A local-first memory system with semantic search, knowledge graph, and diary capabilities. Provides MCP tools for managing tagged entries, querying knowledge graphs, and maintaining agent diaries.
22 tools with consistent verb_noun naming (tagmem_*). All have descriptions (avg ~80 chars, within baseline 34-392). Input schemas present for all tools with proper type definitions. However, descriptions are terse and lack LLM-optimized guidance on WHEN to use each tool vs alternatives. Parameter descriptions are minimal (e.g., 'Filter by tag' lacks format/constraint details). Output schemas are not documented, LLMs cannot predict response structure. Error handling and recovery guidance absent. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels. Composition is sound (single responsibility per tool), but chaining guidance missing.
Add a new entry to the memory system with title, body, tags, and metadata
Check for duplicate or similar entries using semantic similarity
Delete an entry by ID
Read recent diary entries for an agent
Write an entry to an agent's diary with optional topic
Run diagnostic checks on the embedding provider and system health
Find bridge entries connecting two depth levels in the tag graph
Output schemas not documented. LLMs cannot predict response structure for any of the 22 tools. This forces agents to guess field names and types, increasing hallucination risk and breaking downstream tool chaining.
Descriptions lack LLM-optimized guidance. Most descriptions are 50-70 chars and state WHAT the tool does but not WHEN to use it vs similar tools (e.g., tagmem_search vs tagmem_list_entries, tagmem_kg_query vs tagmem_graph_traverse). Baseline is 194 chars with context.
Parameter descriptions lack format/constraint details. E.g., 'Filter by tag' (tagmem_list_entries) does not explain if tags are case-sensitive, comma-separated, or single-value. 'Similarity threshold (0-1)' (tagmem_check_duplicate) lacks guidance on typical values or what threshold means semantically.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | 2026-07-28+ | v2 |
Get hierarchical mapping of tags by depth level
Get statistics about the tag graph including node and edge counts
Traverse the tag graph starting from a tag with optional hop limit
Add a fact to the knowledge graph with subject, predicate, and object
Invalidate a fact in the knowledge graph by marking it as ended
Query the knowledge graph for facts about an entity
Get statistics about the knowledge graph
Get temporal timeline of facts for an entity
List all depth levels and their entry counts
List entries with optional filtering by tag and depth, with pagination
List all tags and their usage counts
Get system paths for data, config, cache, store, and index directories
Semantic search across entries using vector embeddings
Retrieve full details of a specific entry by ID
Get system status including total entries, depth counts, tag statistics, and storage paths
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Risk levels are declared in the spec (READ_ONLY, WRITE, DESTRUCTIVE) but not exposed to the LLM via tool metadata. Agents cannot reason about safety or retry semantics.
No error handling or recovery guidance. Tools do not document what errors are possible, whether they are retryable, or what the LLM should do next. E.g., tagmem_delete_entry offers no confirmation or dry-run pattern.