App Submission: OpenClaw

[lukechilds]

Following on from #4557

This PR adds OpenClaw with a custom umbrelOS optimised Docker image (https://github.com/getumbrel/openclaw-umbrel) that makes OpenClaw onboarding seamless on umbrelOS. Also improves the authentication flow to have less friction and make use of the umbrelOS app proxy. Also handles correct persistence of files and installed programs between app updates.

[github-actions]

⚠️   Linting finished with 1 warning   ⚠️

Thank you for your submission! This is an automated linter that checks for common issues in pull requests to the Umbrel App Store.

Please review the linting results below and make any necessary changes to your submission.

Linting Results

Severity	File	Description
⚠️	openclaw/umbrel-app.yml	"icon" and "gallery" needs to be empty for new app submissions: The "icon" and "gallery" fields must be empty for new app submissions as it is being created by the Umbrel team.

Legend

Symbol	Description
❌	Error: This must be resolved before this PR can be merged.
⚠️	Warning: This is highly encouraged to be resolved, but is not strictly mandatory.
ℹ️	Info: This is just for your information.

[yfrbl]

Following on from #4557

This PR adds OpenClaw with a custom umbrelOS optimised Docker image (https://github.com/getumbrel/openclaw-umbrel) that makes OpenClaw onboarding seamless on umbrelOS. Also improves the authentication flow to have less friction and make use of the umbrelOS app proxy. Also handles correct persistence of files and installed programs between app updates.

hey, this is great. Just installed, but noted that it has "gpt" as provider with 4o as default (old one!). When you access the gateway, it's stuck as the openclaw config is broken with 'gpt' set as the provider. It won't allow any modifications since you can't access 'openclaw config' via the Umbrel-native docker CLI. This also affects any channel pairing, e.g. via Telegram as you need to set "openclaw pairing approve telegram '2FA-code'."

So currently, it can't be used.

[lukechilds]

Awww man, we only tested with Anthropic keys. Thanks so much for the ping @yfrbl will get this fixed and updated today. Will let you know when it's ready.

[Overtorment]

juts tested it

• by default it wants anthropic models, so when during the setup i provided openai keys it didnt work - i checked the raw config, and it tries to use unexistant model like anthropic/gpt or something like that. i had to manually edit raw config and provide correct model name:

...
      "model": {
        "primary": "openai/gpt-4o"
      },
...

• provided tg bot key during the setup, didnt work out of the box too, i had to go to config->channels and add my tg userid as allowed user. in raw format it looks like this:

...
  "channels": {
    "telegram": {
      "dmPolicy": "pairing",
      "allowFrom": [
        "6666666666
      ],
      "groupPolicy": "allowlist",
      "streamMode": "partial"
    }
  },
...

after that, it worked.

• tried to ask it to browse the web - and i think browser integration is not ok, but i havent looked into that yet (tbh idk how to debug this)

[developercrocodiles]

We need access to the OpenClaw CLI.

[lukechilds]

We need access to the OpenClaw CLI.

@developercrocodiles on it 🫡

[lukechilds]

@Overtorment ty for the info sur

by default it wants anthropic models
Yeah we messed the openai model strings up, will push a fix out shortly

provided tg bot key during the setup, didnt work out of the box too

We'll remove the TG step from the setup proxy and only use it to bootstrap LLM connection. OpenClaw is actually really good at configuring itself so you can just ask it to setup TG and if it's connected to an LLM it'll figure it all out.

tried to ask it to browse the web - and i think browser integration is not ok, but i havent looked into that yet (tbh idk how to debug this)

Was it asking for a brave api key? Sometimes it tries to use the Brave search api which needs an API key. But we do have it correctly configured out of the box to use headless Chromium. When i've been testing it with fresh throw away instances about 90% of the time it directly uses Chromium. and 10% of the time it tries to use Brave API or asks you to install a browser extension that gives it access to your tab. In those cases if you just tell it to use the installed Chromium instance it will and then it remembers it for the future.

Or did Chromium work but you just had errors? Sometimes it gets stuck at forms with captchas but it's mostly been working well for me.

[lukechilds]

@yfrbl which OpenAI models are popular with with OpenClaw? 5.2 and 5.2 Codex? Or are smaller cheaper models popular too?

[yfrbl]

hi @lukechilds, yea I think 5.2 is most powerful and best, but also really expensive bc of context size, even for small requests. You should consider adding provider support for the Kimi K2 Model, as many users will request access to it, as it's 80% cheaper.
Also, I think that CLI access is needed the most, for all things regarding setup and config (including Telegram). So the onboarding flow on Umbrel should be 1:1 as the terminal openclaw flow. There may be a solution to map it easily from the CLI flow, even as it changes through updates. If not, users should be prompted to access it via Umbrel Docker CLI.

[lukechilds]

Also, I think that CLI access is needed the most, for all things regarding setup and config (including Telegram).

