Helix: L1 Cache for LLMs

• 🌐 Website: helix-llm.app
• 🔗 Sign up on helix-llm.app to use Helix MCP: MCP Server URL
• 🔌 Docker MCP Gateway: PR made to Docker MCP registry to add Helix MCP Server to Docker MCP gateway

---

Introduction

LLMs are smart, but they forget your context.
Every prompt feels like starting from zero.

Helix fixes that.
It’s the L1 cache for your LLMs.

You can dump in your files PDFs, YouTube videos, documents, even audio and Helix makes them instantly searchable.With support for 20+ file types, Helix ensures you can bring in almost any kind of knowledge source seamlessly.

Then you connect Helix to your favorite LLM chat app like ChatGPT, Claude, or even Cursor and it instantly gets a super-fast search tool, powered by:

• ⚡ Cerebras Systems for blazing-fast inference and indexing
• 🦙 Llama providing the intelligence layer
• 🐳 Docker MCP Gateway allows you to run Helix MCP Server in a container and use it with any MCP-compatible client

Together, these make Helix super fast, reliable, and context-aware, so your AI always has the right memory at the right time.

---

Dashboard Preview

---

System Architecture

Multi-Agent Search System

---

Helix is designed as a modular system that wraps LLMs with a fast, context-aware caching layer. The two main components that make up Helix are:

1. Web Application

• Users log in and upload files or links (PDFs, YouTube videos, documents, audio).
• Users can then plug the Helix MCP into their preferred LLM chat app (ChatGPT, Claude, Cursor etc.) to enable context-aware responses.

2. Chrome Extension

• Allows users to save links directly while browsing, with just one click.
• Works as a lightweight, complementary tool to the web app, making context capture seamless.

Overall Flow

1. User interacts with web app or Chrome extension to provide content.
2. Helix processes, converts, and stores content in the memory library.
3. Helix MCP serves as the bridge between the memory library and any LLM, enabling fast, context-rich responses.
4. The system ensures that the LLM always has the right context, no matter the file type or source.

---

Repository Structure

helix/
│── chrome_extension/   # Chrome extension for one-click link saving
│   └── README.md
│
│── frontend/           # Frontend web application (login, upload, UI)
│   └── README.md
│
│── helix-mcp-remote/   # Helix Remote MCP Server
│   └── README.md
│
│── helix-mcp-docker/   # For Docker MCP Gateway Integration
│   └── README.md
│
│── service/            # Backend services: file processing, storage and multi-agent search
│   └── README.md
│
└── README.md           # Root README (this file)

Folder Purpose Overview

• chrome_extension/
  Save links directly from the browser to Helix’s memory library.

• frontend/
  User-facing interface for uploading and managing content, displaying search results, and interacting with LLMs.

• helix-mcp-remote/
  Helix Remote MCP Server that enables any MCP enabled client to retrieve context from the memory library.

• helix-mcp-docker/
  For Docker MCP Gateway Integration

• service/
  Backend services: file processing, storage and multi-agent search.

Each folder contains its own README.md with instructions on setup and usage.
Check them out to learn more about each component of the project.

---

Features

Helix acts as the ultimate knowledge layer for your LLMs. It offers a unique set of features that combine content ingestion, processing, and seamless LLM integration.

1. Core Intelligence & Speed

• L1 Cache for LLMs: Provides super fast search, ensuring your LLM can access specific knowledge in seconds, not minutes.
• Powered by Cerebras Inferencing: We leverage Cerebras Inferencing to make our search agents and file processing super fast.
• Llama Intelligence: Our search agents are powered by the intelligence of Llama models.

2. Universal Content Ingestion

• Multi-Modal Upload: Easily upload and process a wide variety of file types, including PDFs, documents (DOCX, TXT), audio files, and video links (YouTube).
• Automatic Conversion: All uploaded content is automatically processed and converted into structured markdown for uniform storage and optimal retrieval quality.
• One-Click Capture (Chrome Extension): The dedicated Chrome Extension lets you save web links directly to your Helix memory library while browsing, making context capture seamless.

3. Seamless Integration

• LLM Agnostic Context: Helix MCP can plug into popular apps like ChatGPT, Claude, Cursor etc. instantly giving your favorite chat apps access to your private knowledge base.
• Docker MCP Gateway Integration: You can run Helix MCP Server in a container and use it with any MCP-compatible client.

---

Built with ❤️ by

• Adarsh Dubey
• Shreya Pandey
• Aditya Borse

---

License

This project is licensed under the MIT License. See the LICENSE file for details.