Graph-based memory system for AI agents. Store people, places, organizations as nodes. Build semantic relationships (KNOWS, WORKS_AT, CREATED). Hybrid keyword + semantic search finds partial matches and close variants. Filter by date, traverse relationships, maintain temporal context. Simple tools, LLM handles complexity. 12 tools for complete memory management.
Reverie demonstrates solid tool design with complete schemas, clear descriptions, and thoughtful parameter organization. All 12 tools have descriptions (avg 156 chars) and input schemas with typed parameters. Naming follows verb_noun convention (search_memories, create_memory, etc.). However, output schemas are not documented, LLMs cannot predict response structure for chaining. Error handling lacks recovery guidance. Some parameters accept free-form strings where enums would prevent hallucination (e.g., search_mode has enum, but label and order_by do not). Missing descriptions for relationship properties and memory properties objects. No dry-run or confirmation pattern for destructive operations (delete_memory, delete_connection), violating the confirmation-request pattern.
Create a relationship between two memories. Relationships are directional and can have properties.
Create a new memory node with a label and properties. Automatically embeds the memory for semantic search.
Delete a relationship between two memories.
Delete a memory node and all its relationships.
Consolidate and optimize the memory graph by merging duplicate nodes, removing redundant relationships, and cleaning up properties. Returns a summary of changes made.
Get help on using the memory tools effectively. Topics include connections, labels, relationships, best-practices, and examples.
List all memory labels in use and their counts. Optionally include archived memories.
Output schemas not documented. LLMs cannot predict response structure (fields, types, pagination) for search_memories, create_memory, query_memories, etc. Breaks tool chaining and forces agents to guess field names.
Destructive operations (delete_memory, delete_connection) lack dry-run or confirmation step. Agents can permanently delete data without review. No recovery guidance in error responses.
Properties parameters (create_memory.properties, create_connection.properties, update_memory.properties) lack schema definition. Described as 'Key-value properties' but no type constraints, required fields, or examples. LLMs cannot validate structure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2026-07-28+ | v2 |
Get statistics about the memory graph: node counts by label, relationship counts by type, and total memory usage.
Execute a read-only Cypher query against the memory graph. Query is validated to prevent write operations.
Search for memories using keyword, semantic, or hybrid search. Returns matching memories with optional relationship traversal.
Update properties of an existing relationship between two memories.
Update properties of an existing memory node. Automatically re-embeds the memory.
Parameters accept free-form strings where enums would prevent hallucination: label (should enumerate person, organization, place, project, event, skill, etc.), order_by (should enumerate created_at, updated_at, name, etc.). Guidance tool documents these but tool schemas do not enforce them.
Error handling does not guide recovery. No indication of retryable vs. fatal errors, no suggestions for next steps (e.g., 'Memory not found. Try search_memories() first'). Agents cannot self-correct.