Lury - local agentic AI for free

Explore Lury, a local-first agentic AI for private chat, research, and document work. This talk covers its custom memory and reasoning for small models acting large.

llama-cpp Tauri React Qwen document indexing

Overview

Local Understanding & Reasoning sYstem - Lury
A local-first agentic AI in a box - chat, research, work with documents locally. Build for cases where online exposure and telemetry is not an option ( law firms, medical institutions, concerned citizens ), but AI is needed. Free for individuals. https://lury.org

Links

https://lury.org
Lury provides local-first, zero-telemetry AI chat for private individual use.

Tech stack

llama-cpp

The definitive C/C++ library for high-performance LLM inference: run quantized GGUF models on commodity hardware, CPU-first.

Llama.cpp is the C/C++ engine for highly efficient Large Language Model (LLM) inference, built on the GGML tensor library. It champions a CPU-first philosophy, enabling models up to 70B parameters to run locally on diverse hardware (laptops, Raspberry Pi) with minimal dependencies. This efficiency is driven by the GGUF file format and advanced quantization techniques (1.5-bit to 8-bit), allowing 7B+ parameter models to operate comfortably with just 4–8GB of RAM. The project provides a fast, hackable framework, including command-line tools (`llama-cli`) and an OpenAI-compatible web server (`llama-server`), ensuring state-of-the-art performance across various backends, including robust GPU support.

https://github.com/ggml-org/llama.cpp

View projects
Tauri

Tauri builds tiny, secure, and blazingly fast cross-platform binaries using a Rust core and any web frontend (HTML, JS, CSS).

Tauri is an open-source application toolkit: it delivers lightweight, secure desktop and mobile apps across Windows, macOS, Linux, Android, and iOS. Unlike frameworks that bundle a browser engine, Tauri leverages the host OS's native webview (e.g., WebView2, WKWebView), resulting in minimal app sizes—a minimal binary is often less than 600KB. The architecture pairs a flexible web frontend with a high-performance, security-focused backend written in Rust, utilizing Rust's memory and type-safety features. It includes essential tooling: a built-in bundler, a secure self-updater, and a tree-shakeable JavaScript API for controlled system access.

https://tauri.app/

View projects
React

React is an open-source JavaScript library for building dynamic user interfaces (UIs).

React is a component-based JavaScript library, developed by Meta (Facebook), engineered for building fast, declarative UIs. It mandates a one-way data flow and utilizes a Virtual DOM mechanism to ensure efficient, predictable updates to the user interface. Developers construct complex UIs by composing small, encapsulated components; this architecture promotes code reusability and simplifies state management across large applications. The library employs JSX (a syntax extension) to integrate HTML-like markup directly within JavaScript logic, supporting development for both web (React DOM) and native mobile platforms (React Native).

https://react.dev/

View projects
Qwen

Alibaba Cloud's Qwen is a family of advanced, multilingual large language and multimodal models (LLM/LMM) with both proprietary and open-weight versions.

Qwen is the large language model (LLM) and large multimodal model (LMM) family engineered by Alibaba Cloud, designed for state-of-the-art text, vision, and audio processing. The model series, including the Qwen3 generation, features a comprehensive range of dense and Mixture-of-Experts (MoE) models, scaling from the efficient Qwen3-0.6B to the powerful Qwen3-235B-A22B. Qwen excels in complex tasks: it supports reasoning, agent capabilities (tool use), and instruction-following across 119 languages, with open-weight variants available under the Apache 2.0 license for broad deployment.

https://chat.qwen.ai

View projects
document indexing

Document indexing transforms unstructured data into a high-speed searchable catalog using metadata tags and inverted indexes.

Indexing is the engine behind sub-second retrieval in systems like Elasticsearch and Amazon Kendra. It maps specific terms to their exact locations across millions of PDFs, spreadsheets, and HTML files (the inverted index method). By extracting entities like dates, names, or SKU numbers, the system bypasses slow linear scans in favor of direct pointer lookups. This process reduces search latency from minutes to milliseconds, ensuring that a repository of 100 terabytes remains as navigable as a single page.

https://aws.amazon.com/what-is/indexing/

View projects

Lury - local agentic AI for free

Related talks