jevbox
Permission-aware document library using Jev or Cloudflare Clef for hierarchical search, passage scoring and automatic filing.
Overview
Jevbox is a full-stack, permission-aware document library from Extend. You browse documents in grid, list, column or gallery views, inspect document trees and parsed output, share with organisation members, and ask questions in a chat that answers with inline citations and source previews. Generation goes through the AI SDK, while the decision model, TypeSafe Jev or Cloudflare Clef, performs hierarchical beam search through categories, documents and sections and then scores source passages independently for usefulness.
Search is permission-filtered, uses usefulness thresholds, widens exploration into alternate routes when first-pass evidence is weak, and blocks revoked sources. New uploads are filed automatically: the model walks existing folders, keeps uncertain documents in a suitable parent, and validates occasional model-proposed new branches, with upload-driven reviews of related documents. PostgreSQL with pg-boss handles indexing, filing, chat and MCP search jobs, and SpiceDB handles permissions. Queued messages and generations survive reloads and restarts. It requires Node 24.6+, pnpm and Docker, and can be deployed to Render.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which category, document or section branch should the search descend into? | — |
| Score | How useful is this passage for answering the question? | — |
| Choice | Which existing folder should this new upload be filed into? | — |
How it works
- Upload documents; model files them into existing folders
- User asks a question in chat
- Permission-filtered beam search through the hierarchy; passages scored
- LLM answers with citations to the selected sources
Strengths
- Permission-aware retrieval and filing
- Uncertain documents kept in safe parent folders
- Supports Jev or Cloudflare Clef
- Durable jobs and chat across restarts
Caveats
- Heavier stack (PostgreSQL, SpiceDB, Docker)
- Retrieval quality not benchmarked in the README
Getting started
pnpm install, pnpm setup:local, pnpm services:up, pnpm dev.At a glance
- Type
- Library
- Status
- beta
- Language
- TypeScript
- Runtime
- Node web app + PostgreSQL + SpiceDB
- Has code
- yes
Use cases and integrations
How curated lists describe it
Permission-aware document library: browse document trees and parsed output, search the hierarchy with Jev, and ask source-grounded questions.
MrJev/awesome-jev
Frequently asked questions
What does the decision model do?
Hierarchical beam search through the library, passage usefulness scoring and automatic filing.
Which decision models are supported?
TypeSafe Jev or Cloudflare Clef.
Are permissions respected in search?
Yes, search is permission-filtered and revoked sources are blocked.
Related projects
Python library that replaces hard-coded RAG thresholds with calibrated decisions for stopping, chunking, selection, abstention and caching.
ServicebetaPythonLong-document search that walks a PageIndex tree, asking Jev Choice which page or branch answers the question, with no embeddings.
ProjectprototypePythonGraph-database-agnostic GraphRAG framework where swappable Laya or Jev decisions score edges, route intents and gate hallucinations.
ServicebetaPythonMCP server giving coding agents typed judgment tools (verify, screen, rerank, gate, score) backed by Jev, with auto/review/escalate policy.
MCP serverbetaPythonTyped-decision accuracy layer for RAG: six checks for intent, relevance, clarification, sufficiency, grounding and caveats.
ServicebetaPythonSmall stdio MCP server with ask, rank and verify tools backed by Jev, a local event log and installers for several agent harnesses.
MCP serverbetaTypeScript