Local and arbitrary model support

[zachlloyd]

We are trying to figure out the best way to implement local model support and I wanted to start a discussion on our different potential approaches to see what resonates most with the community.

The reason local model support is not trivial for us to implement is that our harness is split between our client (rust, open-source) and server (golang, not currently open). Moving the harness to be entirely on the client is a fair amount of work.

The options we are considering here (not mutually exclusive):

1. Port the entire harness to the client and open-source it (most work).
2. Implement Warp as an ACP client and allow folks to use other harnesses within our rich terminal UI.
3. Implement a new rust-based "lite" local harness that speaks the same protocol that our client understands but supports local models and arbitrary endpoints.
4. Route local model requests through our server and back to the client via something like ngrok (hacky, but quick).

Questions on my mind:

1. How important to users is it to use our real harness as opposed to a harness that works as an ACP server?
2. How important is it that local model requests are truly local, with no server interaction?
3. Which aspects of our UI are most important for folks wanting local model support?

[djdanielsson]

Answers to the Questions part: I am not sure I have a good answer for number 1 and 3 but 2 it's extremely important to me that local model is truly local that is likely the point of why I am using a local model to start with for that task. I haven't been using warp for a long time because it wasn't open source so I do not have many thoughts on what parts of the UI I want at this time for local models, the little I have done with like Claude cli and the information that the UI provides for that is really nice and to have something like that for local models would be cool but it might depend on what harness people are using idk.

[zachlloyd]

super helpful and tracks with what i expected

[djdanielsson]

going to the options stuff I am kinda interested in number 2 the most I think, I personally want to control my harness and the context I am feeding my agents vs using just another harness.

[FFatTiger]

No server transit for any requests

[harry-xm]

This. At my company it's a policy violation to use something like ngrok.

[officiallymarky]

Warp looks really cool, but the fact it only worked with cloud models always was a deal breaker for me. I would love to have full local AI support for not only for coding, but for the terminal agent when reacting to commands.

[phidauex]

Thanks for opening up some discussion!

I think for your options, #1 is most appealing, but yes, more work. I'd like to think it could also help your architecture long term by relying less on scaling server side components along with client components. #2 is a bit hacky but could be quick. I connect to my Hermes agent using OpenWebUI because it is nicer than the raw terminal client - I'd connect to it through Warp if it were an option, but that doesn't really support Warp being a standalone tool. #3 could be most practical because it would be fully local, not require a second tool, and for most local models, not being feature-complete would be OK. The smaller context windows mean fewer turns, fewer tools available, etc. But my use case in Warp would mostly be "uh help me remember how this command is used" not "build out an entire ansible deployment for a lab". #4 is probably not worth doing - if someone wants to use a local model, its because they want it local.

I'd rank them - 1, 3, 2, 4

For the questions:

1. I'm split because I use another agent tool already. However for new users, only having it work if you already have another harness running feels duplicative.
2. Quite important - I either have no/patchy internet at the time I'm working, or I'm doing something that I need to keep local for confidentiality reasons, and in either case, server routing breaks the whole point.
3. I don't use the most advanced tools in Warp now, so for me, the "advanced command completion" and inline planning directly in the terminal UI is what I use, and would love to use with a local model that would be fully competent at that.

[zachlloyd]

extremely helpful as we think this through

[mastertyko]

My vote is strongly for option 1.

If Warp supports local or arbitrary models, I think it should mean truly local execution with no server transit. I understand that porting the harness to the client and open sourcing it is the most work, but it seems like the right long term architecture for privacy, offline use, trust, and extensibility.

Thanks!

[crazygamerZ783]

hey got a warp fork of my own ,trying to get the ollama support work repo:https://github.com/crazygamerZ783/warp-ollama
i would appreciate a bit of help

[rozsazoltan]

The repo feels a bit strange because the original Git history is missing, and the first commit doesn't show what you actually changed.

