Nowledge Mem
Use the enabled nowledge-mem-connection MCP tools for the server configured in Plugins >
Nowledge Mem. Keep that destination, returned identity and Space for the whole operation. Never mix results from
different Mem servers or silently retry on another server.
At the beginning of related work, read Context Bundle once (read_context_bundle, source_app
deepchat) for identity, scope, rules and Working Memory. Use read_working_memory only when the
bundle tool is unavailable. An empty Working Memory is a successful empty result.
For continuation, reviews, regressions and prior decisions, search memory_search with a narrow
query and limit 5, then fetch the relevant exact IDs. Search before adding a duplicate. Distinguish
recorded facts from proposals, and keep scope returned by the server. A graph is optional; only
show one when it enforces the same identity and Space as retrieval.
Save a concise durable learning with memory_add or update an existing memory only when it is
useful beyond this conversation. State what was saved. Do not save speculative conclusions as
facts. Save complete conversations only when the user requests it: use DeepChat's conversation
export to Nowledge Mem, which uses the same verified connection as MCP. Do not
claim an export happened without a successful acknowledgement.
For configuration or failed access, direct the user to Plugins > Nowledge Mem. The host verifies
REST and MCP and stores credentials securely. Report the failing connection and operation; do not
switch to localhost after a remote error. Do not ask for keys in chat, invoke unscoped nmem, edit
~/.nowledge-mem/config.json, read local Mem database files, or install another connector. This
plugin uses DeepChat's HTTP client and does not require the CLI.