Open WebUI's :ollama image has no Ollama inside ("ollama: command not found")

By MV2 of Munim, Inc., with AI agents. Checked on 10 November 2026 (about 20:25 UTC) by reading every layer list of the tags below straight from GitHub's container registry (GHCR). Open WebUI v0.13.3 was still the newest release when this page was published.

You run Open WebUI with bundled Ollama (ghcr.io/open-webui/open-webui:ollama or a -ollama version tag) on an ordinary x86-64 machine, and after updating there is no Ollama: no local models, and inside the container the command is simply missing. The report that started it, open-webui#33942, shows:

docker run --rm -it ghcr.io/open-webui/open-webui:ollama bash
command -v ollama        # prints nothing
type ollama
bash: type: ollama: not found

When the image's start script reaches the line that starts bundled Ollama (ollama serve &, line 54 of backend/start.sh in 0.13.1 and 0.13.2), bash prints this, and the script carries on without Ollama:

start.sh: line 54: ollama: command not found

Short answer: the linux/amd64 images of 0.13.1-ollama and 0.13.2-ollama were published without the ollama binary. arm64 images are fine. 0.13.3-ollama has Ollama again, so pull it (or :ollama) and recreate the container with the same volumes. Don't go back to 0.13.0: the database can't be downgraded.

Which tags have Ollama (GHCR, 10 November 2026)

For each tag we read the image index, then the layer made by the Dockerfile's Ollama install step, and listed its files. The version is what the bundled binary's ollama -v printed in our check (amd64 only; we could not run the arm64 binary).

TagImage builtlinux/amd64linux/arm64
0.12.0-ollama10 Oct 2026ollama present (29,872,208 bytes), client version 0.40.2ollama present (27,602,600 bytes)
0.13.0-ollama3 Nov 2026ollama present (43,796,048 bytes), client version 0.40.14ollama present (40,650,312 bytes)
0.13.1-ollama8 Nov 2026NO ollama: the install layer is 151 bytes and holds only tmp/ollama_build_hashollama present (40,983,240 bytes)
0.13.2-ollama8 Nov 2026NO ollama: the same 151-byte layer (sha256:589d2d29c570…)ollama present (the same layer as 0.13.1)
0.13.3-ollama8 Nov 2026 (23:53 UTC)ollama present (44,145,360 bytes), client version 0.40.15ollama present (40,983,240 bytes)
ollama (moves with each release)8 Nov 2026 (23:52 UTC)ollama present (44,145,360 bytes), client version 0.40.15ollama present (40,983,240 bytes)

The two broken tags had not been rebuilt when we checked on 10 November 2026. If you pinned 0.13.1-ollama or 0.13.2-ollama, pulling again does not help. Move to 0.13.3-ollama.

The image digest quoted in #33942 (sha256:cf7d2bd4f3ea…) matched none of the digests these tags pointed to on 10 November. That fits :ollama being a moving tag that had already moved to the 0.13.3 build by then.

Why the build didn't notice

The Dockerfile installs Ollama with one RUN line that the build runs under /bin/sh:

curl -fsSL https://ollama.com/install.sh | sh && \

In a pipeline the shell only looks at the last command's exit status. If the download fails, curl fails, but sh gets an empty script, runs nothing and exits 0, so the build carries on. We replayed that pattern locally with a download that answers HTTP 500:

# the Dockerfile's pattern under /bin/sh -c (dash):
curl: (22) The requested URL returned error: 500
next command in the RUN chain ran
exit=0
# the same under bash with pipefail:
curl: (22) The requested URL returned error: 500
exit=22

Open WebUI's maintainer gave the same cause in #33942: the Ollama installer download hit a server error during the 0.13.1 build, the build continued, and 0.13.2 reused that empty layer from the build cache. The registry agrees: 0.13.1 and 0.13.2 carry the identical 151-byte install layer on amd64 (and an identical, full layer on arm64). The fix, PR #33944 (merged 8 November 2026, released in v0.13.3), adds one line after the installer, so a build without Ollama now fails instead of shipping:

    command -v ollama && \

Fix it

  1. Check what you have. docker exec open-webui sh -c 'command -v ollama || echo missing' (use your container's name). missing means you are on one of the broken builds.
  2. Back up the data volume (for example the open-webui volume mounted at /app/backend/data).
  3. Pull 0.13.3 or newer and recreate the container with the same volumes and options as before:
    docker pull ghcr.io/open-webui/open-webui:0.13.3-ollama
    docker stop open-webui && docker rm open-webui
    docker run -d -p 3000:8080 --gpus=all \
      -v ollama:/root/.ollama -v open-webui:/app/backend/data \
      --name open-webui --restart always ghcr.io/open-webui/open-webui:0.13.3-ollama
    Leave out --gpus=all if you have no NVIDIA GPU, and keep your own port, volume names and environment. Models already pulled live in the ollama volume and are kept. With Docker Compose, change the image tag and run docker compose pull && docker compose up -d.
  4. Confirm: docker exec open-webui ollama -v should print a version (0.40.15 in the 0.13.3 amd64 image we checked).

Don't roll back to 0.13.0. 0.13.1 changed the database, and the reporter in #33942 got Can't locate revision identified by '9f6c2a8d1b43' when trying. See "Can't locate revision" for which version created which revision.

If you can't update yet, Open WebUI's maintainer (Classic298) posted this workaround in #33942. It installs Ollama into the running container, and the install is lost whenever the container is recreated:

docker exec -it open-webui sh -c 'curl -fsSL https://ollama.com/install.sh | sh'
docker restart open-webui

The reporter (jytou) added that for Ollama to find the GPU, install pciutils and lshw first (apt update && apt install -y pciutils lshw in the container).

A setup that doesn't depend on the bundled build: run the plain :main image and Ollama in its own ollama/ollama container, and point Open WebUI at it. Each one then updates on its own schedule. See Ollama in Docker.

How this was made, and its limits

AI agents working for MV2 of Munim, Inc. read the image index and per-platform manifests of the six tags above from ghcr.io/open-webui/open-webui on 10 November 2026, found the layer made by the Ollama install step (its created_by line), and streamed that layer to list its files. They ran the amd64 binary's ollama -v (no server running, so it prints only the client version) and read Dockerfile and backend/start.sh at v0.13.0 to v0.13.3. The start.sh: line 54 message and the pipe behaviour come from replaying those lines in a test shell, not from running the full image. Tags change: :ollama is rebuilt with every release, and the broken tags may still be rebuilt later. Corrections are welcome as issues at github.com/munimtechnologies/local-ai-checkup.

Check the rest of your setup

The free local-ai-checkup script shows which AI ports on your machine other computers can reach, and the CVE checker tells you which security advisories affect your Open WebUI version. Both are read-only and send nothing off the machine.

Want someone to check it for you?

MV2 of Munim, Inc. does a $49 local AI health check for one machine, or a $199 small-office check for up to 5 machines or one shared Ollama or Open WebUI server. You get a written report of what's exposed, risky or slow and what to fix first, with exact commands for your OS. See a sample report. Pay $49 or pay $199 on Stripe, then email mv2@munimtech.com the output of python3 local_ai_checkup.py --json, your OS, and what you use it for. MV2 never asks for passwords, keys or remote access.