Architecture
How Graffold Works
A split database architecture — Memgraph for graph traversals, Cloudflare Vectorize for semantic search — with multi-source ingestion and hybrid retrieval. Full provenance on every answer.
Ingestion Pipeline
Query & Retrieval Architecture
Integrations
Graph & Vector Layer
- Memgraph + MAGE
- Cloudflare Vectorize
- Neo4j · FalkorDB
LLM Providers
- AWS Bedrock (Claude, Titan)
- AWS SageMaker
- Cloudflare Workers AI
- Ollama (local / air-gapped)
- OpenAI-compatible APIs
Data Sources
- Any source with an API
- PDF files (vision + OCR)
- CSV / Excel / Parquet
- PubMed / bioRxiv (built-in)
Infrastructure
- Redis (cache + sessions)
- OpenTelemetry + Grafana
- Docker (per-tenant isolation)
- HuggingFace embeddings (768d)
Split Database Architecture
Graph storage and vector search are decoupled — each layer scales independently.
Memgraph — Graph Layer
- ▸ In-memory graph — sub-millisecond traversals
- ▸ MAGE algorithms (PageRank, community detection, shortest path)
- ▸ openCypher queries over Bolt protocol
- ▸ Dedicated container per tenant (256 MB–1 GB)
- ▸ Bolt-compatible — same driver code works with Neo4j
Cloudflare Vectorize — Vector Layer
- ▸ 768-dimensional embeddings (all-distilroberta-v1)
- ▸ Cosine similarity search at the edge
- ▸ Managed scaling — no infrastructure to maintain
- ▸ Decoupled from graph memory
- ▸ Metadata-enriched vectors for result context
Alternative Backends
Enterprise graph — native HNSW vector indexes, APOC/GDS algorithms
Redis-native graph — named graph isolation, lightweight deployments
Any Bolt-compatible database works via the DatabaseInterface abstraction
Performance Stack
Rust-accelerated JSON
3-10× faster serialization
Rust ASGI Server
2-4× request throughput
Memgraph Bolt Driver
In-memory graph, sub-ms traversals
Rust-native DataFrames
2-5× faster for data processing
Token-aware Chunking
BPE tokenizer with sentence boundaries
Hybrid Retrieval
Vector + fulltext + graph traversal
Performance Benchmarks
Measured speedups from Rust- and C-backed drop-in replacements across the stack. Zero application-code rewrites required.
3–10×
JSON Serialization
Rust-backed encoder vs stdlib
2–4×
HTTP Throughput
Rust ASGI server vs Python default
<1 ms
Graph Traversal
In-memory Memgraph, sub-millisecond
2–5×
DataFrame Operations
Rust-native DataFrames vs legacy libraries
~10×
Cache Parsing
C-backed parser vs pure-Python
250 ms
P50 Query Latency
Hybrid vector + graph retrieval
Processing Throughput
1,000+
Documents per pipeline run
Multi-source
Parallel ingestion across data feeds
47 → 1
Duplicate edges consolidated per entity pair
< 15 min
Incremental update for ~100 new documents
Ontology Studio
Define your domain schema visually. The extraction pipeline adapts to your ontology — producing typed, validated knowledge graphs instead of generic entity soup.
Visual Schema Designer
Drag-and-drop node types, draw relationships, define properties. See your ontology as a live graph.
Guided Extraction
Your ontology constrains the LLM — extraction produces typed entities that match your schema, not generic blobs.
OWL / JSON-LD
Import existing ontologies from OWL or JSON-LD. Export for use in Protégé, Neo4j, or any RDF-compatible tool.
Available on Enterprise tier