DeepBlueDynamics/typesafe-arena
Organization repository on GitHub
A playground for TypeSafeAI's Jev Model
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Demos & Experiments
TypeSafe Arena
A local markdown mirror of the TypeSafe documentation plus Lume, a Rust search tool that indexes it with BM25 and a tag graph, so pattern-style queries such as intent routing can walk to the right page without a network round trip.
- Stars
- 3
- Language
- Rust
- Published by
- Community
- Author
- DeepBlueDynamics
- Last verified
- Oct 2026
About this project
Local markdown mirror of docs.typesafe.ai, crawled with grub-crawler on 2026-09-17. Starting page: Intent routing. Lume lives in lume/. It builds a BM25 index plus a semantic knowledge graph from dict/typesafe.csv, then search walks that graph (intent routing → handler, confidence-gated routing, fan-out, …). Run these from the typesafe-arena directory. Verified on 2026-09-17: 46 files, 664 sections, graph search for intent routing ranked docs/patterns/intent-routing.md first.
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- Rust
Highlights
- Lume indexes the mirror into a .lume-index directory using BM25 plus a knowledge graph built from the dict/typesafe.csv tag file.
- Graph boost is on by default at -g 0.4, -g 0 disables it, and -l raises the number of hits returned.
- Dense vectors and Ollama-based entity extraction are opt-in extras, not required for lexical and graph search.
- Verified at 46 files and 664 sections, with a graph search for intent routing ranking the intent-routing page first.
Watch out
This is a crawl of docs.typesafe.ai rather than the live site, so it is a copy that can fall behind upstream changes; the README links back to the official llms.txt index for the current version. Building the search tool needs Rust and Cargo 1.75 or newer plus a release build before anything can be indexed, and the index command has to be pointed at the docs directory and the tag dictionary. The optional paths have their own dependencies: -s dense vectors needs a NUTS token and Shivvr, and -o entity extraction needs Ollama, so neither is part of the base setup. The only quantitative check in the README is the file and section count plus that one example query. The repository also ships its own navigation lists for patterns, primitives, concepts, confidence, cookbooks, demos and SDK pages, which duplicates the official documentation structure. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
The point is local, offline search across the whole documentation set rather than page-by-page browsing. It is a crawl of docs.typesafe.ai, not a live client, so treat it as a copy that can drift from upstream and use the official llms.txt index the README links to when you need the current version.
It builds a BM25 index plus a semantic knowledge graph from the dict/typesafe.csv tag file, where phrases like intent routing, Choice, Noul and confidence become graph nodes, and search walks that graph. Indexing writes bm25.json, entity_graph.json, spelling.json and state.json into the index directory. After doc changes you can update incrementally or force a full rebuild.
No to both. The repository is a documentation mirror plus a search tool for that mirror. Nothing in the README describes calling the TypeSafe API, and the TypeSafe SDK and API reference pages here are just markdown files being indexed, not a client library you would depend on.
No. Rust and Cargo 1.75 or newer are the real requirement, then a release build of Lume. Dense vectors with -s additionally need a NUTS token and Shivvr, and LLM entity extraction with -o needs Ollama. Both are listed as optional extras for lexical plus graph search.
It was crawled with grub-crawler on 2026-09-17 starting from the intent-routing page, and the README records a verification on that date: 46 files and 664 sections, with a graph search for intent routing ranking docs/patterns/intent-routing.md first. That is a snapshot of the documentation set, so use the llms.txt index the README links when you need the current version.
Graph boost is on by default at -g 0.4 and is disabled with -g 0, and -l 20 widens the result count. After the docs change, index update refreshes incrementally against the existing index while -f forces a full rebuild. Each run writes bm25.json, entity_graph.json, spelling.json and state.json into the index directory.
The documentation set as markdown: five patterns including composite scoring, confidence-gated routing, speculative fan-out and intent routing; the Choice, Noul and Score primitives plus an advanced structure page; concepts on state, System One and how to build; a confidence page; nineteen cookbooks from citation double-checking to re-ranking and the SDE cascade; a smart-home demo; and the agent-skill, API reference and Python and JavaScript SDK pages.
More Jev resources like this
32GitHub stars
Jev-protocol random baseline that answers noul, choice, and score questions from fast pseudorandom numbers without reading the prompt.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
karminski/Jev-QuantumRust
karminski/Jev-QuantumRust53GitHub stars
Rust server that exposes a System One compatible API at /v1/systemone for self-hosted Laya checkpoints with token-based batching.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
alvarobartt/sys1Rust
alvarobartt/sys1Rust12GitHub stars
Local decision engine that answers letter-choice questions from one forward pass: shared prefill, one batched decode, and softmax over option letters with no text generation or repair.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
emnlmn/snapRust
emnlmn/snapRust4GitHub stars
A Rust CLI of Jev experiments, led by a PHI screener that judges four separate 45 CFR 160.103 metrics per file and an English part-of-speech tagger that batches only the unresolved words into one request.
Jev Demos & Experiments#community#jevbest#catalog-import
46GitHub stars
Open 150M encoder that reproduces the typed noul, choice, and score interface in one forward pass, trainable on a free GPU and runnable locally for nothing.
Jev Demos & Experiments#community#jevbest#catalog-import
941GitHub stars
Library that adds schema-guaranteed typed outputs to existing autoregressive LLMs without changing weights, served through SGLang or a hosted API.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
TypeLLM/TypeLLMPython
TypeLLM/TypeLLMPythonFrom the Jev guides
Original write-ups on what this entry is for.
A game tick is a state plus a legal action list. How Clash Royale, Minecraft, and the earlier Mario, Pokémon, StarCraft, and drone loops ask Jev one step at a time.
Jev Demos & Experiments#gaming#agents#control
Read article
From the Jev community
Posts from builders shipping with Jev right now.