Emmimal

Emmimal / context-engine

Public

A pure-Python context management layer for LLM systems — retrieval, re-ranking, memory decay, and token-budget enforcement in one pipeline.

context-engineering large-language-models llm prompt-engineering rag
20
5
100% credibility
Found Apr 16, 2026 at 28 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

A pure-Python tool for managing context in AI language model conversations by retrieving relevant documents, handling memory with decay, compressing information, and enforcing token limits.

How It Works

1
🔍 Discover the Helper

You learn about a simple tool that makes AI conversations smarter by picking the right info and remembering chats without overload.

2
📥 Set It Up

You download it to your computer and prepare it with a couple of easy steps, no fuss.

3
📄 Load Your Info

You add notes, articles, or facts you want your AI to know about, like filling a knowledge drawer.

4
🚀 Create Your AI Buddy

You build your personal AI assistant, setting how much info it juggles at once to keep things snappy.

5
💬 Ask Away

You start asking questions, and it grabs the most relevant bits from your info to answer well.

6
🧠 Smart Memory Kicks In

It remembers your chat history, prioritizes important parts, and trims old stuff so chats stay fresh and focused.

🎉 Smarter Chats Unlocked

Your AI now delivers precise, grounded answers every time, feeling like talking to a sharp expert friend.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 28 to 20 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 context-engine?

Context-engine is a pure-Python layer for LLM apps that automates retrieval, re-ranking, memory decay, and token-budget enforcement in one pipeline. It takes your documents and conversation history, builds an optimized context packet for the model—handling what gets included, compressed, and ordered under real token limits—going beyond basic RAG to manage multi-turn agentic context. Install with numpy (sentence-transformers optional for hybrid retrieval), add docs, call build() on queries, and remember() turns; run python demo.py for seven live examples.

Why is it gaining traction?

Unlike fragmented RAG tools, this context engineering kit delivers a single build() call for hybrid keyword/TF-IDF/embedding retrieval, extractive compression, and exponential memory decay—clocking ~92ms end-to-end on CPU with caching. Devs dig the tag-based re-ranking, slot-based budget slots (system/history/docs), and diagnostics output, plus tuning knobs like hybrid_alpha for query types. It's a lightweight augment context engine on GitHub, emphasizing context engineering vs prompt engineering for production-grade LLM pipelines.

Who should use this?

AI engineers building multi-turn chatbots or agentic systems with growing knowledge bases, like customer support bots pulling from docs while enforcing 2k token limits. Backend devs prototyping RAG for codebases or context-engineering-for-ai-agents, especially those hitting context bloat in long sessions. Skip for single-shot queries or sub-50ms latency needs.

Verdict

Solid prototype for context engineering intro—great docs, runnable demos, MIT license—but 20 stars and 1.0% credibility score signal early maturity; no tests or persistence yet. Grab it for LLM experiments if you need decay and enforcement now.

(187 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.