For all this I've just been asking OpenClaw to set it up via chat and he's being doing a good job. Like Telegram etc. Is there anything you're away of that can't be setup this way and can only be done during onboarding? If so we can add it but afaik it can manage it all itself once you have an LLM wired up.

Also, I think that CLI access is needed the most

Yeah working on it now, should be out later today

[yfrbl]

For all this I've just been asking OpenClaw to set it up via chat and he's being doing a good job. Like Telegram etc. Is there anything you're away of that can't be setup this way and can only be done during onboarding? If so we can add it but afaik it can manage it all itself once you have an LLM wired up.

It's mostly about setting up credentials. Only workaround i found was to set up a new Skill and then add the env secrets there, so they are not parsed to LLM input.

[lukechilds]

@yfrbl re cli access are you aware we expose a shell for all apps in settings? If you go to Setting > Advanced > Terminal > App > OpenClaw you should be dropped into a shell inside the OpenClaw app container and be able to run openclaw:

[yfrbl]

@yfrbl re cli access are you aware we expose a shell for all apps in settings? If you go to Setting > Advanced > Terminal > App > OpenClaw you should be dropped into a shell inside the OpenClaw app container and be able to run openclaw:

I've used it before, but only after you send this, I've noticed that it takes like 3-5min until openclaw cli finally opens. So it's kind of stuck forever. I'm using a Pi 4 4GB. Is it device-specific?

[lukechilds]

I just tested it's also slow for me. Not minutes but about 50 seconds. This doesn't seem related to the Umbrel app. I get the same performance running openclaw installed directly on the host on the same machine.

Btw when you said "Also, I think that CLI access is needed the most." did you mean something other than accessing via settings like you can now?

[yfrbl]

Btw when you said "Also, I think that CLI access is needed the most." did you mean something other than accessing via settings like you can now?

No that was exactly what I was referring to, thx for your help!

[michel-dance]

This does not work with OpenAI keys.

[yfrbl]

@lukechilds Hope you don’t mind if I add another improvement suggestion here: accessing files right now is a mess.

Of course, you can use the CLI to preview text files or use Git for syncing, but it’s not really designed for everyday folder and file management. You can also use channels like Telegram to send files “in and out,” but a shared folder between the Umbrel host system and the OpenClaw Docker container would be a major step toward turning it into a real workspace for users.

[lukechilds]

This does not work with OpenAI keys.

@michel-dance yep sorry about that, fix going out today.

a shared folder between the Umbrel host system and the OpenClaw Docker container would be a major step toward turning it into a real workspace for users.

@yfrbl already works! Go to Files > Apps > openclaw > data

http://umbrel.local/files/Apps/openclaw/data

Here's my OpenClaw instance's workspace:

[yfrbl]

@yfrbl already works! Go to Files > Apps > openclaw > data

http://umbrel.local/files/Apps/openclaw/data

Here's my OpenClaw instance's workspace:

Awesome, thank you Luke!

[developercrocodiles]

@yfrbl Yes, I use a Raspberry Pi 5, as my desktop machine and the openclaw cli takes about 30 seconds to load, this is just the thing with NodeJS.

[lukechilds]

OpenClaw 2026.2.3 is out with the OpenAI fixes.

[maluquices]

Thanks for this

Safe to assume we'll be able to use our own self-hosted models via Ollama eventually?

[Overtorment]

are skills supposed to install? 1password skill is not installing for me.

couldnt make it use my credentials to log into websites, thinking maybe 1password skill will help

[lukechilds]

@Overtorment Yeah it should work. It should install the skill and then install the op cli via brew. Do you know which step is failing?

[Overtorment]

these lines on log look relevant:

openclaw_gateway_1    | 2026-02-09T22:47:33.681Z [ws] Loopback connection with non-local Host header. Treating it as remote. If you're behind a reverse proxy, set gateway.trustedProxies and forward X-Forwarded-For/X-Real-IP.

openclaw_gateway_1    | 2026-02-09T13:51:54.379Z [ws] ⇄ res ✗ skills.install 15437ms errorCode=UNAVAILABLE errorMessage=Install failed (exit 1): Error: Failure while executing; `/usr/bin/env PATH=/home/linuxbrew/.linuxbrew/opt/unzip/bin:/home/linuxbrew/.linuxbrew/Homebrew/Library/Homebrew/shims/shared:/usr/bin:/bin:/usr/sbin:/sbin unzip -qq -… conn=ed229a75…f984 id=bfb3c38c…f982

[lukechilds]

@Overtorment Can you ask your agent to install it? Just send it a message saying "Can you install the 1password skill and install the op cli via brew".

I haven't been using the ui at all, I just ask it to install whatever it needs itself.

[Overtorment]

surprisingly, it installed. but man, when i gave openclaw my password it chat it still could not login because every damn website throws cloudflare captcha for headless browser >.<