Loading ToolBench
Local AI memory system — Node.js port of MemPalace. Verbatim storage, 4-layer memory stack, knowledge graph, MCP server. No API key required.
MemPalace provides 19 tools with consistent naming (verb_noun pattern: mempalace_*) and descriptions present for all tools. However, descriptions are often terse (20-80 chars), falling below the 50-200 char LLM-optimized baseline. Input schemas are visible and typed, but many parameters lack descriptions or have minimal ones. Output schemas are not documented. Error handling is minimal, most tools return generic error objects without recovery guidance. The server demonstrates solid foundational structure but lacks the depth needed for production-grade agent integration.
Add a new memory drawer (verbatim content) to the palace
Check if content is a duplicate of existing memories above similarity threshold
Delete a memory drawer from the palace
Read diary entries with optional filtering
Record a diary entry about what happened, what was learned, what matters
Find connections (tunnels) between two wings or all wings
Get AAAK dialect specification for structured memory format
Get complete taxonomy of wings and rooms with drawer counts
Output schemas not documented. Tools return structured objects (wings, rooms, taxonomy) but LLMs cannot see the expected response shape. This forces agents to infer field names and types, increasing hallucination risk.
Destructive tool (mempalace_delete_drawer) lacks confirmation/dry-run pattern. No recovery guidance in error responses. Agents can permanently delete memories without safeguards.
Error handling is generic. Tools return {error: 'No palace found', hint: '...'} but do not categorize errors as retryable, user-fixable, or fatal. Agents cannot determine recovery strategy.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2026-07-28+ | v2 |
Get statistics about the palace knowledge graph
Add a fact (triple) to the knowledge graph
Mark a fact (triple) as invalid/ended in the knowledge graph
Query knowledge graph for facts about an entity
Get statistics about the knowledge graph
Get temporal evolution of facts about an entity
List rooms (named ideas) in a wing with drawer counts
List all wings (topic categories) in the palace with drawer counts
Semantic search across memories with optional wing/room filtering
Get palace overview including total drawers, wings, rooms, palace path, protocol, and AAAK dialect specification
Traverse palace graph starting from a room with BFS up to max hops
Parameter descriptions are minimal (20-60 chars). Many lack format/constraint details. E.g., 'threshold' in mempalace_check_duplicate has no range (0-1? 0-100?). 'max_hops' in mempalace_traverse lacks bounds.
No pagination support. Tools like mempalace_search and mempalace_kg_query accept 'limit' but do not return total_count or next_cursor. Large result sets risk context window exhaustion.