Claude Memory Setup

In response to a question about building persistent memory in Claude, the answers generally converged on not overcomplicating it and starting with a simple file-based workflow.

• Several comments noted that Claude's built-in Memory + custom instructions alone work well enough, especially for cases like a work account where you want to reduce repetitive explanations.
• The most commonly mentioned approach is the CLAUDE.md + local Markdown log combination. Identity/operating rules go into CLAUDE.md, while session summaries are kept in daily logs or per-project files so the next session can read them again.
• Some mentioned attaching tools like Obsidian, Notion, Trello, or SQLite DB to separately manage per-project state and history.
• The key warning is don't cram everything into one file. As the memory file grows it eats up context, so raw logs and a compressed maintained version should be kept separate, with older content archived periodically.
• Ideas like llm-wiki are interesting, but the general sentiment in the comments is that a transparent file system approach usable right now is more practical.
• Many opinions suggested that for developers, a lightweight running loop — things like /init, a per-project CLAUDE.md, and automated session logging — is the most reasonable starting point.