A Memory is durable, account-level context. It can be recalled across Agents
and runs, but it is not shared through Projects. See Core
concepts for the canonical definition.
Write a useful Memory
Keep one reusable fact, preference, pattern, decision, or piece of context in
each Memory. Include the subject instead of relying on the current
conversation.
Valid categories are fact, preference, pattern, decision, and
context. If you omit --category, the CLI uses fact.
The search returns the new Memory. A supported Agent with working Clawdi
tools can recall it in later conversations.
Never store passwords, API keys, private keys, tokens, or connection strings
as Memories. Clawdi blocks likely secrets at this boundary; store them in a
Vault instead. Add and find Memories in the dashboard
Open Memories in the Clawdi dashboard:
- Choose New memory.
- Enter concise content and select a category.
- Save it, then use search or a category filter to find it again.
Open a Memory to inspect its source, creation time, and recall count when that
information is available.
First make sure the supported Session is synced, then identify its local
Session ID:
Extraction asks the service’s configured model to create Memories from that
Session. Some deployments do not configure a memory-extraction model; in that
case the CLI reports that extraction is unavailable and makes no Memories.
Review extracted Memories in the dashboard and delete anything too broad,
duplicated, sensitive, or no longer accurate.
Delete outdated context
The shortest path is the delete action on the Memory card or detail page. For
the CLI, obtain the full ID from JSON output and remove that exact record:
Deletion removes the Memory from future recall. If the underlying preference
or decision changed, add a new, explicit Memory after removing the obsolete
one.
If an Agent cannot recall a known Memory, verify its Clawdi tools with
Connectors and MCP and see the MCP
troubleshooting steps.
Run clawdi memory --help for the command surface shipped with your CLI.