Memory Backup & Migration (export / import)

claude-mem-go's export/import subcommands are its only backup and cross-backend migration story — there's no other way to get data out of the store, or move it between the SQLite and Postgres backends.

When to use

• "Back up my memories" / "export the database" → export
• "Restore from a backup" / "import that file" → import
• "Move my memories to Postgres" (or from Postgres to SQLite) → export from the source backend, import into the destination — the same file works for both, since both backends implement the same interface.

export

Writes every observation as JSON Lines (one per line) to a file.

"$CLAUDE_PLUGIN_ROOT/claude-mem-go" export -out backup.jsonl

Add -db <postgres DSN> to export from the Postgres backend instead of the default SQLite file. No -project filter exists on export — it's always everything, since a partial backup is a worse default than a complete one.

import

Reads a file export wrote and re-inserts every row.

"$CLAUDE_PLUGIN_ROOT/claude-mem-go" import -in backup.jsonl

Add -db <postgres DSN> to import into the Postgres backend (the migration path). This is safe to run more than once, or against a database that already has some of the data: each row keeps its original content-hash, so an already-present row is skipped, not duplicated — tell the user this if they're worried about re-running it. Each row's embedding (if it had one) travels with it too, so a SQLite→Postgres migration doesn't leave semantic search with nothing to search — the user shouldn't need to re-embed anything just because they moved backends. If mem-doctor reports embedding_dims_consistent=false afterward (e.g. the source and destination were configured with different embedding models at some point), that's a mem-reembed job, not an export/import problem.

(Fall back to claude-mem-go export/import on PATH if $CLAUDE_PLUGIN_ROOT isn't set — see mem-doctor's skill for why.)

Not destructive, but be clear about direction

Neither command deletes anything from the source — export only reads, and import only adds. If the user's actual goal involves also deleting old data (freeing up space, not just backing it up), that's a separate, genuinely destructive step — see the mem-prune skill, and don't combine the two without the user explicitly asking for both.