Warp
Warp copied to clipboard
warpdotdev
Make Warp work with Local Language Models (like Ollama models)
Discord username (optional)
No response
Describe the solution you'd like?
Due to safety concerns, many users are getting paranoid about Warp's forced login and online AI assistance. Since terminals are used to access critical data on local machines and servers, adding the ability to use local language models like Llama 2 using Ollama instead of online AI engines would be a great plus to Warp terminal emulator.
Is your feature request related to a problem? Please describe.
Mostly safety concerns when using Warp for accessing password protected systems and documents.
Additional context
No response
How important is this feature to you?
4
Warp Internal (ignore) - linear-label:39cc6478-1249-4ee7-950b-c428edfeecd1
None
Some local LLM tools even expose an OpenAI API-compatible server, which I believe could allow for rudimentary support with just a setting to change the OpenAI endpoint?
Thanks for this feature request @hmdz105!
To anyone else interested in this feature, please add a 👍 to the original post at the top to signal that you want this feature, and subscribe if you'd like to be notified.
As a warp user and system's administrator, this feature alone would make getting buy in from our CTO 100x easier. I love using Warp in my home lab / personal environments but no way in its current iteration could I even get it in my professional environment to demo a proof of value. Having the option to show case the AI integrations with local models would allow us to begin the discussion on an enterprise plan with zero data retention.
Yes, please. Ollama is easiest. LMStudio great too. Groq (not the X model) would be rather interesting. Let people use the LLM they want to use. Warp team, this INCREASES the value of this tool greatly! Your value is NOT in being a niche gateway to popular LLMs.
I think this is one of the important features missing in warp. 2 reasons - Most of the big companies dont let access gpt or any other LLM which captures data. Also, if they allow, there is only so many request that can be made for free.
Consider using the keyboard shortcut action to generate prompt based on last entered command with printed output and ask it each time in order to avoid using too much CPU/data processing usage.
This is a perfect feature for those developer who have security concerns. It's inevitable, soon enough all devices will use LLM natively just like its part of a system by default (like a RAM, CPU).
Highly suggest they would add this to the roadmap, it's easy to deploy an LLM server locally nowadays.
If they won't allow other services to connect to warp, maybe they can add this to the build natively: https://cortex.so/
So everyone is happy and not leaving warp desktop app just to run an llm server, its within the warp app itself.
This shoud have been like 2nd on your list of features to implement right after plugging OpenAI API... There is no way anybody would run such a tool as an agent over a production server over ssh You can put as much safechecks as you want, you can't prove that OpenAi or other company won't save anything that goes through the endpoint, I'm also pretty sure that everything is routed through your own enpoints which are prone to attacks as well. If I'm not trusting my paid VPN nor my company one, there is no way I'd could trust an outsourced/offshore LLM messing with critical data. Regardless of what it can actually run on its own or not.
I concur, I'm already running Ollama for handling other things, and ideally I'm already using qwen2.5-coder so just adding this to my terminal would mean I wouldn't need to pop out of my terminal to check something and I don't intend to use OpenAI if I can help it.
Agreed that this option, especially at the free level, would allow for greater adoption in corporations if anything to validate the concept especially in organizations where they run their own LLM.
This feature will make me enable AI in the app because it allows to use local AI instead of US-hosted AI(which is a huge privacy issue for me, i trust China more than US and running local is the best for me). Yes, i seen no training policy but the data can be used not only for training but for example for US government organizations to extract code parts and find vulnerabilities that can be secretly exploited by them.
Warp without this feature is a dealbreaker for me. I work with too much sensitive information to not have an option to utilize a local LLM.
Sorry, but I will never be able to trust any cloud-based AI with any kind of sensitive information (personal or professional). I've checked the network logs, and I see calls to your GraphQL API with pretty much the output of all my previous commands in it...
Having the ability to plug the terminal into my locally ran Ollama would help me alleviate my concerns. But so far, it's unfortunately a no-go for me.
This seems extremely important to users, would be nice to hear from the warp team whether or not they intend to implement? If not, we can save ourselves the time watching this thread.
Just learned about warp and have been giving it a test drive. If I could link it back to my Ollama instance, it would be a daily driver for sure!
Agree with all of the above. Amazing product otherwise, but being able to integrate with a private, local LLM would make it perfect for competent privacy minded users.
Leaving it as an 'Enterprise Only' feature is just a cash grab. Not cool.
I am a novice programmer at best and I'm trying to learn by doing. I've used the free version of Warp and ran out of tokens in less than a week. I really like Warp and it has helped me learn but since I already am paying for Github Copilot and a couple more services, I am not going to add another subscription in which the AI can only be used for one purpose. So I'm uninstalling it and just using VS Code terminal or other method where I can use my existing subscriptions or use my local llm.
It can be used to avoid using API data limits and without processing data to server.
agreed, when working with many proprietary dataset, it's not possible to take advantage of warp at all regardless of the privacy/security policies associated with warp.
Adding my voice to the din: this looks like a really cool product but this feature is a hard requirement. Years ago, I was initially turned away by the always-online account-required policy, but it looks like that has been updated. That said, it's a nonstarter for me without being able to target any OpenAI-compatible backend.
