ldraw-nova
Agentic LEGO builder that turns an idea into an LDraw model, using Jev reranking to find suitable parts and example models.
Overview
ldraw-nova lets you give an AI agent a model idea, guide it, and get back a buildable LEGO model in LDraw format, with 3D viewer and player views, VR on Meta Quest 3, a Blender-editable glTF file with metadata, and the chat history and agent reasoning. It runs as a Dockerised web app built from two sibling repositories.
Jev's role is in search. The tools agents use to find suitable parts and example models rely on jev-rerank, the same author's semantic search with reranking backed by Jev. With a TypeSafe key set in the app's settings, search results are reranked by Jev's judgment of how well each part or model description matches; without a key, reranking does not work and agents fall back to full-text search, which the README says may yield worse models. The app has no login, so it should only run on trusted networks.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Score | How well does this part or example model match what the agent is looking for?(inferred) | — |
How it works
- User describes a model idea
- Agent searches the LDraw parts and model library
- Jev reranks search candidates by relevance
- Agent assembles the model in LDraw
- App renders 3D, VR and glTF outputs
Strengths
- Better part selection with semantic reranking
- Graceful fallback to full-text search
- Rich outputs including VR and Blender export
Caveats
- Reranking requires a TypeSafe key
- No login; trusted networks only
- Docker build needs about 5 GB
Getting started
Clone ldraw-nova and ldraw-nova-docker at the same tag, then docker compose build and docker compose up -d in ldraw-nova-docker.At a glance
- Type
- Project
- Status
- beta
- Language
- Python
- Runtime
- Dockerised web app
- Has code
- yes
Use cases and integrations
How curated lists describe it
Agent tooling for building LEGO models, where Jev judges part and model descriptions during search.
MrJev/awesome-jev
Frequently asked questions
Where does Jev come in?
It reranks search results for parts and example models through jev-rerank.
Does it work without a key?
Yes, but search falls back to full-text, which may yield worse models.
What do I get at the end?
LDraw source, 3D and VR views, a glTF file and the agent's chat history.
Related projects
Web app that has Jev score emoji categories and candidates 0 to 3 for relevance to a short text and returns many matches.
DemoprototypeTypeScript- Jev Search★ 534
Plain-language web search where a decision model like Jev picks query, sources and time range, then ranks Search1API results.
AppproductionTypeScript MCP server over Apple developer docs and WWDC transcripts that uses hybrid search, then Jev relevance scores to rank results.
MCP serverproductionTypeScriptLocal long-term memory for coding agents with an optional Jev reranker that orders recalled memories.
ProjectproductionTypeScriptVolcengine's context database for agents, with a Jev rerank client that scores candidate documents with jev-latest as relevance.
LibraryproductionPython- jev-rerank★ 17
Jev as a calibrated reranker: query plus up to 30 documents in one call, one relevance Noul per document, benchmarked on BEIR.
ProjectbetaPython