Redis document agent

This app is a Bun + Express document agent that uses Redis for:

• Session storage
• Short-term chat history
• Long-term, episodic, and semantic memory
• JSON document storage
• Vector search over document chunks
• Redis Streams-backed logging

By default, the app runs in a deterministic local demo mode that loads bundled markdown docs from data/documents into Redis. You can optionally switch to live crawl mode with Tavily.

Requirements

• Bun
• Docker for the bundled Redis service
• One LLM provider key
  • Recommended: OpenAI
  • Optional: Google Vertex AI or Anthropic

Quick start

1. Create a local env file:

cp .env.example .env

2. Install dependencies:

bun install

3. Start Redis in Docker:

bun run docker:redis

4. Start the app:

bun run dev

Open http://localhost:8080.

Demo modes

Local demo mode

This is the default mode.

• CRAWL_SOURCE=local
• The app loads bundled docs from data/documents
• No Tavily key is required
• Best fit for the tutorial flow and local testing

Live crawl mode

Switch to Tavily-backed crawl when you want live document ingestion.

CRAWL_SOURCE=tavily
TAVILY_API_KEY=...

In live crawl mode, the app uses the URL and crawl instructions extracted from the project prompt.

Scripts

bun run dev         # app + CSS watcher
bun run test        # test suite
bun run ts          # TypeScript checks
bun run build       # production build
bun run format      # Prettier
bun run docker      # Redis + app in Docker
bun run docker:redis

Redis setup

The bundled Docker setup uses:

• redis:alpine
• Port 6300 on the host
• redis://redis:6379 inside Compose

For local development outside Docker, the app reads REDIS_URL from .env.

Redis Cloud

You can also point the app at Redis Cloud by setting REDIS_URL to your Redis Cloud connection string.

Example:

REDIS_URL="redis://default:<password>@redis-xxxxx.region.provider.redns.redis-cloud.com:12345"

What the app does

1. Create a project and provide a title plus a working brief.
2. Load source docs into Redis.
3. Chunk and embed those docs for vector search.
4. Ask questions over the indexed docs.
5. Edit markdown and reuse editing preferences through Redis-backed memory.

Learn more

• Redis docs
• Redis tutorials
• Redis Cloud