Arch Linux User Repository

@sashank thanks for pointing this out.

You're right about the mistake, I'll fix python-pptx, review the package list, and get everything sorted.

Also I will keep hash verification.

Hey Oskar, I appreciate your taking on maintaining the build scripts. Several of the packages you list seem to be wrong and do not exist in the aur, probably because of typos. I'd recommend keeping the hash checking of the files downloaded for the build and packaging. The most important one is python-pptx instead of python-python-pptx.

@mistersmee, I appreciate the work you did. I learned a lot about how Arch is designed from your work on this. If you're still around, what greener pastures are you moving onto?

Update (2026-04-05)

The package has been cleaned up and fixed to ensure a consistent and reproducible build.

Changes:

• Bumped pkgrel to reflect packaging fixes
• Simplified PKGBUILD to build and install the upstream wheel cleanly
• Removed incorrect install script (.install)
• Added proper system configuration (/etc/conf.d/open-webui) with backup support

• Improved systemd service:

• correct ExecStart and EnvironmentFile

• added DynamicUser, StateDirectory, CacheDirectory
• basic hardening applied

• Fixed runtime paths (no writes to /usr):

• DATA_DIR=/var/lib/open-webui

• HF_HOME=/var/cache/open-webui/huggingface
• Updated dependencies and build requirements
• Regenerated .SRCINFO

The package now builds cleanly with makepkg and installs correctly.

Notes about dependencies:

• The majority of dependencies are open-source Python libraries.
• Some integrations (e.g. OpenAI, Anthropic, Azure, Google services) rely on external proprietary APIs.

• During installation, you may be asked to choose a provider for sentencepiece:

• sentencepiece → builds from source (recommended)

• sentencepiece-bin → prebuilt binary (faster, less transparent)

License note:

Upstream Open WebUI (>= 0.6.6) uses a custom license with branding requirements, so it is not strictly OSI-compliant open source.

Please report any issues if you encounter them.

Today I will update the package :)

P.S. I reported the issue and linked to the upstream bug fix on gitlab with the package maintainers.

Does anybody else have python-tensorflow installed and getting the following error: ValueError: The field annotations for StructuredTensor are invalid. Field FieldName is missing a type annotation.

Apparently the bug was fixed in November, but it has not been released yet https://github.com/tensorflow/tensorflow/pull/104620

The current version is from back in August 2025.

Anyways, thank you evorster and MarekZ for the hints/help, I actually stepped away from all this for a bit until now, so I think the bugs were fixed in that time, but I appreciate the help.

I got this running on my system without downgrading.

Ever helpful ChatGPT summarizes the fix like this:

Open WebUI 0.6.40 currently fails to start on Arch due to upstream LangChain 1.x module restructuring. The AUR package depends on the newer split python-langchain-* packages, but the code still imports old monolithic LangChain paths such as:

from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.retrievers import ContextualCompressionRetriever

These modules no longer exist in LangChain ≥1.0, causing:

ModuleNotFoundError: No module named 'langchain.text_splitter'
ModuleNotFoundError: No module named 'langchain.retrievers'

Temporary workaround (working fix)

Patch the following imports to support both old and new layouts:

In /usr/lib/python3.13/site-packages/open_webui/routers/retrieval.py:

try:
    from langchain.text_splitter import (
        RecursiveCharacterTextSplitter,
        TokenTextSplitter,
    )
except ImportError:
    from langchain_text_splitters import (
        RecursiveCharacterTextSplitter,
        TokenTextSplitter,
    )

In /usr/lib/python3.13/site-packages/open_webui/retrieval/utils.py:

try:
    from langchain.retrievers import (
        ContextualCompressionRetriever,
        EnsembleRetriever,
    )
except ImportError:
    try:
        from langchain_community.retrievers import (
            ContextualCompressionRetriever,
            EnsembleRetriever,
        )
    except ImportError:
        ContextualCompressionRetriever = None
        EnsembleRetriever = None

This allows Open WebUI to run normally with the Arch python-langchain-* packages.

Other notes

The warning about huggingface-hub>=0.34.0,<1.0 is non-fatal and does not stop the service.

Hi sashank, I hit the same issue as you did. The problem came with python-langchain 1.0, since the internal structure of the package(s) changed. I downgraded packages python-langchain, python-langchain-core, python-langchain-text-splitters and python-langchain-community to the last available version before 1.0 and everything works as expected. Hope it helps, at least for now. Cheers.

Well, whatever they changed just broke everything, lol. Can't figure out if its a database error (which I get when I run /bin/open-webui serve --port 8080 or if it's this: ModuleNotFoundError: No module named 'langchain.text_splitter' which what I get when I check the systemd logs. But I have to focus on other stuff. I just got a new AMD APU machine for this sort of stuff, and I have to work on a couple of other things before I get distracted by this again, haha. Anyways, thank you again for keeping up on all this. This has been such a great linux learning experience beyond basic shell scripts and whatnot.

P.P.S I checked your dependency_discover.py script, and I trust it more than namecap, haha.

Edit: I just saw that you replied. So yes, I agree with you.