Anychat ( our tum.ai hackathon submission )

Anychat the first ai supportet multi-lingual group chat for project works planning and reports!

the team

• Lukas Braum
• Jorit Vásconez Gerlach
• Tim Schupp

some of the sills

Currently deployed at anychat.t1m.me

You can login with the test user: testUser1, pw: Test123! And If you want to test the peer to peer chat you can login with any of testUser<ID> for ids in 1-20.

Send a chat message and see it appear in all connected user chat with translations in all languages. Want to ask a topic spcific question just chat with the project ai @ai ...

• fully scalable production ready kubernetes deployment, that can be run completely locally in development
• micro service contain architecture
• custom django channels redis implementation for manging multiple project groups and distribting messages to connected users
• fully websocket enabled group chat
• topic specific ai assistant in any group chat, just start you message with @ai ...
• intelligent ai agent that can device when to use specic tools and is customizable with multiple tools
• agent implementation is based on langchain and support ANY LLM as base, so we can deploy our own model in the future or switch the api whenever we want
• the ai has internet access, one of the tools is that the ai can decide when to lookup something in the internet
• autmatic translation in ALL languages of a projects participants: everybody can send messages in their choosen language and still read everything!
• chat is fully file supportet send images or other attachments!
• intelligent attachment processing, automatic image content annotation and voice message to text
• loging screen and admin user management comes packages and ready
• automatic api doc and shema generation, fully typed api serializers: /redoc/
• CI ready deployment, local development on any individual component at the time.

The stack

The anychat stack Stack is a production-ready Kubernetes deployment for building dynamic web applications. It includes a Django backend and a React + Next.js frontend.

This project is uses Tim's Stack.

Prerequisites

• make is required for managing the commands.
• Docker is required for local development.

Backend

The backend is a Django application with various pre-installed packages and addons.

• Build the development backend: make backend_build
• Start the development backend: make backend_run

The server will start and be exposed at port 8000. Note that you cannot access localhost:8000/ without the Next.js frontend container. However, you can visit the admin dashboard at localhost:8000/admin/ and the API documentation at localhost:8000/api/schema/redoc/.

In development, the ./back folder is mounted into the container for hot reloading.

Frontend

The frontend uses Next.js with some pre-built components and pages. A global state is set up for all pages containing the data provided by the Django backend.

The ./front folder is mounted into the container for hot reloading during development.

• Build the frontend container: make frontend_build
• Start the frontend container: make frontend_run

The frontend can overwrite any page path except /api, /admin, /static, and /media. This allows you to configure your pages as needed.

daisyui & tailwindcss are also set up for the frontend.

To ensure that Next.js pages are hydrated with the backend data from the Django container, access the frontend through localhost:8000/ instead of localhost:3000/. Pages should be developed to work both with and without backend data.

Kubernetes

You can also build a replica for the production build or fully develop in kubernetes locally! I recommend microk8s for a quick mininmal but complete kubernetes dist.

Use make microk8s_setup to prepare the local microk8s setup.

Use make frontend_build_push and make backend_build_push to build frontend and backend container images and push then to the local registry which was initalized in the setup.

Now you can install the helm chart with make helm_install, check that all deployments and servies are running with make microk8s_status and check the running pages at local host http port 80 by vising plain localhost.

this stack was provided by Tim Schupp, (Tim Benjamin Software)
tim@timschupp.de
t1m.me
(C) `README.md`, `helm-chart/*`, `back/*`