ZeroMemory

ZeroMemory gives AI agents persistent cognitive memory. Three memory tiers — working, episodic, and semantic — with automatic consolidation, decay scoring, and GraphRAG hybrid retrieval.

Why ZeroMemory

Problem	Solution
Agents forget between sessions	Persistent memory across sessions
Flat vector search misses connections	GraphRAG combines vectors + knowledge graph
No way to prioritize memories	Importance-weighted decay scoring
Manual memory management	Automatic consolidation between tiers
Entity relationships lost	Auto-extraction into knowledge graph

Memory Tiers

Tier	Purpose	Lifespan
Working	Active task context	Hours
Episodic	Past interactions and events	Days-weeks
Semantic	Long-term knowledge and facts	Permanent

Memories automatically consolidate from working → episodic → semantic based on access patterns and importance scores.

Quick Start

1. Store a memory

curl -X POST https://api.ainative.studio/api/v1/public/memory/v2/remember \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "User prefers dark mode and uses Python for backend work",
    "metadata": {"user_id": "u_123", "source": "onboarding"}
  }'

2. Recall by meaning

curl -X POST https://api.ainative.studio/api/v1/public/memory/v2/recall \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "What does the user prefer?",
    "user_id": "u_123",
    "limit": 5
  }'

3. Forget

curl -X POST https://api.ainative.studio/api/v1/public/memory/v2/forget \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "user_id": "u_123"
  }'

Core Endpoints

Method	Path	Description
POST	/memory/v2/remember	Store a memory
POST	/memory/v2/recall	Semantic search across memories
POST	/memory/v2/forget	Delete memories
POST	/memory/v2/reflect	Agent self-reflection
GET	/memory/v2/profile	Build user profile from memories
POST	/memory/v2/relate	Find entity relationships
POST	/memory/v2/process	Batch process memories

Scoring Algorithm

Memories are ranked using blended scoring:

final_score = (similarity × 0.5) + (importance × 0.3) + (recency × 0.2)

• Similarity — Cosine distance between query and memory embeddings
• Importance — Assigned at storage time, increases with access
• Recency — Decays over time, refreshes on access

Benchmarks

ZeroMemory achieved 100% Recall@1 and 94% QA accuracy on the LongMemEval benchmark (ICLR 2025):

System	Recall@1	QA Accuracy
ZeroMemory	100%	94%
GPT-4o (full context)	~70%	~70%
Letta/MemGPT	40%	—
Mem0	25%	—

Next Steps

• GraphRAG — Hybrid vector + knowledge graph search
• Context Graph API — Entity management and traversal
• MCP Memory Server — 6-tool MCP integration