Memory & persistence
Alfred Memory stores the decisions, conventions, and lessons your agents accumulate. Its server runs only on your computer; the managed hosted service is retired.
Memory is independently usable. Workspace and Agent Harness are optional callers, connected only through explicit endpoint, credentials, and scope. Selecting them in setup does not create a connection.
Local process, chosen storage
The HTTP server accepts only loopback addresses. SQLite works for coding tenants; user-chosen Postgres storage remains available, including a remote database. A remote database does not make the HTTP service remote. The compatibility profiles web and both still require Postgres and use local clients.
Workspaces & tenants
A Memory filesystem workspace is a directory binding, distinct from the Workspace application. Existing registry IDs and tables keep their meanings. A tenant is a memory partition. Workspaces are not tenants: tenant_access is many-to-many with explicit inheritance.
Start locally
alfred memory init --profile=coding --backend=sqlite --name my-memory ALFRED_MEMORY_REQUIRE_AUTH=true alfred memory serve
Namespaced commands are available in source and require a release after 1.1.0. Existing alfred init and alfred serve aliases remain supported on 1.1.0 and later. Standalone alfred-memory uses the same Memory entrypoint in releases with the composition layer.
Save the API key printed by init. Keep the server running and configure local clients with your loopback endpoint and that key.
Local console
Open the local console and unlock it with your key. The console and API share an origin, including custom ports. Use alfred memory serve --port=4321 and open http://127.0.0.1:4321/console/ to select a different port.
The public website contains instructions only. Cloud ChatGPT Actions and Google AI Studio cannot reach the local service.
Existing data
Existing tenants, API keys, OAuth, MCP, SQLite and Postgres data backends remain supported. No database is deleted as part of hosted retirement. Keep independent backups and use the local API or console to manage your data.