Improve quickstart setup experience

[nsarrazin]

We would like to improve the experience of running chat-ui locally and reduce friction in the setup.

Current pain points:

• Having to setup your DB
• Having to point to an inference server
• Having to deal with either docker or git + npm

The goal is to reduce the friction with the following steps:

• Make the DB optional, use an in-process mongo server if it's not set (feat(db): use in memory db when MONGODB_URL not set #1773)
• Make it possible to download and run models directly from chat-ui (with node-llama-cpp?) feat: add a local endpoint type for inference directly from chat-ui #1778
• Allow admin login from CLI à la jupyter notebook (feat: admin CLI login #1789 )
• Add an onboarding UI for setting up new models from the app
• Provide an easy setup like npx @huggingface/chat-ui or pip
• Update the README so it focuses on quickstarting and move the rest of the doc to the docs page

[arch-btw]

Completely agree with this, I was on the verge of opening an issue asking if this would be possible, looks like you already thought of it.

For us that don't use Docker, the setup gets pretty difficult. I'm also concerned about maintaining my instance and losing data if I were to migrate all of it to a different device due to the complexity.

I like the idea of making the DB optional, maybe something like optional SQLite (which is embeddable, no separate server needed) might even simplify things further? I haven't checked to see if something like this is possible already, though.

The final thing I wanted to suggest is that maybe GitHub Actions could auto-build releases.

Excited about these changes! Looks like the first one is done already :)

[nsarrazin]

So the DB is now embedded! @arch-btw If you run the app without setting MONGODB_URL there should be a db folder that gets created where the DB data is added. If you backup that folder you should be able to recreate your data on any device.

The final thing I wanted to suggest is that maybe GitHub Actions could auto-build releases.

Curious what build output you would be interested in if you're not using docker, do you mean the output of npm run build?

[arch-btw]

@nsarrazin

Sorry for the late reply! And thank you for explaining how to backup the MongoDB instance. I didn't know that backing up that folder is all it takes to allow me to recreate the data on other devices or possibly a new installation.

As for the build output, I was indeed thinking of the output of npm run build, or maybe even being able to run chat-ui as a single binary. However, the more I look into that the more I realize that might not be realistic (I'm still going over a lot of the code and learning about typescript in general) 😅. Also, it might not really be necessary, depending on your thoughts on this of course.

Regardless, I really think this is a great ui and I use it daily. Also, thank you for your help and for making these changes happen :-)

[nsarrazin]

No worries! I think we might be able to distribute the node app as a binary in the future actually, that's a good idea 😄

[BlueskyFR]

Hey!
I just set up the chat-ui, and I know I'd have appreciated if a sample Docker compose configuration was provided in the README.
I made my own (Podman rootless compatible), so I'll paste it here in case you are interested!

services:
  # Example inference service
  ollama: [...]

  chat-ui:
    image: ghcr.io/huggingface/chat-ui
    ports:
      - 3000:3000
    environment:
      OPENAI_BASE_URL: http://ollama:11434/v1
      OPENAI_API_KEY: ollama
      MONGODB_URL: mongodb://chat-ui-db:27017
      ALLOW_INSECURE_COOKIES: true
      COOKIE_SAMESITE: lax
    volumes:
      - ./chat-ui-data:/data
    restart: always

  chat-ui-db:
    image: mongo
    ports:
      - 27017:28017
    volumes:
      - ./db-data:/data/db
    userns_mode: keep-id:uid=999,gid=999
    restart: always