> ## Documentation Index > Fetch the complete documentation index at: https://docs.claude-mem.ai/llms.txt > Use this file to discover all available pages before exploring further. # Obsidian > Copy selected project memories into a chosen vault without replacing your notes. Export selected Claude-Mem observations as Markdown notes in an existing Obsidian vault. This is an explicit, one-way export: it does not scan or import the vault, capture conversations, install a plugin, or run in the background. It uses the local worker's existing selected-observation API. Obsidian stores notes as [local Markdown files](https://help.obsidian.md/data-storage), so no community plugin, Obsidian API token or CMEM Pro subscription is needed. You choose both the source project and the observation IDs. A note in a different project cannot be selected by passing its ID. ## Requirements Use a Claude-Mem source checkout with its dependencies installed, Bun 1.4.4 and a running local worker. This repository script is not a command shipped inside the installed `claude-mem` package. The worker and filesystem checks were verified on macOS arm64 with Claude-Mem 15.18.0 and Node.js 25.1.0. Other operating systems have not been verified here. Choose an existing vault with the default `.obsidian` configuration directory. The export folder must be a regular directory, not a symlink, and the vault's filesystem must support hard links for atomic, non-overwriting publication. ## Export selected notes Use [mem-search](https://docs.claude-mem.ai/usage/search-tools.md) to find relevant observation IDs in the chosen project. From the source checkout, run this with your selected IDs, replacing the example values: ```bash theme={null} bun scripts/export-obsidian.ts \ --vault '/absolute/path/to/My Vault' \ --source laptop \ --project checkout-service \ --ids 12,34 ``` `--vault` must be an absolute path. `--project` is one literal project key, including any commas or spaces; the exporter does not widen it to worktree parents or adopted projects. Select between one and 100 IDs per invocation. `--source` is a stable label you choose for this worker installation, using lowercase letters, digits, underscores or hyphens. Use a different label for another database: observation IDs are local, so two computers can both have a memory numbered `12`. Keep the same label when rerunning an export from the same installation. It does not select a worker or account. The script uses your existing `CLAUDE_MEM_DATA_DIR`, worker URL, host and port configuration. Pass the same settings used to start that worker. It does not start or reconfigure the worker automatically. Requests have a ten-second deadline and are never retried automatically. The output reports created and skipped IDs: ```json theme={null} {"created":[12,34],"skipped":[]} ``` The files are `Claude-Mem/laptop/12.md` and `Claude-Mem/laptop/34.md` beneath the selected vault. Open them from Obsidian's file explorer. Each contains a quoted title, source, project and original observation ID in frontmatter, followed by the selected observation's readable prose. Raw tool bodies, internal record metadata and entire sessions are not exported. The prose is inside a literal text block so Markdown image links, HTML and Obsidian embeds from retrieved memory do not activate on opening the note. Memory remains reference material, not instructions to run tools. If the standard prose projection would shorten a record, the exporter refuses it instead of silently losing text; use [JSON memory export](https://docs.claude-mem.ai/usage/export-import.md) for that record. ## Preserve edits and handle retries An existing destination is always skipped, including its frontmatter and any edits you made in Obsidian. The script never updates or removes a prior note, even if the source memory later changes. Each new file is completed before being published without overwriting its destination. Repeating the command therefore creates no duplicate notes and preserves unrelated vault files. Missing IDs, a wrong project or malformed worker results are rejected before any note is written. A filesystem failure can leave earlier completed notes; inspect the vault and rerun the same selection after resolving it. Existing destinations will be skipped. A deleted note is recreated only if you explicitly select that ID in a later export; nothing watches for deletions or silently imports notes back into Claude-Mem. ## Privacy and disconnect Exported notes are plaintext. Only the chosen records are copied, but a vault's Obsidian Sync or another synchronization service may then copy them elsewhere. Choose the destination deliberately and review notes before sharing the vault. No vault content, user edits or frontmatter is sent to Claude-Mem. There is no persistent connection to uninstall. Stop running the script to disconnect. Keep or remove exported notes yourself in Obsidian; removing the script leaves your vault and saved Claude-Mem memories intact. ## Verify and troubleshoot ```bash theme={null} bun test tests/scripts/export-obsidian.test.ts ``` The tests cover project/ID scope, repeated exports, frontmatter preservation, inert embeds, traversal and symlink rejection, and refusal to shorten content. For a real-worker check, save a harmless note in a disposable project, export its acknowledged ID, open it in a test vault, add a personal property, and repeat the same export. The second result should list that ID under `skipped` and leave the file unchanged. If an ID is not returned, check its project and that it still exists. If the worker cannot be reached, check its status and matching connection settings. For a symlink or filesystem error, choose a regular writable export directory on a filesystem with hard-link support; the script does not fall back to overwriting notes. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.