness.mikelyons.org
Warp had an account requirement in the beginning which spoke volumes about the misalignment of values. Now the terminal is not called a terminal, it is "the agentic development environment" (whatever that means) which also lowkey implies that it might have some kind of online features. But at the same time I understand that it is now an absolute requirement to mention AI on any web page for any product.
Does it call home?
Absolutely, in heaps. The very moment you start Warp, it sends 5 HTTP requests, (1 version check, 1 LLM model list, 3 telemetry events). The former two go to app.warp.dev, the latter 3 go to warpianwzlfqdq.dataplane.rudderstack.com and include a persistent UUID, your operating system and version, Warp version, and the tracked event name and its properties.
All of this happens before you even see the window. After you're done clicking 'No' to all the SaaS nags, you can turn off telemetry in the settings, but for some reason it gets turned back on when you restart the terminal.
While the terminal is running, it calls home whenever you trip one of the events listed in this 7000-line long file: https://github.com/warpdotdev/warp/blob/d0f045c01bacbd845a63.... Besides all the hosts listed earlier, it also makes requests to o540343.ingest.sentry.io.
All in all, a privacy nightmare.
edit: nevermind, it was quite discoverable...
I don’t think the approach of open source as a substitute for a quality program is going to last.
i currently use tmux and ghostty for my workflow
Could we have just released it? Absolutely. But I think everyone who contributed felt better knowing that what was released had one final "ready for public" review. Then our regular review process handled that going forward.
https://x.com/mitchellh/status/2049159764261925005
And the motivation was warp is doing a little bit more than a terminal.
https://con.nowledge.co
Glad to see now warp is open-sourced
tangential: I've seen Mitchel tweet that people in SF have ran up to him showing him how they fully riced their Ghostty setup. How many people here have done this and how easy/manageable is it? e.g. just forking the repo and implementing whatever Warp feature I like?
Ghostty remains incredible stable and usable and fast compared to competition.
That would be awesome!
Do you regret having this requirement in the first place?
Personal feedback: I live in a terminal 24x7 for the last 30+ years and once Warp came out I wanted to try it out immediately, but I was impressed by the requirement. So I never had a chance to try it out.
I'm actually pretty proud of the final setup I've created with it.
Each time I start to implement a new ticket, superset will pull the ticket from linear, create a worktree/workspace, reserve ports, start the servers, start a browser and start Claude with the ticket as instructions.
The cool thing with this setup is, I can have like 10x the same servers running on different ports/worktrees. Each time an agent is done, I switch to the workspace, look at the browser and can immediately test things.
It's like having 10 virtual desktops. Wonderful!
Vercel Security Checkpoint
Failed to verify your browser
Code 99
The AI integration only came years later, and they probably figured maybe they could compete in the space with Cursor, repl.it, etc. And then Claude Code came and just devoured everything, while Google bought Windsurf, Microsoft pushes Copilot, and OpenAI has Codex (while at the same time kitty and ghostty also built really nice, fast terminal emulators)
I agree I don't think they have any real shot and I certainly wouldn't recommend investing in their next round, but it's not like this was their plan all along, they went where the winds were blowing.
Nice...
... it's not about OS/2 3 or 4, but some AI thing, and apparently feels no need to disambiguate or even acknowledge the prior usage of its name
... closes tab
for zsh:
Alt-e for fish
Ctrl-g for Claude code
What you describe sorta works, but you lose things like file/dir-based autocomplete, since your editor doesn't know about your shell session.
EDIT: well looks like this is not OS/2 Warp. I wish the title would have noted this is somekind of app instead of just saying "warp".