[Other] Opinion: Strata app should be separate repo from Strata Engine

I understand the demand for yet another harness app. But it is better for the app to be a separate repo that bundles Strata engine instead.

Combining the app and engine increases codebase complexity. Focus is shifted from model optimizations and stability to the app itself. And there will be a new flood of issues related to the app, leaving the engine worse off. Especially given the engine was receiving too many issues & PRs by itself already.

I suggest while still possible to "walk back" the app release, and redo it as a separate thing from the engine.

Anyone else agrees?

Version: v0.1.45
Hardware: RTX 4090 24GB, Ryzen 5950X, DDR4 64GB, Linux v7.2.9

(It is also my opinion you should enable the Discussions tab and why does [Other] issues need version & hardware info?)

Thanks for the issue. To keep the tracker readable, every item needs an area prefix in the title and the parts of the issue form filled in. This one is missing:

Add the Strata version (Settings > About, or strata --version).
Add your graphics card with its VRAM, your RAM and your operating system.

Edit the issue (title and description) and this note goes away on its own. If it is still missing parts in 5 days it will be closed, which is not a rejection: fix it and comment /reopen.

Details: CONTRIBUTING.md

Thanks, this follows the format now.

/reopen

The Strata Engine (v1.45) is awesome!

But I do not need anything more than a drop-in replacement of llama-cpp-server running in a container (I'll happily take the light-weight web dashboard, too, which is nice for testing / monitoring).

Best follow the old and proven Unix philosophy: do one thing right.
Additional functionality, like the app only a part of the users want to use, is best in a separate repo.

I used to keep an eye on performance-improvement PRs almost every day. Today I saw a flood of Harness-related PRs show up, and it honestly felt discouraging—I just couldn’t muster any enthusiasm. I can already see a wave of GUI PRs coming into this project. Strata is certainly getting a lot of attention, and I agree that a visible app may well have more commercial value. If the author’s goal is to use Strata as the entry point for Harness, then I think it’s reasonable to consider splitting the Engine out separately.

In my opinion, Strata should concentrate on improving the engine - this is its current strength.
Focus on further performance improvements, broader hardware support and, especially, support multiple LLMs (with similar MoE architecture) like GLM Flash, DeepSeek Flash, etc.

It would be very difficult for Strata to compete with other, more established, harness products. And if this is where the devs put their main resources, Strata would loose its edge quickly on the engine part.

+1 to keeping the engine focused. Speaking as someone who accidentally ended up on the "app" side of this 😅 — I've been building Strata Manager, a separate MIT desktop manager that just drives the Strata engine from the outside (install/update the engine, pick models, tune configs, run it as a tray service):
https://github.com/hongfaqiu/strata-manager

It just added support for the 0.2.0 engine packages. Building it as a separate project has worked fine so far, so I think a split is very doable: the engine stays a clean inference server with a stable CLI/API, and GUIs (official or community) live on top of it.

Manager deliberately doesn't ship its own agent/harness, so you can keep using OpenCode, Hermes or whatever you prefer.

That said, the official app clearly lowers the barrier for a lot of new users, so I get why Niko shipped it. Maybe a separate repo + a documented engine interface would give us both.

At this point harness UI's are dime a dozen, and reinventing the wheel for the nth time is just a waste of resources.
The only GUI needed would be for managing the strata engine.

I agree with this suggestion.

First of all, I really appreciate the work being done on Strata. The new desktop app is an exciting addition, and I can see how it could make Strata more accessible to a wider audience.

However, as a Strata 0.1.45 user, my primary interest is still the inference engine itself — especially performance, stability, reliability, and efficient hardware utilization.

My concern is that maintaining a full desktop application within the same project could gradually divert development resources away from improving the core inference engine.

I would personally prefer to see Strata remain a focused, lightweight, high-performance inference engine, with the desktop app maintained as a separate project that uses the engine.

This would allow both projects to evolve independently, while keeping the core engine as stable and efficient as possible.

I'm not against the desktop app at all. I simply believe that a clear separation between the engine and the application would be healthier for Strata's long-term development.

Thanks to the developers for all their hard work. I genuinely hope Strata continues to improve! ❤️

+1 absolutely, please

+1

We already have a bunch of harnesses and we don't need another slop one.
Strata should focus on it's core identity and what makes it good.

+1

The current state of Strata is even more complex than ever, for new users and those supporting it pre-2.0 . For instance, failing to setup new models via strata launch result in instructions to retry via START_HERE.bat. Then why do I even bother with the strata-based terminal app?!

It's adding unnecessary complication, going against the initial reason why so many of us gave Strata a go back then -- the ease of running Qwen3.8-Flash-Next easily on our hardware that was not supported well.

At this rate, it would not be long before this gets replaced by another comparable project or fork that just focuses on ease of use and performance. That will be a pity.

For starters, could've spent the effort to improve the docs, and list the different options available. Right now, it is almost unusable.

Harness / app is pretty much useless, most of the users have their own preferred ones, either from frontier AI companies or one of the popular ones (e.g.) Deepseek Harness, Hermes Agent, OpenClaw, etc.

+1

+1
This project has been amazing and being able to achieve 120+tok/s locally has been a dream come true. It's only the beginning and I think the efforts are best spent towards the engine itself and the docs, without complications.

+1

I understand the draw to bring in new users, users with "e-waste adjacent" hardware (in terms of AI at least) are not likely to be the most technical, but keeping everything combined to a harness is going to weaken the engine development and create another exploit vector. Just look at Hermes Agent and how many PRs are related to patching security issues, it's much better to have harness devs just work on a harness, and there's plenty of them already. Strata Engine is novel and needed, the focus deserves to be there.

+1

+1 absolutely

Navigation Menu

Repository navigation

[Other] Opinion: Strata app should be separate repo from Strata Engine #3330

1 remaining item

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

[Other] Opinion: Strata app should be separate repo from Strata Engine #3330

Description

Activity

github-actions commented on Oct 20, 2026

github-actions commented on Oct 20, 2026

Interpause commented on Oct 20, 2026

BenjaminLungwitz commented on Oct 20, 2026

plus1998 commented on Oct 20, 2026

BenjaminLungwitz commented on Oct 20, 2026

hongfaqiu commented on Oct 20, 2026

JohnArneBirkeland commented on Oct 20, 2026

qwqqwqqqw commented on Oct 20, 2026

4f1sh3r commented on Oct 20, 2026

SchizoDev commented on Oct 20, 2026

EstatorMator commented on Oct 20, 2026

ppc64el commented on Oct 20, 2026

1 remaining item

xandertreat commented on Oct 20, 2026

0xDEADC0DE55 commented on Oct 20, 2026

jacob8919 commented on Oct 20, 2026

StoicaTielemans commented on Oct 20, 2026

HazyBruh120 commented on Oct 20, 2026

Silenttttttt commented on Oct 20, 2026

Alper2022 commented on Oct 20, 2026

The-CEO-of-internet commented on Oct 20, 2026

Luigi-Pizzolito commented on Oct 20, 2026

stegon98 commented on Oct 20, 2026

punal100 commented on Oct 20, 2026

thecannawiki commented on Oct 20, 2026

Societus commented on Oct 20, 2026

Hotcooler commented on Oct 20, 2026

samolukadjo commented on Oct 20, 2026

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions