Glia

Product Introduction

1. Definition: Glia is a 100% offline, open-source memory bridge and hybrid RAG (Retrieval-Augmented Generation) engine. Technically, it is a dual-mode system comprising a Chrome extension for web-based AI chats and a native MCP (Model Context Protocol) server for AI-powered coding agents, both sharing a single local SQLite database.
2. Core Value Proposition: It exists to solve the pervasive problem of AI context loss, providing persistent, project-isolated memory for AI coding agents and browser chats without any cloud dependency, subscriptions, or data privacy risks. Its primary value is enabling AI tools to remember past conversations and code decisions locally.

Main Features

1. Hybrid RAG Engine: Glia employs a three-layer fused search architecture combining Sentence Vector, Chunk Vector, and FTS5 keyword search. This allows for surgical precision, returning only the most relevant sentences from stored memories, achieving up to 95% context compression versus raw chunk injection.
2. Dual-Mode Operation (Web Extension & MCP Server): The browser extension auto-intercepts and enriches prompts on platforms like Claude.ai, ChatGPT, Gemini, and DeepSeek with relevant project context. Simultaneously, the MCP server provides native tools (recall_context, store_memory, search_memory) to coding agents in Cursor, Claude Code, Windsurf, and VS Code, with both interfaces reading from and writing to the same glia.db SQLite file.
3. Knowledge Graph & HyDE Retrieval: Conversations are automatically parsed into a D3 force-directed graph of entities and relationships for visual browsing. The system also uses Hypothetical Document Embeddings (HyDE), generating a synthetic ideal answer to a query and then searching by that embedding's vector, significantly improving recall on rephrased queries.

Problems Solved

1. Pain Point: AI tools like Claude and ChatGPT suffer from severe context window limitations and have no inherent memory between sessions, forcing developers and users to repeatedly re-explain project details, architecture decisions, and past conversations.
2. Target Audience: The primary users are software developers, engineers, and technical teams using AI coding assistants (Cursor, Claude Code) and engaging in detailed technical discussions on web-based AI chat platforms. Secondary users include anyone requiring persistent, private memory for AI interactions.
3. Use Cases: Essential for a developer who needs their AI pair-programmer to remember the API schema discussed yesterday; for a team to maintain context across multiple AI coding sessions on a shared codebase; for a researcher to query a long history of AI conversations about a specific topic without manual search.

Unique Advantages

1. Differentiation: Unlike cloud-based memory services or simple prompt wrappers, Glia is a local, open-source infrastructure. It offers 100% project isolation at the SQL level, ensuring zero cross-project data leaks—a critical differentiator from solutions that use a single vector store. It also requires zero Docker, running entirely on SQLite with the sqlite-vec extension.
2. Key Innovation: The seamless, bidirectional memory bridge between web chat and local coding environments is its core innovation. The shared SQLite database acts as a universal memory layer, meaning context saved in a casual ChatGPT conversation is instantly recallable by the Claude Code agent in your editor, creating a truly unified AI memory ecosystem on your machine.

Frequently Asked Questions (FAQ)

1. Is Glia AI secure and private? Yes, Glia is 100% offline and open-source. All data—conversations, context chunks, and knowledge graphs—is stored locally in a SQLite database on your machine. No data is sent to any cloud service or external server, ensuring complete privacy and security.
2. How does Glia work with Cursor and Claude Code? Glia integrates via the Model Context Protocol (MCP). You configure the Glia MCP server in your editor's or Claude Desktop's settings. Once connected, the AI agent can natively call tools like recall_context to automatically fetch relevant memories based on your current project's file path.
3. Can I share memories between different computers or with my team? Yes, Glia supports portable JSON sessions. You can export any project's memory session as a clean JSON file from one installation and import it into another, enabling easy context sharing and synchronization across machines or team members.
4. What AI chat platforms does the browser extension support? The Glia Chrome extension auto-injects context on claude.ai, chatgpt.com, Gemini, DeepSeek, Grok, Copilot, and Mistral AI chat interfaces, covering the major platforms for AI-assisted conversation.
5. How accurate is Glia's memory retrieval? According to published benchmarks (v1.5.1) audited against 1,000-chunk noise haystacks, both the Web and MCP context engines achieve 90.0% recall accuracy with significant context compression (95% and 81.3% respectively), demonstrating high precision in retrieving relevant information.