[crazygamerZ783]

it was probably because of github glitches

[regismesquita]

ollama supports openai-compatible as far as I remember, and there are already some implementations out there supporting that.

[VicZhang6]

Honestly, I want to use DeepSeek V4 Flash inside Warp — it’s cheap, and it allows me to interact with the terminal using natural language.

[jensenojs]

same here, Although I can understand from a business logic perspective why supporting an open method of simply providing a URL + API key is not allowed, from a user demand standpoint, I think it would be more natural.

[FunkyFresh67]

Local models in Ollama or similar should be configurable as sources within Warp. Once available, you should be able to select a model during a session, either manually or by directing Warp to use it automatically based on the task or preference.

[FelixZoe]

6666

[regismesquita]

There are already a handful of "local warp server" implementations on your PR list , and on the wild forking from this repo.

People just want to be able to use a software that they really like (warp) without going through something that they don't need (your servers). we might end up with some opensource spin-off leading this if you don't just release a minimalist opensource server that simply allows people to use warp with a openai-compatible upstream.

In the future you can add something feature-rich and supporting a bunch of stuff... but for now people just want to use warp and remote models without touching someone else servers.

[apetti1920]

Model selection should be allowed to be

1. local (lm studio, ollama, etc) 2. also allowed to be configurable and tiered, small model for cmd suggestions (with bash history injection) as well as large for agent interactions

[zachlloyd]

All this feedback makes sense. We will have a proposed solution here shortly.

[bernardodsanderson]

I am mostly interested as I want to use one source of models (openrouter/GLM Coding Plan) for it.

[chukwunonsomichael189-boop]

Warp looks really cool, but the fact it only worked with cloud models always was a deal breaker for me. I would love to have full local AI support for not only for coding, but for the terminal agent when reacting to commands.

[chukwunonsomichael189-boop]

There are already a handful of "local warp server" implementations on your PR list , and on the wild forking from this repo.

People just want to be able to use a software that they really like (warp) without going through something that they don't need (your servers). we might end up with some opensource spin-off leading this if you don't just release a minimalist opensource server that simply allows people to use warp with a openai-compatible upstream.

In the future you can add something feature-rich and supporting a bunch of stuff... but for now people just want to use warp and remote models without touching someone else servers.

[Sitric1]

Adding my two cents here for what it's worth:
I think it is really necessary long term to have the harness client side and open source it. Warp seems to have a very capable UI/UX team that as long as that stays consistent I can see people, including myself, sticking around on the Warp terminal for its nice UX over any forks or competitors. Option 2 is also preferable, to those wanting to use specific harnesses but within the Warp experience.
On your questions: I'd be in the mind that local has to be fully local (no server interaction). That feels like an absolute. Can't really answer the other questions well tbh.

[Vahn84]

We are trying to figure out the best way to implement local model support and I wanted to start a discussion on our different potential approaches to see what resonates most with the community.

The reason local model support is not trivial for us to implement is that our harness is split between our client (rust, open-source) and server (golang, not currently open). Moving the harness to be entirely on the client is a fair amount of work.

The options we are considering here (not mutually exclusive):

1. Port the entire harness to the client and open-source it (most work).
2. Implement Warp as an ACP client and allow folks to use other harnesses within our rich terminal UI.
3. Implement a new rust-based "lite" local harness that speaks the same protocol that our client understands but supports local models and arbitrary endpoints.
4. Route local model requests through our server and back to the client via something like ngrok (hacky, but quick).

Questions on my mind:

1. How important to users is it to use our real harness as opposed to a harness that works as an ACP server?
2. How important is it that local model requests are truly local, with no server interaction?
3. Which aspects of our UI are most important for folks wanting local model support?

