Khoj

KhojSince 2025-04

Khoj is an open-source AI desktop assistant and personal second brain that allows users to search, chat, and synthesize insights across their personal notes, documents, and code repositories. Operating seamlessly with local SLMs (via Ollama) or hosted frontier models, Khoj prioritizes user privacy and offline autonomy. Whether indexing Obsidian markdown vaults, PDF research papers, Emacs org-mode files, or browser bookmarks, Khoj acts as a unified knowledge retrieval agent that answers complex multi-hop questions directly from your private workspace.

Khoj supports scheduled automated research agents that browse the web, compile daily intelligence briefings, and synthesize relevant industry updates directly to your inbox or WhatsApp. The architecture combines dense embeddings with hybrid local vector indexes, ensuring sub-second response times on standard consumer laptops. Developers can run Khoj as a native desktop client, Emacs package, Obsidian plugin, or private self-hosted Docker server with complete offline privacy.

Mem0Since 2024-06

Mem0 (formerly Embedchain) is a universal, persistent memory architecture designed to solve the critical context amnesia problem in modern AI applications. While foundational LLMs forget user preferences and past interactions the moment a session ends, Mem0 maintains a continuous, self-improving memory graph across user sessions, agents, and applications. With Mem0, developers can build personalized AI assistants, customer support agents, and autonomous workflow bots that remember user preferences, past project decisions, and communication styles over months and years. Mem0 operates as both an open-source self-hostable Python/TypeScript library and a managed cloud platform, providing sub-100ms vector search, episodic memory extraction, and automated memory consolidation without manual prompt engineering.

Mem0 utilizes a multi-layered memory architecture comprising short-term working memory, long-term episodic memory, and semantic user preference graphs. When a user interacts with an AI agent, Mem0 automatically analyzes the conversation, extracts persistent facts, updates existing memory records, and prunes conflicting or redundant information. Under the hood, Mem0 integrates with leading vector databases including Qdrant, Pinecone, Chroma, and pgvector. When queried, it retrieves only the most relevant memories for the current context, minimizing token consumption while maximizing personalization accuracy. Mem0 supports multi-agent shared memory, allowing a team of specialized agents (such as a research agent and a coding agent) to collaborate with synchronized awareness of the user's ongoing project state.