From kagura-memory
Searches Kagura Memory Cloud for past knowledge and patterns using semantic, keyword, or hybrid search with optional filters, reranking, and explore hints.
How this command is triggered — by the user, by Claude, or both
Slash command
/kagura-memory:recallThe summary Claude sees in its command listing — used to decide when to auto-load this command
Search Kagura Memory Cloud for relevant past knowledge and patterns. Use the Kagura Memory Cloud MCP tools to search for: $ARGUMENTS ## Steps ### 1. Resolve the target context If only one context exists, use it. If multiple, pick the one most relevant to the current project. If unclear, ask the user. ### 2. Search Use `recall` with the resolved context_id: **Search modes** — choose based on the query type: - `hybrid` (default): Best for most queries — combines semantic understanding with keyword matching - `semantic`: Use when you know the concept but not the exact words (e.g., "...
Search Kagura Memory Cloud for relevant past knowledge and patterns.
Use the Kagura Memory Cloud MCP tools to search for: $ARGUMENTS
list_contexts()
If only one context exists, use it. If multiple, pick the one most relevant to the current project. If unclear, ask the user.
Use recall with the resolved context_id:
recall(context_id=..., query="$ARGUMENTS", k=10)
Search modes — choose based on the query type:
hybrid (default): Best for most queries — combines semantic understanding with keyword matchingsemantic: Use when you know the concept but not the exact words (e.g., "how we handled auth token expiry")keyword: Use for exact term matching, hiragana queries, or when semantic returns noiserecall(context_id=..., query="$ARGUMENTS", k=10, search_mode="keyword")
Reranking — enable for higher-quality results when the user has a reranker configured:
recall(context_id=..., query="$ARGUMENTS", k=10, use_rerank=true)
Explore hints — get suggestions for follow-up explore() calls to discover related memories via the knowledge graph:
recall(context_id=..., query="auth token handling", k=10, include_explore_hints=true)
When enabled, the response includes up to 3 explore_hints — each with a memory_id and a reason (top_result, high_centrality, or unexplored_neighbor). Use the suggested memory_id as a seed for explore() to discover related knowledge beyond keyword/semantic matching.
Filters — narrow results by type, tags, importance, date, or source:
recall(context_id=..., query="...", k=10, filters={"type": "decision"})
recall(context_id=..., query="...", k=10, filters={"tags": ["python", "fastapi"]})
recall(context_id=..., query="...", k=10, filters={"tags": ["python", "fastapi"], "tags_match": "all"})
recall(context_id=..., query="...", k=10, filters={"importance": {"gte": 0.8}})
recall(context_id=..., query="...", k=10, filters={"created_after": "2026-01-01T00:00:00Z"})
recall(context_id=..., query="...", k=10, filters={"source_uri_prefix": "vault://my-vault/"})
recall(context_id=..., query="...", k=10, filters={"source_type": "file"})
tags: matches ANY of the listed tags by default (OR). Add "tags_match": "all" to require all tags (AND). Before building a tag filter, call list_tags(context_id=...) to discover the actual tag spellings in this context — troubleshoot vs troubleshooting drift silently kills tag filters.source_uri_prefix: Filter by origin URI prefix (e.g. "file://", "vault://my-vault/"). Useful for querying memories from a specific vault or directory.source_type: Filter by origin type — "file" | "url" | "vault" | "api" | "manual".Cross-context search — to query 2-20 contexts at once, use context_ids instead of context_id:
recall(context_ids=["<uuid-1>", "<uuid-2>"], query="...", k=10)
All listed contexts must:
workspace_mismatch)context_privacy_mismatch)embedding_model_mismatch)Filters can be combined:
recall(context_id=..., query="...", k=10, filters={"type": "bug-fix", "tags": ["auth"], "created_after": "2026-03-01T00:00:00Z"})
recall(context_id=..., query="...", k=10, filters={"source_uri_prefix": "vault://", "source_type": "vault", "importance": {"gte": 0.7}})
Show results in a table: memory_id, summary, type, importance, tags.
reference for detailed content on the most relevant match:
reference(memory_id=<id_from_results>, context_id=...)
keyword if hybrid missed, or vice versalist_contexts() — the memory may be in a different contextnpx claudepluginhub kagura-ai/memory-cloud --plugin kagura-memory/recallRecall memories from past sessions via memoir. Delegates to the memory-recall skill. Pass --include-metrics to include machine-generated turn statistics (excluded by default).
/recallLoads the spectre-recall skill to search and retrieve project knowledge using the provided query.
/recallRetrieves a memory by exact key, partial key, value search, or fuzzy query from project/global .nemp/memories.json files. Logs access and updates vitality tracking.
/recallSearches session history by keyword, returning matching sessions with concept tags, Done bullets, and Decisions.
/recallSearches Obsidian vault for past documents, decisions, and troubleshooting matching query. Returns file paths, excerpts, dates, categories, and types in formatted list.