I basically wouldn't change anything about it...I like the folder navigation, the agent tool belt, the ui is clear and appealing...the fact that at each session it can auto-detect the user intent, be it a terminal command or a natural language request to the agent. Probably too many sidebars. Also local models should be truly local, but let the user decide, and keep the account signin/signup if anyone wants to tinker with what warp offers as its models. Just don't paywall features pls

[zachlloyd]

lol on too many sidebars. i don't disagree. challenging design problem

understood on the local means local point

[markhellier]

Thanks for opening this up for discussion. I wanted to share my perspective as a user, as my requirements for local LLM support are driven by privacy , Corporate policy and workflow needs.

1. On the "harness" vs. ACP client:

For my workflow, the specific protocol matters less than the capability. As long as the implementation (whether it's an ACP client or a lite harness) can seamlessly pull in the context I need to be productive, I don't have a strong preference between the two.

2. On the importance of "truly local" requests:

This is the most important point for me. In my professional environment is privacy and governance. Policies are strict: code, secrets, and sensitive data must stay local.
Because of this, I couldn't use a solution like Option 4 (routing through a server/ngrok). Even if the model itself is local, the act of routing my terminal data through an external server creates a compliance hurdle that would prevent me from using the feature in a professional setting. I am looking for a "Zero Trust" experience where I can be certain that my code and secrets never leave my machine.

3. On the most important UI aspects:

The reason I use Warp is the "rich" terminal experience. If I move to local models, I still want the AI to feel integrated into the terminal, be context-aware, be able to see my current command buffer, and capable of assisting within the flow of my work.

Privacy is my primary driver: I am looking for a way to use LLMs without compromising my data.
I don't mind paying a subscription for a tier that explicitly guarantees this local-first privacy.

[zachlloyd]

super helpful

[snowyu]

A local terminal should, first and foremost, function offline; online services should be optional. However, Warp has completely inverted this paradigm—which is precisely the primary reason I do not use it. That said, now that the project has been open-sourced, I felt compelled to offer my two cents.

So, is this so-called "terminal" essentially just a conduit for piping all local data to an AI service provider?
Furthermore, the fact that this terminal only functions within a GUI environment is absolutely laughable.

[zachlloyd]

to be clear, warp does function fully offline, no server is requried at all for the terminal functionality.

it's the ai only right now that requires a server, which i think makes sense in general, but not for local model access.

[geert2705]

Voting option 2 (Warp as ACP client) as the immediate unlock, with option 3 (a lite local Rust harness) as a longer-term parallel track. Option 4 (ngrok-style routing) doesn't satisfy the local-model use case at all in our experience.

Why option 2 first:

• The harnesses people most want to drive local models with are already ACP-shaped: vibe-acp (Mistral Vibe, Apache 2.0), Gemini CLI, OpenCode, Claude Code via adapter. ACP-client gets all of them at once.
• It satisfies "truly local, no server transit" (Q2) for any user whose chosen harness is local — the data never leaves the harness↔Warp pipe.
• It does not require porting the proprietary server-side harness, so it stays scoped to a quarter (and matches the May–June 2026 commit in Roadmap: May-June 2026 #9233).
• It's cleanly separable from the Warp-as-ACP-server direction (Support for Agent Client Protocol (ACP) #7326 also covers that), which can follow later without redoing client-side work.

Answers to your three questions, from a small-team-using-Warp-daily perspective:

1. Real Warp harness vs ACP-attached harness? For local-model work, ACP-attached is fine — preferable, even. We want Warp's UX (blocks, command suggestions, attachments, tab/pane model, session history). The agent harness behind it is interchangeable.
2. Truly local? Hard requirement. Confidentiality and offline cases are the whole reason for running local models; any server hop is a non-starter (and a policy violation in some environments).
3. Which UI aspects matter most? Block-structured output, attachment/buffer model, tab+pane workspace, command suggestion / inline planning, session history, settings & keybindings parity. Diff/review surfaces too if they can be wired through session/update.

Concrete grounding: we're targeting ~128GB unified-memory boxes (Jetson Thor / Apple silicon) running Devstral Small 2 (24B, Q8) and Qwen3-Coder-30B-A3B locally via vibe-acp. The minute Warp is an ACP client, that pipeline is real, fully local, and doesn't touch any server.

