Claude DB: Provide Claude Code with a fully local long-term memory, where decisions, dead ends and code maps are stored in your own SQLiteClaude DB：给 Claude Code 装一个完全本地的长期记忆库，决策、死胡同和代码地图都存进你自己的 SQLite

Common problem of all programming agents: as soon as the context window is cleared, the working memory returns to zero. You decided last week to "use Zustand instead of Redux", stepped on the dead end of "this function can not be called concurrently", and reset the next session.

Claude DB's approach is two-pronged: use lifecycle hooks to silently capture decisions, dead ends, and rules at the end of a session; and scan your codebase to build a symbolic map of who defines, who calls, and who imports, supported by TypeScript, JavaScript, Python, Go, Rust, Java, and Kotlin.

The key restraint lies in the "recall on demand": each new question is only injected with one or two matching historical memories, without a match, nothing will be injected, and the context will not be burst.

/catchup: Come back after a few days and ask “Where did I do it last time?” and it answers you with a memory bank.

/handoff: Leave a handoff note for the next conversation (or co-worker who takes over) without losing key conclusions.

/cdb-scan: Sweep an existing code base into the memory map as a whole, and connect to the old project for the first time.

claude-db status/doctor: Always check if the hook is connected, what is recorded, and where there is a problem with the configuration.

Node 22.16 or higher is required and is used within the git repository.

Three steps: npm install -g claude-db install → cd globally Your project claude-db → install --project Write the hook into the project configuration and restart the Claude Code. After that, the capture and recall will be fully automatic without manual intervention.

If you take over the stock project, first run claude-db flush to empty the demo data, and then run claude-db scan (or/cdb-scan) to build the existing code map.

The download button provides the Claude DB official repository source code package (zip, 0.78 MB), which contains the complete plugin source code, hook scripts and documentation. You can install it yourself after offline review, or you can install it directly according to the above npm command.

The data is stored natively in SQLite by default; the team wants to share memory, cut the storage backend to Postgres or MongoDB, and still host it themselves.

This article is compiled from public sources and ships with the matching resource. Product features and pricing are subject to the official page. Resources are for learning and exchange only — please respect the original license.