A modern, real-time collaborative chat platform built with Effect-TS, TanStack DB and Cloudflare Durable Objects.
Hazel Chat is a full-stack chat application featuring:
- Effect-TS for type-safe, functional backend programming
- Type-safe RPC for seamless client-server communication
- Real-time sync from Durable Objects (one per organization, channel and user) into TanStack DB
- Background work (notifications, RSS, webhooks, thread naming) inside those objects
- Real-time Messaging - Channels and direct messages with instant delivery
- User Presence - Live online/offline status and typing indicators
- Message Reactions - React to messages with emoji
- Pinned Messages - Save important messages for quick access
- File Attachments - Share files with Cloudflare R2 storage
- Notifications - Configurable notification system with muting
- Team Management - Organizations with member roles and invitations
- Rich Text Editor - Plate.js powered editor with formatting
- Bot Integration - Create and manage bots with a simple API
- React 19 with TypeScript
- Vite for fast development and builds
- TanStack Router with file-based routing
- TanStack DB
- TailwindCSS v4
- React Aria Components
- Durable Object sync over WebSockets (
libs/effect-db-collection)
- Cloudflare Workers and Durable Objects, deployed and run locally with Alchemy
- Effect-TS functional programming framework
- Effect RPC for type-safe APIs
- effect-orm (
@maple-dev/effect-orm) on SQLite: Durable Object storage and D1 - Hazel auth (
@hazel/auth) for sign-in and sessions
- Turborepo for monorepo orchestration
- OXC for linting and formatting
- Vitest for testing
├── apps/
│ ├── web-foldkit/ # Web frontend
│ ├── backend/ # api Worker: HTTP API, RPC and the Durable Objects holding the data
│ ├── bot-gateway/ # Bot gateway Worker and its per-bot Durable Object
│ └── link-preview-worker/ # Link preview generation
├── packages/
│ ├── db/ # effect-orm SQLite tables, per-store table lists, migrations
│ ├── sync-do/ # Durable Object sync runtime
│ ├── domain/ # Shared types, RPC contracts, sync protocol
│ ├── backend-core/ # Reusable backend services and repositories
│ ├── schema/ # Shared schema definitions
│ └── ...
└── .context/ # Library documentation (Effect, Effect Atom, TanStack DB)
- Bun v1.2.14 or later (required for workspace catalogs)
- Docker (MinIO, the local S3 for uploads)
- Clerk account (for authentication)
-
Clone the repository
git clone <repository-url> cd app
-
Install dependencies
bun install
-
Run the setup wizard
bun run setup
This interactive CLI will:
- Start Docker services (MinIO)
- Validate your environment
- Generate secrets
- Write
.env.local(whatbun run alchemy:devreads; see.env.local.example) and the bots'.envfiles
-
Start development
bun run dev
This runs
bun run alchemy:dev: every Worker (the api with its Durable Objects and the directory D1, the bot gateway, link preview, actors) runs locally in workerd, plus the web app's dev server. Local state lives in.alchemy/.To start local bot apps, run:
bun run dev:bots
This project uses Clerk for authentication. After creating a Clerk application, grab your Publishable key and Secret key from the dashboard and provide them to bun run setup (or set CLERK_PUBLISHABLE_KEY / CLERK_SECRET_KEY in the generated .env files).
Enable Organizations in your Clerk dashboard (Organizations → Settings). Hazel uses Clerk organizations for multi-tenant workspaces, with members synced to our database via webhooks.
Add a webhook in the Clerk dashboard pointing at http://<backend-host>/webhooks/clerk (e.g. http://localhost:3003/webhooks/clerk for local dev, exposed via a tunnel). Subscribe to at least:
user.created,user.updated,user.deletedorganization.created,organization.updated,organization.deletedorganizationMembership.created,organizationMembership.updated,organizationMembership.deleted
Copy the signing secret into CLERK_WEBHOOK_SECRET.
| Command | Description |
|---|---|
bun run dev |
Start the whole stack (alchemy:dev) |
bun run dev:bots |
Start local bot apps |
bun run build |
Build all apps and packages |
bun run typecheck |
Run TypeScript checking across all packages |
bun run format |
Format and lint code with OXC |
bun run test |
Run tests in watch mode |
bun run test:once |
Run all tests once |
bun run test:coverage |
Run tests with coverage report |
The backend uses Effect-TS for:
- Dependency injection via layers and services
- Error handling with typed errors
- Resource management with scoped resources
- Concurrency with fibers and streams
Type-safe client-server communication using Effect RPC:
- Contracts defined in
packages/domain - Full type inference across the boundary
- Automatic serialization/deserialization
Each organization, channel and user is a Durable Object with its own SQLite
(infra/durable-object-sync-plan.md):
- Clients open a WebSocket per object and receive their visible rows, then live changes
- Writes return a
transactionIdthe client awaits before dropping its optimistic state - Side effects between objects go through a transactional outbox, retried by alarms
- CLAUDE.md - Detailed architecture and development guidelines
- packages/db/README.md - Database schema documentation
- .context/ - Library-specific guides for Effect, Effect Atom, and TanStack DB
See CONTRIBUTING.md for guidelines on how to contribute to this project.
See LICENSE for details.