A week of Jev repos: routers, MCP, graphs, games

Forty-eight hours after launch, discussion on X shifted toward what practical tasks Jev could handle.

A roundup from @0xLogicrw, framed as an Awesome list for developers receiving early API access, grouped initial projects around a shared structure: ask a bounded question, receive a probability, and let code act on the result. A follow-up post linked to a generated directory claiming over 130 projects updated via GitHub Actions. Independently, hellogumbo/awesome-jev compiled a separate community catalog. Project counts on both lists continue to shift.

The original fourteen entries outline how builders initially applied the model:

1. jev-ultrafast: Browser Use’s indexed action-space browser agent.
2. typesafe-mcp: a single-binary MCP server exposing Choice, Score, and Noul to tools like Claude Code, Claude Desktop, and Codex.
3. jev-mcp: verification, screening, and ranking tools for agents offloading deterministic checks from frontier models.
4. SemDecide: a semantic utility shaped like grep or jq for command pipelines and CI.
5. Jev Codex Router: per-turn model and reasoning-depth routing, with the author reporting a 60% cost reduction on a 237-turn replay.
6. Winnow: a context filter for oversized tool results in Claude Code.
7. jev-review: a first-pass review tool for correctness, security, and reliability that escalates flagged issues to larger models.
8. Blink: a semantic codebase traverser that scores whether subdirectories warrant expansion.
9. neo4jev: graph traversal using edge classification and beam search.
10. jev-desktop: an accessibility-tree desktop loop integrated into agent-desktop.
11. TypeSafe AI Playground: a Rust CLI containing small experiments for PHI detection, comment moderation, and tone checks.
12. Prism: a market analysis tool where Jev scores toxic flow and mean reversion, running in advisory mode without executing orders.
13. OneVOneJev: a first-person shooter demonstration evaluating movement, aim, and firing decisions at roughly 9 Hz.
14. safer-with-jev: a router directing queries ahead of larger chat models. A related Neon article is available at typesafe-on-neon.

The roundup suggested using Jev for bounded evaluations like selecting options, calculating scores, or deciding next steps. Generating code patches, essays, or complex plans remains the domain of generative language models.

Savings and latency figures in those repositories were reported by their respective authors without independent verification.