[brooksc]

Thank you for being open to this and engaging the community.

I think you've overwhelming heard from others that they want option 1. I'd say the same.

If you implement this, I'll absolutely return to warp. I'm looking forward to it!

[chukwunonsomichael189-boop]

Answers to the Questions part: I am not sure I have a good answer for number 1 and 3 but 2 it's extremely important to me that local model is truly local that is likely the point of why I am using a local model to start with for that task. I haven't been using warp for a long time because it wasn't open source so I do not have many thoughts on what parts of the UI I want at this time for local models, the little I have done with like Claude cli and the information that the UI provides for that is really nice and to have something like that for local models would be cool but it might depend on what harness people are using idk.

[EdenBendheim]

spam ^

[EdenBendheim]

I agree with the consensus that #1 is of course the best option, and I would like to point out that I am most interested in warp in its capabilities as a harness.

Like most programmers, I still do use cloud models for the majority of my tasks, and the ability to use my OpenAI Codex sub while still taking advantage of the harness (like Opencode for example) would push me to use warp more.

If this can be achieved in a lightweight manner, option #3 is the clear choice. If not then #1

[zachlloyd]

Updates on our plan here...

With respect to BYOK and arbitrary endpoint support:

1. In the next 1-2 weeks we plan on making BYOK available on our free plan and extending it to support arbitrary OAI compatible endpoints. For now this will still go through our server harness, as it’s a much bigger lift to fully move that harness to the client in a short timeframe.
2. BYOK and arbitrary endpoint support will be available for individuals and companies of up to 10 employees. It will be login-gated, but free to use within that scope. For larger businesses and enterprise customers, they need to go through Business / Enterprise because they are using Warp’s managed harness as part of a broader platform offering, and that usage is metered through platform credits.

With respect to local model support, our plan is to

1. Work on a client side harness that connects directly to local models. We are currently slotting this into our priority list, but I expect work to start within the next few weeks.
2. Implement Warp as an ACP client and allow folks to use other harnesses within our rich terminal UI. This addresses the need to BYO inference without connecting to Warp’s servers. See Support for Agent Client Protocol (ACP) #7326
3. These will not require login and will be completely client side.

Thanks to everyone who has weighed in on this thread.

[dagmfactory]

As discussed above, arbitrary endpoint support is coming soon! 🎉
Warp will support any OpenAI compatible endpoint with your own API key.
Here is a quick demo of what's to come
https://www.loom.com/share/0ba7a43bc1df4d44a73081f562d24d3d

[SergioNR]

@dagmfactory this looks great! However, I noticed the video only showcases for /agent mode - will it also work with autocomplete suggestions, or suggested prompts (eg: stop all docker containers)?

thank you!

[xbmc4lyfe]

Was just trying to make a build of the code with this exact functionality. Any way we can get a link to the source and make a custom build?

[tao12345666333]

Thank you! : ❤️

[Cjava08]

When is available?

[MichaelSL]

Main obstacle for using Warp at our company is the privacy: we have the strong governance model and we can't use 3rd party LLM providers.

We would use BYOK and point it to the private instance hosted in Azure, AWS etc. Buying a license is not an issue: we just need to make sure that data never goes to any of Warp servers, which probably makes billing a bit challenging for Warp.

We tried other solutions with BYOK, but none of them work with agents as good as Warp.

[zcg]

https://openwarp.zerx.dev/ https://github.com/zerx-lab/warp/releases/tag/v2026.05.06.preview Hello, I noticed this open-source project has been modified and it's already usable. I've downloaded it, installed it locally, and used it. It's really impressive and works great. If you can't wait for the official release, you can use this one for now.

[regismesquita]

@zcg btw I saw openwarp and the problem with it is that they heavily modified the client, making it a hard fork, it will be really hard for them to merge upstream changes.

