JevLoop is a zero-dependency TypeScript agent loop that routes control decisions to TypeSafe AI's Jev decision model, reserving LLM calls for generation. It includes a runnable demo, DECISION.md policy compilation, and benchmark/latency comparisons.
该仓库是实质性的 TypeScript 项目，使用 TypeSafe AI 的 Jev 决策模型构建 Agent 循环，将分支判断从 LLM 生成中分离，并包含可运行 demo、DECISION.md 编译、规则回退、基准和延迟测试，是真实的 Jev 集成与研究。
walidboulanouarwalidboulanouar/jev-agent-kitCLI and MCP tools for agents using TypeSafe Jev to make fast typed decisions: route, triage, guard, grep, rank, compact, and judge. Includes runnable use-case demos and an offline evaluation script.
buberlobuberlo/dsh-jevA DeepSeek Harness plugin that uses TypeSafe Jev/System One for dynamic tool selection, pre-execution call assessment, model routing, and skill routing. Includes runnable examples, 120 tests, 25 evaluation fixtures, and a with/without-Jev benchmark.
tonyzdevtonyzdev/pijevPiJev is a terminal coding agent built on Pi that uses TypeSafe Jev to rank files, pick skills, and triage tool failures while the coding model writes code. The repository includes runnable code, tests, and controlled experiments showing fewer tool calls and faster file discovery.
