Giving Claude and Codex a shared memory that outlives the session

What it does

MemoraX Code hooks into Claude Code and Codex to capture four kinds of context—coding lessons, repository structure, personal preferences, and reusable procedures—and injects the relevant bits back into later sessions. It runs as a Node.js package that integrates via client hooks and a bundled skill command, quietly writing memory in the background while allowing manual retrieval through a local viewer. Without an API key and explicit activation, the memory features stay unavailable.

The interesting bit

Instead of stuffing entire chat histories into the context window, the system tries to surface only the small slice of memory relevant to the current task. It also deliberately avoids uploading full conversation traces, which is a refreshing bit of restraint in an ecosystem that usually equates “memory” with “more tokens.”

Key highlights

• Four distinct memory boundaries: Coding Memory (verified fixes and pitfalls), Repo Memory (architecture and ownership), Personal Memory (tone and format preferences), and Procedure Memory (reusable checklists and validation gates).
• Background writeback extracts lessons from completed turns without manual note-taking, while a local Memory Viewer at 127.0.0.1:8787 tracks activity counts and status.
• Hook integration means agents can trigger retrieval and reminders natively via $memorax-code in Codex or /memorax-code in Claude Code.
• Cloud data scope is narrowly scoped: MemoraX receives selected queries and agent responses for extraction, but not raw client trace artifacts or model-provider credentials.

Caveats

• Requires Node.js 24 or newer and Python 3 for repository memory operations, so older environments are out of luck.
• Core search, retrieval, and writeback depend on the MemoraX cloud service and an API key; the local installation alone does nothing useful.
• The README notes there is “no second writeback confirmation,” which means automatic background extraction happens silently once enabled.

Verdict

Worth a look if you are already living inside Claude Code or Codex and tired of re-explaining your codebase’s quirks every morning. Skip it if you need fully offline, self-hosted memory or if your stack is pinned to older Node versions.