richmondalake/memorizz/memory-transfer

Export and import memory

MemoRizz archives use format: memorizz.memory, version: 1, and the suffix .memorizz.json. They contain the 13 canonical taxonomy stores and a SHA-256 integrity check. Records retain their identity, scope, timestamps and explicit relationships. Available evolution/context records can travel with them.

Project archives

Use the project's memory ID from the session context (or memorizz plugin memory-id). For local file operations use:

memorizz plugin export-memory ./project-memory.memorizz.json
memorizz plugin import-memory ./project-memory.memorizz.json
memorizz plugin import-memory ./project-memory.memorizz.json --apply

The import command previews by default. It maps a single project namespace to the current project. --preserve-namespace retains the source project instead. --new-ids makes a separate copy and rewrites internal links. Existing records stop the restore; --conflict skip retains them, and --conflict replace replaces them when that behavior is authorized by the user. Do not add --overwrite to export without authorization to replace that file.

If the user asks to import, validate/preview the file and apply the authorized restore when the preview succeeds. Do not request another approval for a restore they already requested. Explain conflicts and only apply the selected conflict policy. Report actual committed/skipped counts and any partial errors.

MCP / hosted memory

memorizz_export_memories(memory_id=PROJECT_ID) returns an inline archive. memorizz_import_memories(archive=ARCHIVE, dry_run=true) previews it; set dry_run=false to apply. Tools enforce the caller's tenant and server write policy. Remote transports exclude global agent/tool/shared configuration. The plugin CLI also supports its configured hosted server. Inline archives must fit the server request-size limit (4 MiB by default); use the local CLI file interface for larger archives. Do not print the full archive into chat.

Agents and whole stores

memorizz memory export ./agent.memorizz.json --agent-id AGENT_ID
memorizz memory import ./agent.memorizz.json --new-ids
memorizz memory import ./agent.memorizz.json --new-ids --apply

Agent exports include configured delegates and their separate namespaces by default. --no-delegates, --no-history, --no-context, and --types narrow an export. The UI has Memory types → Export & import, also linked from Playground and Memory evolution. The SDK uses MemoryArchive(provider).

Limits

Archives omit credential-shaped fields, runtime objects and embeddings; ordinary memory text and captured prompts are included. Imported agent automations start disabled; no agent/tool is executed during restore. Semantic search requires regenerating embeddings (memorizz memory import --reembed, which may call the configured embedding service). Notion projections remain pending until explicitly reindexed. Historical/deleted values cannot be reconstructed from metadata-only history. External attachments and harness SQLite runs are not included. Writes can overlap export; restore is not atomic. See https://richmondalake.github.io/memorizz/guides/memory-transfer/.