extend-hq / jevbox

# Jevbox
A full stack, permission-aware document library. Browse with Extend UI Finder, inspect document trees and parsed output, share with organization members, search the hierarchy with JEV, and ask source-grounded questions through the AI SDK.
Render setup requirements: prepare the shared environment group, private S3 bucket, and SMTP credentials before deploying.
### Self organizing drive & search through folder trees in chat
Ask questions across your library, follow inline citations, and preview the original source without leaving the conversation.
Switch between visual browsing, a compact outline, folder columns, and full-page previews.
| Grid | List |
| :-----------------------------------------------------------------------------------------------: | :----------------------------------------------------------------------------------------------------: |
| | |
| **Columns** | **Gallery** |
| | |
 Captured from the app using a demo workspace. 
## Run locally
Requires Node 24.6+, pnpm 12.9.1, and a running Docker engine with Compose.
For the first run, from the repository root:
 [code] 
`pnpm setup:local` creates `.env` and a stable encryption key. `pnpm services:up` starts PostgreSQL and SpiceDB in Docker and waits for both to be ready. `pnpm dev` then starts the app server, Vite frontend, and pg-boss consumers in your terminal. **`pnpm dev` does not start PostgreSQL or SpiceDB.** Keep that terminal open while using the app.
On later days, start Docker and run `pnpm services:up` followed by `pnpm dev`. Open http://localhost:4310, create an organization, and configure **Settings → Connections**:
- **Extend API key** for parsing PDFs, Office documents, and images. Plain text, Markdown, code, JSON, HTML, and CSV index locally.
- **Decision model: TypeSafe or Cloudflare Clef** for hierarchical beam search through categories, documents, and sections, followed by independent usefulness scoring of source passages. Search and document questions require this connection; there is no keyword or embedding fallback.
- **Chat providers** and model IDs for answers with page and section citations. Indexing stores the parsed structure, heading outlines, and original source passages without generative summaries. Automatic filing can use a separately selected model for occasional folder names and descriptions. Members can choose among administrator-configured models and attach up to eight accessible, indexed documents to scope an answer.
Keys are encrypted on the server and never sent back to the browser. Leave a key blank to retain it, or use **Clear on save**. Configure additional provider options through the encrypted advanced JSON field.
 [code] 
Alternatively, run the complete app and its services in Docker after `pnpm setup:local`:
 [code] 
PostgreSQL stores accounts, sessions, document metadata, parsed output, and chats. Production uses private S3 storage for original files and thumbnails; local development defaults to PostgreSQL bytes. See file stora