Persistent AI memory for coding agents. Browse, search, promote, and manage short-term and long-term memories right from VS Code.
iGenius Memory provides 10 memory-management tools with consistent naming conventions and reasonable parameter schemas. All tools follow verb_noun naming (memory_briefing, memory_ingest, etc.) and have descriptions. However, descriptions are generally brief (averaging ~90 chars), parameter descriptions are sparse or missing in several cases, and output schemas are entirely undocumented. Error handling guidance is absent. The tools are cohesive (all memory-focused) but lack the richness expected of production-grade tools. No destructive operation confirmations, no pagination support, and no guidance on when to call which tool.
Get a consolidated intelligence briefing from all previous sessions containing decisions, preferences, context, and open threads
Merge all interaction extracts into a master briefing, typically done before a context reset
Delete a memory by ID
Ingest a message (user or agent) into the memory system for automatic extraction of facts, decisions, and context
Promote a short-term memory to long-term persistent storage
Get raw persistent interaction extracts from the memory system
Output schemas completely undocumented. No tool declares what fields it returns, forcing LLMs to guess at response structure and preventing tool composition. E.g., memory_briefing returns unknown structure; memory_search results are not documented.
Destructive operations (memory_delete) lack confirmation pattern. No dry-run, no explicit warning in description, no per-item success/failure reporting. Agent could silently delete wrong memory.
memory_consolidate has empty input schema and generic description ('Merge all interaction extracts into a master briefing, typically done before a context reset'). Unclear when to invoke this vs memory_briefing. No guidance on side effects or prerequisites.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 54 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 47 | - | v1 |
Get the review queue of short-term memories for user triage and promotion decisions
Search through all memories by query string
Directly save content to a specific memory layer (short_term, long_term, persistent, or pinned)
Update an existing memory's title, content, or importance score
Parameter descriptions are missing or minimal. memory_search accepts only 'query' with description 'The search query', no guidance on syntax, field scope, or result limit. memory_store's 'importance' parameter defaults to 50 but lacks range constraints (0-100? unbounded?).
No error guidance. Descriptions lack recovery hints. If memory_search returns no results, the agent has no guidance on fallback strategies. If memory_promote fails (memory already long-term?), no actionable error message documented.
No pagination support on list-like operations (memory_review, memory_recall). If user has 1000 memories, these tools will either explode the token count or truncate silently.
Tool selection guidance missing. Descriptions do not explain when to call memory_briefing vs memory_recall vs memory_consolidate. Overlapping names create ambiguity.
memory_store's 'layer' enum values ('short_term', 'long_term', 'persistent', 'pinned') are not explained in the parameter description. What is the difference? When should each be used? Lacks actionable guidance.