Turn documents into knowledge you can question.

Verdant reads the scans, tables, and long agreements other tools skip. GraphForge turns what they contain into a graph your team and your agents can query, with the evidence attached.

Talk to us

Open source to run yourself, a hosted Hub to publish and share, and a team that will build the pipeline with you.

Evidence graphagreements · 31 pages · 4 entities

MATCH(p:Person)-[:SIGNED]->(a:Document)RETURNp.name, a.title, a.evidence

Apache-2.0 and AGPL-3.0openCypher compatibleRust core, Arrow resultsParquet persistenceMCP server for agentsRuns in your environment

From a pile of PDFs to an answer a reviewer can check.

Two open-source tools, one path. Verdant gets knowledge out of documents. GraphForge holds it in a graph you can question. Each is useful on its own.

1Extract
Verdant reads the pages other parsers give up on.
Scans, tables, charts, and long PDFs become structured Markdown with page-level provenance. Every run is traced and scored against pages you approve, so you know what it got right before anything downstream depends on it.
```
## Supply agreement
| Party     | Role     |
| Northwind | Supplier |
> p.14 · signed 2024
```
2Connect
GraphForge turns what they contain into a graph.
The people, organizations, events, and claims inside that text become nodes and edges in an embedded, openCypher-compatible engine. It runs inside your process. There is no database server and nothing to provision.
3Ask
Every answer arrives with its evidence.
Analysts and agents question the graph in openCypher or plain language. Each answer carries the page it came from and how much to trust it, so a reviewer can check the reasoning instead of taking it on faith.
MATCH (p:Person)-[:SIGNED]->(a:Document) RETURN p, a
A. Reyes signed the supply agreement with Northwind Ltd.
supply-agreement-2024.pdf · p.14 · confidence 0.92

Same code. Three ways to run it.

Nothing is held back for a paid tier. The open-source tools, the hosted Hub, and our services all run the same GraphForge and Verdant, and your documents stay where they are.

Open source

Install it and go. Apache-2.0 for GraphForge, AGPL-3.0 for Verdant, the same code we run everywhere else.

$ pip install graphforge
$ git clone DecisionNerd/verdant
$ docker compose up -d

Cloud

GraphForge Hub gives your graph projects a stable address to publish, share, and clone. It runs no computation. Your data stays with you.

graphforge.sh / your-team / case-graph

v1v2v3Public

$ gf clone graphforge.sh/your-team/case-graph

Services

A fixed-scope pilot on your documents, in your environment. You keep the pipeline, the code, and the evaluation.

wk 1Extract
wk 2Connect
wk 3Ask
wk 4Decide

Two tools that meet in the middle.

Verdant

Knowledge extraction from visually rich documents. Image and PDF to structured Markdown and HTML, with charts rebuilt as diagrams, scored on your own pages.