What is bhd?

BHD Memory is a personal memory and knowledge base system that runs locally. Its main purpose is to store things you want to remember across AI coding sessions, documents, and notes, and then let you retrieve them later through search or by hooking into tools like Claude Code and Codex. The system has two main ways to add content. The first is scanning conversation transcripts from AI coding tools (Claude Code, Codex, or JSONL files), archiving those sessions, and automatically extracting memories that seem worth keeping long-term. The second is uploading documents directly: plain text, URLs, Markdown files, HTML, PDF, Word documents, PowerPoint files, and spreadsheets are all supported. Uploaded content is parsed and broken into chunks for storage. Storage uses two databases working together. SQLite holds the structured facts, evidence records, and a log of changes. Qdrant (a vector search engine that runs on your own machine) holds numeric representations of the content for similarity-based search. When you search, the system queries both and combines the results. Memories can be in different states: active, pending review, paused, archived, or deleted. Low-confidence items, sensitive information, and conflicting memories go into a review queue for manual inspection before they become active. The system exposes a REST API and includes a web management interface built with React and Vite, accessible at localhost:8767. There is also a command-line tool for uploading content, writing memories, and querying the system directly. Background ingestion jobs run through a queue backed by SQLite, so large uploads process asynchronously without blocking the interface. A graph layer tracks entities, relationships between them, and how those relationships change over time, which is compatible with the Graphiti graph memory format. The README is written in Chinese.