Stop drowning in documents.
Start finding answers.

Graffold turns scattered files, databases, and APIs into a single knowledge graph you can query in plain language — with every answer traced to its source.

290+

Modules

53K

Lines of Code

6

Database Backends

100%

Source Provenance

Deploy Your Way

☁️ Cloud

Managed databases and hosted LLMs. Deploy with Docker Compose — scales to millions of entities.

Memgraph Neo4j Bedrock Cloudflare Workers AI

🏠 Local

Full stack on your machine. No external API calls. Air-gapped deployments for sensitive data. One docker compose up.

Ollama Memgraph

🔀 Hybrid

Mix and match. Local database with a cloud LLM. Remote graph with a local model. No lock-in at any layer.

Local DB + Cloud LLM Any combination

What's Under the Hood

Multi-Source Ingestion

PDFs, CSVs, REST APIs, spreadsheets — any source with an API plugs in. Multi-strategy PDF extraction with vision-based OCR, Nougat for scientific papers, and parallel batch processing.

Provenance & Temporal Tracking

Every entity traces to its source. Temporal validity windows, contradiction detection, and confidence decay over time.

Rust-Accelerated Hot Paths

orjson (3-10× JSON), Granian (2-4× HTTP), Memgraph MAGE (native graph algorithms), Polars, tiktoken — measurable throughput gains, zero rewrites.

Full Observability

5 Grafana dashboards, Prometheus metrics, structured audit logging, per-session LLM cost tracking, and regression alerts on eval metric drops.

Frequently Asked Questions

If you have any other questions, don't hesitate to contact us.

About Us

Headquarters
Stockholm, Sweden
Founded
2025
Stack
Python · Rust · TypeScript
Codebase
290 modules · 53K LOC

Critical knowledge is buried across millions of documents. Graffold consolidates it into a queryable graph so teams discover connections in seconds instead of months.

Do I need a graph database to get started?

No. Graffold provisions a Memgraph container automatically — it runs in 256 MB of memory. For production, dedicated per-tenant containers provide full isolation. Neo4j and FalkorDB are available as alternative backends via a single config change.

What file types can Graffold ingest?

PDFs (with vision OCR), CSV, Excel, Parquet, JSON, XML, Markdown, and any REST API. The connector pattern means adding a new source is a single adapter class.

Can I run it fully offline?

Yes. Local mode uses Ollama for LLM inference and an embedded graph database. Zero external API calls — suitable for air-gapped environments.

How does provenance work?

Every entity and relationship stores its source document, extraction timestamp, and confidence score. Answers cite specific sources with page numbers and passages.

How long does setup take?

Local mode: one docker compose up and you're querying in minutes. Cloud deployments use Docker Compose with pre-built stacks — typically under an hour.

Is there vendor lock-in?

None. Database, LLM provider, and embedding model are all swappable via the DatabaseInterface abstraction. Move from Memgraph to Neo4j, or Bedrock to Ollama, without changing application code.

What databases are supported?

Memgraph (primary, in-memory graph) + Cloudflare Vectorize (vector embeddings). Neo4j and FalkorDB available as alternative graph backends via the DatabaseInterface abstraction.

Which LLM providers work?

AWS Bedrock, Cloudflare Workers AI, Ollama, and any OpenAI-compatible API.

Is Graffold open source?

Core platform is open source. Enterprise features (SSO, monitoring, dedicated support) available separately.

How fast is it?

250ms P50 query latency. Rust-accelerated JSON and HTTP, in-memory graph via Memgraph. Parallel ingestion for bulk runs.

Are you drowning in documents?
Try us out.

Or email us directly at hello@graffold.com