I made one that follows the suggestion #3 implementing a fake server that allows for external models (that way I can avoid touching actual warp code as much as possible and therefore making it easier to merge upstream changes)

And I remember seeing two or three other PRs here in the repo with some other implementations.

Hopefully my project will be superseded by the

Work on a client side harness that connects directly to local models. We are currently slotting this into our priority list, but I expect work to start within the next few weeks.

until there at least I get to use warp fully local and with my internally hosted models.

[zcg]

@zcg btw I saw openwarp and the problem with it is that they heavily modified the client, making it a hard fork, it will be really hard for them to merge upstream changes.

I made one that follows the suggestion #3 implementing a fake server that allows for external models (that way I can avoid touching actual warp code as much as possible and therefore making it easier to merge upstream changes)

And I remember seeing two or three other PRs here in the repo with some other implementations.

Hopefully my project will be superseded by the

Work on a client side harness that connects directly to local models. We are currently slotting this into our priority list, but I expect work to start within the next few weeks.

until there at least I get to use warp fully local and with my internally hosted models.

you are right! we cant wait so try

[musaabhasan]

From an enterprise/privacy perspective, the important distinction is not only "local model support" but "local execution boundary." If prompts, repository context, tool traces, or intermediate reasoning still route through a remote harness, many regulated users will not be able to use it even if the final model endpoint is self-hosted.

I would rank the options this way:

• Short term: arbitrary OpenAI-compatible endpoint support is useful, but should be labeled clearly if the harness remains server-side.
• Medium term: ACP client support can unlock local harnesses without Warp needing to own every provider integration.
• Long term: client-side/open harness is the strongest trust model because users can verify where code, prompts, and tool outputs flow.

A good product surface would show an "effective data path" for each provider: what stays local, what goes to Warp, what goes to the model endpoint, and what is logged. That is the question security teams will ask before approving local/private model workflows.

[AkikoOrenji]

Came here as heard Warp now supporting Windows. Installed and then immediately uninstalled after realising the product is effectively useless without sign up and sending data to yet another provider. With most workstation level laptops now coming with a dedicated NPU or GPU with a few gig of VRAM (even shared RAM is OK for lower end qwen models) people may as well make use of to make terminal life easier. Personally just want an AI shell for system management and basic automation:

• Handling of remote sessions including auto switching from Windows powershell to whatever shell the remote system has. Can't install anything on the remote end so the ability for the harness to recognise the change in shell and keep operating.
• Ability for AI to log into remote system and carry out tasks at direction with strong granular guardrails (regex).
• Password hand-off back to user on logon for password input (either via SSH key password or remote user password)
• Automatic detection of password prompts in remote session for hand-off e.g. sudo, keystores, secondary SSH connections.
• The diff tool would be good for other tasks such as comparing configurations before and after changes.
• Basically any use case where remote desktop is not available but complex operations need to be carried out e.g. modifying registry keys
• generation of quick single shell one liners to manipulate and write data. what’s that sed or awk flag i needed.
• generate complex powershell without consulting reams of documentation.

Not interested in coding capabilities as use other tools for that.

1 and 3 are the better options. Given there are already forks in the wild why not encourage them to PR (if they haven't already)

[smthpickboy]

It's frustrating how big companies constantly try to seize control of your computer and data. LLMs and their harnesses should act as assistants to the terminal, not as supervisors. If Warp continues with its closed mindset, open alternatives like OpenWarp or other terminal+LLM apps will thrive and take its place.

[mz135135]

+1

[Mikewhodat]

If I can contribute in any way whatsoever whether it's bug bounty for your project or contributing code. Please let me know I'd be highly interested. I have a personal vendetta against warp. I was just thinking along the lines of instead of starting. My own repository in starting this whole task, out from scratch, I would join the community.

This is something that I did not do with deep seek

[compgeniuses]

now that's music to the ears,Also add ability to fetch model, for /models endpoint, as well as context windows, and whether model has vision or not.