Production inference server for open Jev-like decision models (Jev/SystemOne /v1/systemone). Clef-Flash on vLLM: 45 ms p50, 2x throughput, verified against the reference.
-
Updated
Oct 2, 2026 - Python
Production inference server for open Jev-like decision models (Jev/SystemOne /v1/systemone). Clef-Flash on vLLM: 45 ms p50, 2x throughput, verified against the reference.
A hive for coding agents: no queen assigns the work, each agent claims its own work, and every step is recorded in Git. Built on Cloudflare Workers, Artifacts, Durable Objects, Containers, D1, Queues, Clef and Clef-flash.
A curated directory of projects, tools, SDKs, demos and resources built with Cloudflare Clef.
Cloudflare/clef-flash(9B の構造化判定モデル。テキストと画像)を、onnxruntime と numpy だけで推論する最小依存の推論実装。 ※GPU への切り替えも可
本地 clef-flash 判断 + 本地文本模型起草的聊天回复助手:全程离线、零 API key、零账号、零遥测。Apache-2.0(上游 MIT)。
To associate your repository with the clef-flash topic, visit your repo's landing page and select "manage topics."