Repository navigation
Conversation
Sportarr
force-pushed
the
add-open-webui
branch
from
October 20, 2026 03:35
816e829 to
a63c863
Compare
Contributor
Author
|
I just noticed your |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What it does
Open WebUI is a chat interface for AI models. This script wraps the TrueNAS community app
open-webuiand picks its image with Ollama bundled, so it works on its own with no other app installed.How it installs
There are no questions. The first person to open it creates the admin account, which is Open WebUI's own first-run step. Models are downloaded from inside the app.
image_selectorisollama_image, which runs Open WebUI and Ollama in one container, on CPU or NVIDIA.gpusis$GPU_CONFIG().webui_secret_keyandredis_passwordcome from$RANDOM_STRING(32).open-webui/dataand is snapshotted before updates. Ollama models go to ApplicationsCapacity atopen-webui/ollama, because they are large and can be downloaded again. The script sets no owner, because the container runs as root.$CPUS(75%, 2)and$MEMORY(50%, 4096). The memory limit covers Ollama too, so loaded models have to fit inside it.8GBRAM, since loaded models need memory. I left outGPUbecause no other script uses it, and your schema page describes it as required hardware.Testing
bun _lib/validate.tspasses with 0 errors and 0 warnings.use_all_gpusfor the Intel GPU. The bundled Ollama started and reported the limit as its memory./healthreturned 200, and/api/configreportedonboarding: true, so the first visit goes to admin account creation. Idle, it used about 900 MB.