t5-chat

Try it at: https://t5-chat.fly.dev/

Note: This app was NOT vibe coded. Most of the code here was written before I had access to good quality AI.

How to deploy locally:

1. Clone the repo

2. You need:

   • Docker
   • Google Oauth keys (client ID and Client Secret) for authentication
   • OpenRouter or OpenAI API key

3. Put that envs (see docker-compose.yml) in a .env file

4. Run docker compose up --build

5. Run pnpm db:push:docker to push schema to local database

How to deploy on Fly.io:

You need to deploy first a Postgres database (see fly-postgres/ dir).

Then you need to deploy the app itself (see fly.toml file)

You can use the fly.io's internal ipv6 net to connect to the database (for example DATABASE_URL=postgres://appuser:1234@t5-chat-db.internal:5432/chat).

Set POSTGRES_PASSWORD as a secret on the database app, then use that same password in the app's DATABASE_URL.

Set the corresponding envs using fly secrets set ENV_NAME=env_value command. Make sure to replace ENV_NAME and env_value with your actual environment variable names and values.

What works:

• Chatting with different models
• Using OpenAI api and or using OpenRouter
• Streaming responses
• Markdown and latex support
• Authentication with Google Oauth
• Saving conversations in the database
• PWA and offline support (readonly, you can't prompt offline)
• Generating multiple responses in parallel
• Leaving a chat and returning later doesn't stop the generation
• Reloading doesn't stop the generation
• Pseudo resumeability (when you reload, will still generate but it will be actually polling the server each second for the pending message)
• Large amount of chats (with virtual scrolling)
• Fast and responsive UI
• Working on mobile
• Preview color, emoji, title and last sent message in the chat list
• Delete chats