Jev 搭配 20 个开源工具的决策层组合清单

Kiyoro @0xKiyoro原推发布于 2026.09.30

在 X 上打开原推

THIS IS PURE TREASURE FOR EVERY AI BUILDER. 20 OPEN-SOURCE TOOLS. 1 DECISION LAYER. STEAL EVERY COMBO. Jev gives you 3 primitives: Choice, Score, Noul. It makes one narrow judgment. Code enforces the rules. Another tool does the work. ROUTE THE WORK 01 Jev × Hermes Agent - Choice: which skill to load ▸https://github.com/NousResearch/hermes-agent 02 Jev × LangGraph - Choice: which node runs next ▸https://github.com/langchain-ai/langgraph 03 Jev × MCP SDK - Noul: is this tool call safe to run ▸https://github.com/modelcontextprotocol/typescript-sdk 04 Jev × Temporal - Choice: retry, replan or wake a human ▸https://github.com/temporalio/sdk-typescript 05 Jev × E2B - Score: how risky is this code before it runs ▸https://github.com/e2b-dev/E2B FIND THE EVIDENCE 06 Jev × Meilisearch - Score: rerank the top hits ▸https://github.com/meilisearch/meilisearch 07 Jev × Qdrant - Score: which of 20 matches actually answer ▸https://github.com/qdrant/qdrant 08 Jev × Docling - Choice: what kind of document is this ▸https://github.com/docling-project/docling 09 Jev × Graphiti - Choice: new fact, duplicate or contradiction ▸https://github.com/getzep/graphiti 10 Jev × LlamaIndex - Noul: is this context enough to answer ▸https://github.com/run-llama/llama_index VERIFY THE RESULT 11 Jev × Langfuse - Score: grade every trace ▸https://github.com/langfuse/langfuse 12 Jev × Promptfoo - Noul: did the output pass the rubric ▸https://github.com/promptfoo/promptfoo 13 Jev × Semgrep - Score: real vulnerability or noise ▸https://github.com/semgrep/semgrep 14 Jev × Sentry - Choice: which team owns this error ▸https://github.com/getsentry/sentry 15 Jev × Playwright - Noul: is the task actually done ▸https://github.com/microsoft/playwright OPERATE THE PRODUCT 16 Jev × GitHub MCP - Choice: label and route the issue ▸https://github.com/github/github-mcp-server 17 Jev × PostHog - Score: churn risk of this user ▸https://github.com/PostHog/posthog 18 Jev × Resend - Noul: send this email now or hold ▸https://github.com/resend/resend-node 19 Jev × Stripe - Score: fraud risk before capture ▸https://github.com/stripe/stripe-node 20 Jev × Cloudflare Workers AI - Choice: which model takes this request ▸https://github.com/cloudflare/workers-sdk The loop: collect state → ask one Choice, Score or Noul → check confidence and policy in code → run the allowed step or ask a human → log it Jev doesn't do the work. It decides which work gets done. These are buildable architecture ideas, not official integrations between the projects. Save this before you build your next agent ⭣