Skip to content

Latest commit

 

History

History
183 lines (129 loc) · 7.65 KB

README.md

File metadata and controls

183 lines (129 loc) · 7.65 KB
Raw

👍 Congusto Chat

Vision: To build the best open-source AI Chat application for developers and power users.

Features:

  • Any model from any API provider
  • Multi-modal
    • Text
    • Images
    • Audio
    • Video
  • Artifacts
  • Code Interpreter
    • One-off calculation
    • Jupyter notebook session
  • Esay deployment
    • Postgres + Vercel (or any other host with node support)
    • Local deployment with SQLite an no auth.
  • MIT license allows for commercial use and modification.

Milestones:

  • GitHub milestone details
  • GitHub milestone details
  • GitHub milestone details
  • GitHub milestone details

Hosted Version

  • Go to Congusto Chat and create an account.
  • Go to Settings/Providers, add your API Key to a default provider, or configure a new provider and models.
  • Go to Settings/Assistants, create a new assistant that combines the model, API Key, and your system prompt.
  • Start chatting.

Deploy your own

When deploying Congusto Chat, you will need:

Postgres database.

Some options (not sponsored):

  • Neon - Free tier available. Has a nice UI for managing the database.
  • Supabase - Free tier available. Has a nice UI for managing the database.
    • Make sure you get the "Session mode" connection string:
    • Go to https://supabase.com/dashboard/project/[your project]/settings/database
      • Select [x] Display connection pooler: Mode: Session and copy the Database URL.
    • We only want the database, not the Auth/API functionality.
  • Deploy on Railway - Free tier available. The UI is minimal, you might want to use pgAdmin or similar to manage the database.
  • Deploy on any vps - Digital Ocean, AWS, etc. - You will need to manage the database yourself.

You will need the set the DATABASE_URL env variabled to the Postgres connection string, that should look like this:

DATABASE_URL=postgres://user:password@host:port/database

TODO: Local deployment with SQLite and no Auth.

OAuth Authentication with Google and/or Github

Github

Github is the easiest to set up.

  • Go to Github Developer Settings/OAuth Apps and create a new OAuth App.

    • Set the Homepage URL to your domain.
    • Set the Authorization callback URL to https://yourdomain.com/login/github

  • Github does not allow multiple callback URLs, so if you want to both deploy publicly and do local development, you will need to create an extra OAuth App for local development.

    • Set the Authorization callback URL to http://localhost:5173/login/github

Set the following environment variables:

GITHUB_CLIENT_ID=your client id
GITHUB_CLIENT_SECRET=your client secret

Make sure to set the correct one for prod/dev. The redirect will fail if the redirect URL does not match the one in the OAuth App.

If you don't want Github auth, you can disable the login button in the UI by setting the following environment variable:

PUBLIC_DISABLE_GITHUB_LOGIN=true

Google

To set up OAuth authentication with Google:

  • Go to the Google Cloud Console.
  • Create a new project or select an existing one.
  • Navigate to "APIs & Services" > "Credentials".
  • Click "Create Credentials" and select "OAuth client ID".
  • Choose "Web application" as the application type.
  • Set the "Authorized JavaScript origins" to your domain (e.g., https://yourdomain.com).
  • For local development, add http://localhost:5173 as an authorized JavaScript origin and http://localhost:5173/login/google as a redirect URI.

After creating the OAuth client, you'll receive a client ID and client secret. Set these as environment variables:

GOOGLE_CLIENT_ID=your client id
GOOGLE_CLIENT_SECRET=your client secret

If you don't want Google auth, you can disable the login button in the UI by setting the following environment variable:

PUBLIC_DISABLE_GOOGLE_LOGIN=true

Email login

Coming soon.

Local development

  • Clone the repository: git clone https://github.com/ConGustoAI/chat

  • Copy .env.example to .env and set the environment variables.

  • Make sure you have NodeJS v20+ installed.

  • Install the dependencies

    • npm i

  • Populate .env with your Postgres connection string, Google and/or Github OAuth credentials.

  • Run migrations (populate the table structure) and seed the database with default data.

    • npm db:migrate
    • npm db:seed

  • Start the dev server

    • npm dev

  • Open the browser at http://localhost:5173 and sign up as a new user.

Note: Run migrations when you make changes to the database schema (src/lib/db/schema/*.ts):

  • npm db:generate to update the migrations.
  • npm db:migrate to apply the migrations. This will sync the database schema with the code.

If you update the default providers/models/assistants, don't forget to reflect this in src/scripts/seed.ts

Deployment on Railway.app (Recommended)

  • Fork the repository on Github.
  • Create a new project, "Deploy from GitHub" and point it to our repository.
  • Set the envorpnment variables in the settings.
  • Set the domain name to your domain, make sure it matches the auth callback URL for Google/Github.

@xl0 You can also deploy Postgres on Railway, but the management UI is minimal. I prefer Supabase or Neon if I have to touch the database.

Deploy to Vercel (Not recommended)

The issue with Vercel is, instead of the normal long-lived node runtime, they wrap your code into "server functions" that are short-lived, and then charge per millisecond of their runtime. On the hobby plan, the default runtime cap is 10 seconds that can be extended to 60. When we execute a chat request, the backend function runs for the time the message is being streamed from the LLM provider. This typically takes a few seconds, and can easily be minutes for long message from a slow provider. This will result in a timeous and/or an expensive deployment.

Still, if you choose to deploy to Vercel, you can:

  • In svelte.config.js, set the adapter to vercel (uncomment vercel, comment out node).

  • In settings, use the following commands:

    • Build command: npm run deploy - this will build the app and run migrations/seed if needed.
    • Install command: npm i
    • Development command: npm dev

  • Node version: 20

  • Runtime limits: 60 seconds on Hoby plan, 300 seconds on Pro plan.

  • Set the environment variables.

  • Set the domain name to your domain, and make sure the auth callback URL for Google/Github matches the domain.

Admin users

Congusto Chat keeps the providers (OpenAI, Anthropic, etc.), models and assistants in the database. It comes with a set of default ones, and the users are able to define their own. The default ones are seeded in the database when you run npm run db:seed.

To edit the default providers, models, and assistants through the UI, the user needs to be an Admin. To make a user an Admin, set the is_admin field to true for that user in the public.users table, you will see the Admin entries in the Settings page.

Development

For backend logs, set DEBUG=app:* to see all logs. You can also set it a subset of the logs, e.g., DEBUG=app:db* to see only the database logs. For frontend logs, set debug to * in the browser local storage. See the debug package for more information.

Sponsors:

Possibly, you: info@congusto.ai