- GiperONE — Business Automation & AI Agents
- GiperMEMORY
GiperMEMORY: Customer Vector Search
Production-grade hybrid RAG system with Knowledge Graph. Multi-functional mixed memory for agents. Foundation for any AI service.
System Architecture
Technology Stack v3.0
Frontend
React 18 + TS + Vite + shadcn/ui
Backend
FastAPI + Pydantic v2
Task Queue
Celery + Redis
Database
PostgreSQL 17 + pgvector
OCR
Docling (IBM)
LLM
OpenRouter / OpenAI / Anthropic
Key Features
System Capabilities
Structured OCR
Docling extracts text, tables, heading hierarchy and bbox.
Hybrid Chunking
Smart chunking with semantic boundaries.
LLM Enrichment
MetadataAgent and EntityExtractor enrich each chunk with context.
Knowledge Graph
Entities and relationships are saved in a graph structure.
Vector Search
pgvector for fast semantic search over embeddings.
Async Processing
Celery workers for background processing of documents of any size.
Processing Pipeline
Docling-native Pipeline v3.0
Loading
STEP 01Document loading, status check and template selection.
Template
STEP 02LLM provider and chunking parameters configuration.
OCR (Docling)
STEP 03Structured OCR: text, tables and structure extraction.
LLM Setup
STEP 04Agents and embedding services initialization.
Summary
STEP 05Document summary generation for context.
Enrichment
STEP 06Parallel metadata and entity extraction.
Aggregation
STEP 07Entity and relation deduplication (Knowledge Graph).
Embedding
STEP 08Chunk content vectorization.
Saving
STEP 09Saving chunks and graph to PostgreSQL.
Finalizing
STEP 10Task completion and status update.
Database Schema
PostgreSQL 17 + pgvector
Integration
REST API Reference
{
"id": "upload-abc123...",
"filename": "contract.pdf",
"status": "pending"
}data: {
"status": "PROGRESS",
"step": "enrichment",
"progress": 65,
"step_label": "Metadata enrichment"
}Deployment
Docker Compose & Environment
Services
- • cvs-backend (FastAPI)
- • cvs-frontend (React)
- • cvs-worker (Celery)
- • cvs-docling (OCR Service)
- • supabase-db (PostgreSQL)
- • queue-redis (Redis)