Skip to main content
The Memory Schema page (route /schema, under the EXPLORE section of the sidebar) defines how Cognee extracts knowledge from a brain and rebuilds its memory. It shows a schema-view visualization alongside a config toolbar where you set the graph model, prompt, and ontology used during processing. The graph-model, prompt, and ontology configuration — together with the Re-process action — were moved here from the Mindmap page, which is now focused on visualization only.

​
Selecting a brain

A BrainSelector at the top chooses the active dataset (brain). The H1 reads Memory Schema and the subtitle shows Define extraction rules and re-process to rebuild the memory for . The page renders a schema-view visualization embedded from /v1/visualize for the selected brain. Use the Refresh button to reload the visualization. If the selected brain has no schema yet, you see an empty state — No memory schema yet — with a Go to Brains button that takes you to the Brains page.

​
Config toolbar

The config toolbar has three dropdowns — Model, Prompt, and Ontology — each defaulting to Automatic. Configuration is per brain (dataset) and is also editable from the brain detail page.
Changing the model, prompt, or ontology when a brain already has files marks the brain Outdated until you Re-process. The new configuration applies on the next run, and the Outdated indicator appears on the Brains list and brain detail pages.

​
Graph model

The Model dropdown selects the graph model that defines which entity types and relationships Cognee extracts:
  • Automatic (default) — Cognee infers the schema from your data.
  • A saved model — a model you previously created and named.
Each saved model offers an Edit option (opens the Graph Model editor) and a Create new option. Graph models are reached only from Memory Schema or a brain detail page — there is no Graph Models nav item.

​
Graph Model editor

The editor is a visual canvas of entity types as draggable nodes. The right sidebar edits the selected type’s Name, Description, and Fields. Field and relation types are:
TypeDescription
stringText value.
numberNumeric value.
booleanTrue/false value.
dateDate value.
relationA link to another entity; needs a target entity and a cardinality of one or many.
The header shows a rename control, a types, relationships count, an Unsaved indicator, and Delete, Regenerate, and Save buttons. Regenerate analyzes the selected brain’s files to propose a schema. (The same “infer from data” action is also available as Infer from data on the brain detail page.)

​
Custom prompt

The Prompt dropdown selects the extraction prompt:
  • Automatic (default) — Cognee uses its built-in extraction prompt.
  • A named prompt — a prompt you created.
Create new offers two starting points:
  • Generate from graph model — seeds the prompt from a selected graph model (requires a selected model).
  • Start blank — begins with an empty prompt.
Named prompts are editable and deletable.

​
Ontology

The Ontology dropdown selects an optional OWL ontology to guide extraction:
  • Automatic — no custom ontology.
  • An uploaded ontology — an OWL file you provided.
Choose Upload new to add one. An upload needs a Key, an optional Description, and a .owl file.

​
Re-process

Click Re-process to rebuild the brain’s memory. This re-runs cognify with the chosen model, prompt, and ontology, and shows a Re-processing started notification. Re-process to clear the Outdated state after a config change.

Configuration & Ontologies

API equivalents for graph models, prompts, and ontologies.

Brains

Edit per-brain config from the brain detail page.

Mindmap

Visualize the memory the schema produces.