Dust

Dust is a free, open-source task-management app, hosted at https://dust.luketurner.org. See the manual for more information.

Development

Dust uses Next.js and Prisma with a Postgres database.

Development dependencies:

• Node 16+
• Postgres 15+ w/pgvector extension (or podman to run containerized postgres locally)

# One time setup
npm install
npm run db:build:dev
npm run db:setup:dev
npm run db:migrate

# Launch the app in dev
npm run db:start:dev
npm run dev

# run after schema updates
npm run db:migrate

Deployment

Dust is deployed at Fly.io. Deploy with:

# first time -- run this
flyctl launch

# then afterwards run this
flyctl deploy

Required secrets:

NEXTAUTH_SECRET
AUTH_GITHUB_ID
AUTH_GITHUB_SECRET
DUST_SSH_KEY_PASSPHRASE

Recommended scale for self-hosting:

flyctl scale count app=1 cron=1

Postgres

Dust uses a custom version of the postgres-ha app. The vendored version is based on commit 44ab26a7cd0c3f8345887e9e7b9f48a73851eaf5.

Changes from the base app:

• Add pgvector extension.

To deploy:

cd postgres
flyctl deploy

Once you do this, the flyctl postgres commands won't work anymore, unfortunately.

To connect to the DB, use:

cd postgres
flyctl proxy 5432

To get the password, use:

cd postgres
flyctl ssh console

# get password
echo $OPERATOR_PASSWORD

To get started, need to do the following queries (only need to do this once):

CREATE DATABASE lt_dust;

LLM Server

To run a tiny LLM server locally:

cd local
wget https://huggingface.co/jartine/phi-2-llamafile/resolve/main/phi-2.Q2_K.llamafile
./phi-2.Q2_K.llamafile --embedding --nobrowser

And put the following in your .env:

DUST_LLM_SERVER="http://127.0.0.1:8080"
DUST_LLM_VERSION="phi-2.Q2_K"
DUST_LLM_EMBEDDING_VERSION="phi-2.Q2_K"

Quote loading

Dust provides daily motivational quotes. Quotes are stored in a Quote table in the database. They have to be manually loaded in -- Dust doesn't have UI for creating/editing quotes.

There is a scripts/load_quotes.mjs file that can convert an easily-authored YAML file containing quotations into a series of SQL statements that can be pasted into psql to load quotes.

The script expects a file called local/quotes.yml with content like:

quotes:
  AUTHOR:
    BOOK1:
      - Quote1
      - Quote2
    BOOK2:
      - Quote3
      - Quote4

And will produce SQL INSERT statements, one for each quote, in local/quotes.sql.

AI Features and Notes

I've been experimenting with adding some optional AI features to Dust.

IMPORTANT: Currently, it's not possible for users to opt into AI features without deploying their own Dust instance.

AI features:

• Calculate embeddings for tasks and use them to find similar ones