WHAT NEW IN X11.05 THU

X 上的 AI 合集档案 · 23 个合集 · 69,408 条

jevIMAGE

7 个把 Jev 接进 Claude Code 与 Postgres 的开源项目

合集列出 typesafe-mcp、jev-mcp、jev-codex-router、Winnow、SemDecide、jev-curate、pg-jev,用 Jev 做 MCP 决策、任务路由、上下文裁剪与数据库查询。

damkin
@damkina72026.09.26收录原推发布于 2026.09.25
7 个把 Jev 接进 Claude Code 与 Postgres 的开源项目
damkin @damkina7原推发布于 2026.09.25
在 X 上打开原推

this is pure f*cking insane The remaining 7 open-source projects turning Jev into a full-stack typed decision substrate for Claude Code, Codex, and production databases: protocol servers. agent judgment. task routing. context pruning. cli binaries. parquet streaming. postgres extensions. 08 typesafe-mcp — plug Jev directly into Claude Code, Claude Desktop, Codex, and Pi over Model Context Protocol to execute Choice, Score, and Gating anytime. https://github.com/itsmostafa/typesafe-mcp 09 jev-mcp — ready-made agent judgment toolkit: fact-checking, content screening, semantic ranking, and classification before state commits. https://github.com/jkudish/jev-mcp 10 jev-codex-router — evaluates coding task difficulty in 180ms, dynamically routing between speed models and deep reasoning tiers to slash 70% in compute cost. https://github.com/0xNatoshi/jev-codex-router 11 Winnow — context garbage collection for Claude Code. When bash, grep, and terminal output dump 10,000 lines, Jev judges and prunes low-utility noise before token burn. https://github.com/GhalebDweikat/winnow 12 SemDecide — turns Jev into a command-line binary. Directly classify, score, and filter in the shell for web crawlers, data pipelines, and CI approval gates. https://github.com/sharziki/semdecide 13 jev-curate — high-throughput dataset curation: streams Parquet and JSONL rows through typed confidence filters at 1,500+ rows a second, keeping only what clears the bar. https://github.com/AkashPriyadarshii/jev-curate 14 pg-jev — native PostgreSQL extension that lets you run typed plain-English queries directly inside your database tables with sub-millisecond execution. https://github.com/realZachi/pg-jev the core rule: Over 80% of tool-calling agent spend goes into the trash because engineers ask models to write sentences when they need a binary verdict. Stop asking language models to write essays when you need an if-statement. Wire typed decision engines into the loop and save reasoning compute for actual reasoning. Bookmark this before building your next autonomous agent ↓

这条是引用着另一条写的:x.com/damkina7/status/2102860956724335039

x.com/damkina7/status/2103493356860911901采集于 2026.09.26 10:11