Claude Code already has native persistence, so an Obsidian vault is optional. It is a folder of Markdown notes you can curate for human review or deliberately share with other tools. The difference matters if you want a portable archive: Claude Code’s own auto memory is tied to a repository and machine, while a vault gives you a separate set of notes to organize and maintain.
In this article
- What Claude Code already remembers
- What an Obsidian vault adds
- Use an index to guide retrieval
- Keep agent edits under your control
- Local files and session records
What Claude Code already remembers
Claude Code uses two complementary kinds of persistent context. You write instructions in CLAUDE.md; Claude can also create auto-memory notes based on a project and previous sessions. Both provide context rather than hard enforcement. Auto memory is stored locally for each Git repository and does not automatically travel across machines or cloud environments.
At session start, Claude Code loads the first 200 lines or 25 KB of its MEMORY.md index, whichever limit it reaches first. It can read topic files on demand. Auto memory does not save something every session, and Claude skips information it can derive from the codebase or existing CLAUDE.md files. Anthropic describes the two mechanisms and their limits in its Claude Code memory documentation.
| Approach | Storage and scope | What loads at startup | Best fit |
| Claude Code native memory | User-authored CLAUDE.md instructions plus Claude-written auto memory, stored per repository on the local machine | CLAUDE.md context and the first 200 lines or 25 KB of the MEMORY.md index, whichever comes first; topic files can be read on demand | Project instructions and learned context for Claude Code on a particular machine |
| Obsidian vault | A folder of Markdown notes that can be organized for people and made accessible to an agent | The vault does not load in full just because it exists; an agent needs access and direction to relevant notes | A human-curated archive you can organize or deliberately share across tools |
What an Obsidian vault adds
A vault gives you a readable place to organize project knowledge as Markdown files, with links between notes and, if useful, folders, tags, or metadata. A workflow described by Journal du Net has Claude Code read and edit those local files directly, without relying on Obsidian’s interface. That makes the vault an optional archive alongside Claude Code’s own memory, not a prerequisite for it.
The graph view is a visual map for you. Claude Code still needs access to the files and a useful route to the notes relevant to its task; seeing connected dots in Obsidian does not by itself select or retrieve those notes. One Claude Code user reported using a one-line index to help an agent choose a few notes from a vault of about 70. That is an individual account, but it points to the practical value of a clear index over expecting the whole vault to be available at once.
Use an index to guide retrieval
Start with a short index that names the topics and points to the relevant notes. Keep each note focused, and use links or tags to connect related material. Journal du Net’s described workflow tells Claude Code to consult an index and follow links to notes relevant to the task.
Keep the index current as decisions change. One Obsidian user described rolling session notes into higher-level summaries and pruning outdated or reversed decisions. For conflicting project notes, ask Claude to surface the notes that disagree so you can review them, rather than choosing one silently. An index can make useful material easier to locate, but it cannot ensure that Claude will retrieve every relevant note or resolve contradictions correctly.
Keep agent edits under your control
A setup prompt can ask Claude to inventory a vault, propose changes, and wait for approval before editing or deleting files. Those instructions are guidance. Anthropic says CLAUDE.md is not an enforcement mechanism; use Claude Code settings or hooks for controls that must apply regardless of the model’s choice.
Local files and session records
A vault stored on your computer can still have content recorded when Claude Code reads it. Anthropic says content passed through Claude Code tools is written to local session transcripts in plaintext, with operating-system file permissions as the protection it describes. A separate route for remote agents adds other data paths: a workflow described by Journal du Net uses GitHub or Obsidian Sync with a community MCP server and tunnel to expose notes remotely.
