| name | nowledge-mem |
| description | Use Nowledge Mem from CodeBuddy for startup context, memory and thread search, durable saves, and CodeBuddy transcript import. |
Nowledge Mem is the user's cross-tool memory. Use it to resume with current context, recall exact prior work, save durable knowledge, and make CodeBuddy sessions searchable elsewhere.
Startup Context
Read Context Bundle at the beginning of meaningful work or when resuming:
nmem --json context --source-app codebuddy
If an older CLI rejects context, use nmem --json wm read. Do not read both unless the user asks.
Recall
Search when the user references prior work, resumes a project, investigates a regression, asks for rationale, or needs exact conversation history.
Prefer MCP:
memory_search for decisions, preferences, procedures, and learnings.
thread_search for prior conversations.
thread_fetch_messages only after identifying a relevant thread.
CLI fallback:
nmem --json m search "what to recall"
nmem --json t search "conversation to find" --source codebuddy -n 5
Save Durable Knowledge
Search before saving. Update an existing memory when the new information evolves it; add a new memory only for a durable decision, reusable procedure, stable preference, correction, or non-obvious learning.
nmem --json m search "existing concept"
nmem --json m add "content" -t "Title" --unit-type decision -l "label" -s codebuddy -i 0.8
CodeBuddy Threads
Lifecycle hooks automatically sync the active transcript. For a deliberate historical import:
nmem t sync --from codebuddy --limit 20
nmem t sync --from codebuddy --apply
Real capture runs on the CodeBuddy machine because MCP cannot read local files under $CODEBUDDY_CONFIG_DIR/projects or ~/.codebuddy/projects.
Status
nmem --json status
If status succeeds but CodeBuddy-specific commands fail, update the same CLI installation before debugging hooks or MCP.
Respect NMEM_AGENT_ID, NMEM_HOST_AGENT_ID, and NMEM_SPACE when present. source_app=codebuddy is provenance, not an AI identity.