kobolingfeng

kobolingfeng / ntk

Public

NTK — 基于信息密度路由的多智能体框架。用最少的 token 做最多的事。

agent-framework ai llm mcp multi-agent
15
0
100% credibility
Found Mar 31, 2026 at 15 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
TypeScript
AI Summary

NTK is an open-source multi-agent framework that applies a 'need-to-know' principle to minimize context for AI agents, achieving cost savings and efficiency in tasks like code generation and analysis via adaptive pipelines and editor integrations.

How It Works

1
🔍 Discover NTK

You hear about NTK, a smart helper that makes AI assistants focus better by sharing only what they need to know for tasks like writing code or comparing ideas.

2
📥 Add to your editor

Download and add NTK to your favorite code editor like VS Code or Claude Desktop with a simple setup guide.

3
🔗 Link your AI service

Connect your preferred AI thinking service so NTK can use it wisely without wasting effort.

4
💬 Ask for help

Type a request like 'write a sorting function in Python' or 'compare React and Vue' right in your editor chat.

5
Get quick smart results
🚀
Super fast for easy asks

Simple tasks finish in seconds with perfect focus.

🧠
Deeper for complex ones

Tough tasks get team-like help but still efficient.

🎉 Enjoy efficient results

You receive clear, accurate code or advice quickly, saving time and money compared to regular AI chats.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 15 to 15 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is ntk?

NTK is a TypeScript multi-agent framework that slashes token usage in AI tasks via information density routing. It enforces a "need-to-know" rule: agents get minimal context through adaptive pipelines (direct to full depth) and dual-model setup, routing 95% work to cheap models like gpt-4o-mini. Users get CLI commands like `npx ntk run "Design REST API"`, MCP tools (ntk_run, ntk_run_fast, ntk_compress) for VS Code Copilot/Claude Desktop, and an HTTP API server.

Why is it gaining traction?

Unlike bloated frameworks dumping full history on every agent, NTK prioritizes selective forgetting and regex fast paths (63% tasks LLM-free), delivering 90%+ cost savings with zero quality loss per benchmarks. Dual-model isolation and compression backtracking prevent info loss, while MCP plug-and-play turns it into a token-efficient sidekick for chats. Devs notice faster runs (10s vs 37s) and precise outputs on code gen, debugging, comparisons.

Who should use this?

AI-heavy devs crafting code, APIs, or analyses via Copilot Chat/OpenClaw, especially those burning tokens on simple tasks. Backend teams optimizing ntk landau/neu-ulm-style pipelines, or frontend folks needing quick refactors/debugs without ntk ngk bloat. Token-watchers integrating ntk shop/youtube workflows with MCP clients.

Verdict

Strong benchmarks make NTK a clever token miser for TypeScript/MCP stacks, but 15 stars and 1.0% credibility mean it's raw—docs solid, tests cover core, yet maturity lags. Prototype it for high-token use cases; AGPL suits open experiments.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.