Advanced: work directly with GraphForge

This path assumes you can read basic Python, understand database tables and queries, and run commands in a terminal. It does not assume graph-database, research-platform, or storage-engine expertise. For guided research without writing code, choose the Basic path.

Bring what you already know

Familiar idea	GraphForge idea	What you do
A row representing a person or document	A node with a label and properties	add_node("Paper", title="Survey") creates one paper
A link between rows, often queried with a join	A named, directed relationship	add_edge(survey, "CITES", methods) records which paper cites which
A query producing a table	An openCypher query	execute() returns a result table in Python
Data stored between runs	A project directory	Open GraphForge(path) and close it when finished

A label groups similar nodes; a property stores a detail. A direction such as CITES is part of your model, not a causal claim. Unlike a declared SQL schema, an exploratory graph can start without an ontology. An ontology is an optional set of rules about entity types, relationships, and values; learn it when you need shared definitions or validation.

Run, inspect, then adapt

1. Install a matching package. If you prefer cells to a script, start a local notebook.
2. Run Your first graph, including its expandable Python steps. Confirm the expected result before changing the example.
3. Query, analyze, and save. Supply query values as parameters and reopen the saved project in a new process.
4. Construct your own graph, then query it with Cypher. Choose identifiers and what each relationship means before importing rows.

For example, this pattern follows a stored citation:

MATCH (source:Paper)-[:CITES]->(target:Paper)
RETURN source.title AS source, target.title AS target

MATCH selects matching connections. Parentheses name nodes; :Paper selects the label; -[:CITES]-> selects an outgoing citation. RETURN chooses the result columns. It does not infer a connection that you did not store.

Tabular query and analysis results in Python are Arrow tables, a column-based data format. Start with result.to_pylist() to inspect ordinary Python rows. Conversion to pandas or Polars and graph algorithms are explained in analytics integration. In Node, those tables arrive as Arrow IPC bytes; the integration guide shows decoding. Construction can return handles, and metadata or lifecycle operations can return collections, scalars, or no value. You do not need serialization internals to use the Python path.

Add one capability for the next task

Need	Read next	Concept introduced there
Combine numerical and interview evidence	Mixed-methods worked example	Unit of analysis, coding, integrated interpretation
Save a challenged hypothesis and find it again	Record an inquiry	Scope and evidence, stored as ordinary graph records
Keep an exact past state while editing	Keep research history	Version: a retained state unaffected by later edits
Move a project to another folder or machine	Move and share	Exported package and verified import
Explore optional model-assisted applications	Application examples	Human-reviewed extraction and retrieval

A UUID is an identifier rather than a row number. Some advanced operations also require an operation identity, which lets the engine recognize an exact retry. Reuse the original request and identity for that retry; a changed action is a new operation. Follow the concrete example for the operation you need.

Look up exact method signatures in the API reference. Architecture documents, wire formats, and contributor requirements explain how the engine is built. They are separate specialist material and are not the next level you must complete to use these advanced guides.