Back to Home
ODAM Documentation
API Reference & Guides
ODAM Architecture
System architecture and technical implementation overview
High-Level Overview
🏗️
Hybrid Architecture
ODAM combines graph databases, vector search, and intelligent caching to achieve 95% personalization accuracy with 700ms response times.
Data Flow
1
Message Reception
User message received via REST API
2
Memory Lookup
Redis cache check, then CosmosDB graph query
3
AI Processing
Azure OpenAI processes message with memory context
4
Memory Update
New entities extracted and stored in graph
5
Response Generation
Personalized response returned to user
System Components
Azure OpenAI
AI Service
LLM processing and entity extraction
CosmosDB
Graph Database
Entity storage and relationship mapping
Redis Cache
Caching Layer
High-speed memory retrieval
FastAPI
Web Framework
Async API endpoints and routing
Azure Cloud
Infrastructure
Scalable cloud hosting
Key Features
Graph-Based Memory
Entities and relationships stored in CosmosDB graph structure
Hybrid Search
Combines vector similarity with graph traversal
Multi-Layer Caching
Redis cache with intelligent invalidation
Enterprise Security
Azure security with data encryption
Technical Specifications
Memory Storage
CosmosDB Graph Schema
// Entity Node
{
"id": "user_123_person_alex",
"type": "Person",
"name": "Alex",
"confidence": 98.5,
"attributes": {
"first_name": "Alex",
"mentioned_at": "2024-01-15T10:30:00Z"
},
"user_id": "user_123",
"session_ids": ["session_abc", "session_xyz"]
}
// Relationship Edge
{
"id": "rel_123",
"from": "user_123_person_alex",
"to": "user_123_location_kyiv",
"relationship": "LIVES_IN",
"confidence": 95.2,
"created_at": "2024-01-15T10:30:00Z"
}Performance Metrics
| Metric | ODAM | Traditional | Improvement |
|---|---|---|---|
| Personalization Level | 95% | 74% | +21% |
| Response Time | 700ms | 15s | 20x быстрее |
| Memory Utilization | 100% | 60% | +40% |
| Fallback Rate | 0% | 25% | -25% |
Scalability & Infrastructure
Horizontal Scaling
- • Azure Container Instances for API servers
- • CosmosDB auto-scaling based on demand
- • Redis cluster for distributed caching
- • Load balancer with health checks
Capacity Limits
- • 10M+ entities per user supported
- • 1000+ concurrent requests/second
- • 99.9% uptime SLA guarantee
- • Sub-second cache retrieval
Security & Privacy
Data Privacy
All user data is encrypted at rest and in transit. ODAM never shares or uses personal data for training. Each user's memory is completely isolated.
Encryption
- • TLS 1.3 for all API communications
- • AES-256 encryption at rest
- • Key rotation every 90 days
- • Azure Key Vault integration
Access Control
- • API key authentication
- • Rate limiting per user/IP
- • User data isolation
- • Audit logging enabled