Graph

The Graph tab shows the grounded entities and relationships that have been imported from a Hindsight memory bank. The left side lists entities and edges; the right side shows the details for whichever item you select.

Layout

The Graph tab is split into three resizable panels:

• Entities (top left) — grounded nodes such as people, organizations, projects, products, or any other entity type imported from Hindsight.
• Edges (bottom left) — grounded relationships between those entities.
• Details (right) — the full information panel for the selected entity or edge.

You can drag the resize handles to change panel sizes, and double-click a handle to reset it.

Entities list

Entities are displayed as rows in the top-left panel. Each row shows:

Field	What it tells you
Label	The human-readable name of the entity.
Type and ID	The entity type (for example person, project, product) followed by its stable ID.
Left border color	A color assigned to the entity type so you can scan the list quickly.

Click any entity to open its details in the right panel.

Edges list

Edges are displayed in the bottom-left panel. Each row shows:

Field	What it tells you
Source → Target	The two entities connected by the relationship.
Description / type	A short description of the relationship, or the relationship type if no description is available.
Edge count badge	When multiple physical edges are grouped under a single edge context, the badge shows how many edges are in that group.

Click any edge to open its details in the right panel.

Why are edges grouped?

Mental-model-driven edge contexts can represent many concrete Hindsight relationships as one conceptual edge. The group row lets you see that an edge context exists, while the child rows beneath it show the individual physical edges.

Details panel

When you select an entity or an edge, the right panel shows everything architxt knows about that item. The sections are the same for both nodes and edges unless noted below.

Header

• Title — the entity label, or the edge description if one exists.
• Subtitle — the entity ID, or the source_id → target_id pair for an edge.
• Badges — entity labels (for nodes) or the edge type (for edges).

Mental model refs

Lists any contextual mental models attached to this entity or edge. For each model you see:

Field	Meaning
Scope badge	Whether the model is attached to a node, edge, seed, or other scope.
External ID	The stable Hindsight ext_id of the model.
Template role	The template role the model was derived from.
Attached / fetched	When the model was attached to the graph and when its content was last fetched from Hindsight.
Content hash	A hash used to detect content changes.
Refresh status	Whether the last refresh succeeded, was skipped, or errored — and the error if one occurred.

Provenance

Shows where the item came from — for example, whether it was imported from Hindsight, created by a sync job, or produced by a specific pipeline stage. Empty provenance means the source was not recorded.

Stored data

The raw properties stored for this entity or edge, excluding provenance and legacy computed fields. This is useful when you need to see the underlying values that feed into models or sync logic.

Timestamps

• Last refreshed — most recent mental-model content refresh across all attached models.
• Created — when the item first appeared in the contextual graph.
• Updated — when the item's record was last modified.
• Last seen — when the item was most recently observed during a sync.

Interactions

Action	What it does
Click an entity or edge	Opens its details in the right panel.
Search	Filters both entities and edges by label, ID, type, source, or target.
Drag resize handles	Resizes the panels.
Double-click a resize handle	Resets that split to its default size.
Export	Opens the export dialog for the current bank's graph.

Export

The Export button sits on the same row as the graph search bar and the node/edge count. Clicking it opens a dialog where you choose how to download the current bank's contextual graph.

	Option	What it controls
	Format	The export file format.
	Filename base	An optional prefix for the downloaded files. If you leave this blank, architxt uses the bank name and today's date.
	Undirected edges only	When checked, the export contains only edges explicitly marked as undirected.
	Entity Content (JSON only)	When checked, every mental-model ref attached to a node is expanded with its normalized envelope content.
	Edge Content (JSON only)	When checked, every mental-model ref attached to a grounded co-occurrence edge is expanded with its normalized envelope content. Other edges are left unchanged.

Entity Content and Edge Content are JSON-only

The two content checkboxes are only available when Format is set to JSON. They are disabled for all other export formats because those formats do not carry mental-model model_refs in a way that can meaningfully include full envelope content.

Embedded mental-model content

When Entity Content is enabled, each node properties.provenance.model_refs[] entry gains a content field containing the normalized envelope of the referenced Hindsight mental model. The envelope follows the standard architxt shape — narratives, graph, tables, and diagrams — and is produced by toEnvelopeWithFallback from the model's reflect_response.

When Edge Content is enabled, the same content field is added to model_refs on grounded co-occurrence edges (edges whose properties.labels contains 'grounded'). Non-grounded edges keep their model_refs but receive no content field, so their mental models are not fetched.

If a referenced model cannot be fetched, the content field is set to null and the export continues. Each referenced model is fetched individually, so large graphs with many attached models may take longer to export when these options are enabled.

Supported formats

	Format	Download
	JSON	A single .json file containing the nodes, edges, export metadata, and (optionally) embedded mental-model envelope content.
	GraphML	A single .graphml file for tools that support the GraphML format.
	GEXF	A single .gexf file for Gephi and other GEXF-compatible tools.
	Cypher	A single .cypher file containing Neo4j-flavoured Cypher statements.
	CSV	Two files: nodes.csv and edges.csv. The node file has an id and labels column plus any stored properties; the edge file has id, source_id, target_id, type, and directed columns plus any stored properties.
	C4 (Structurizr DSL)	A single .dsl file in Structurizr DSL format, using the C4 element and relationship mappings defined on entity/edge types.
	C4 (Mermaid)	A single .mmd file in Mermaid C4 diagram format, using the same C4 mappings.
	ArchiMate XML	A single .xml file in ArchiMate Open Exchange Format.

API filtering

The export dialog calls the /contextual-graph/export endpoint. The same endpoint can be used directly with the following query parameters:

	Parameter	Effect
	server_id (required)	The Hindsight server to export from.
	bank_id (required)	The bank to export from.
	format	json (default), graphml, gexf, cypher, csv, c4, c4-mermaid, or archimate.
	labels	Comma-separated node labels. Only nodes that have all of these labels are included.
	id_prefix	Include only nodes whose ID starts with this prefix.
	ids	Comma-separated list of exact node IDs to include.
	source_id	Include only edges that originate from this node ID.
	target_id	Include only edges that point to this node ID.
	type	Include only edges of this relationship type. Pass an empty value to match untyped edges.
	undirected	true to include only edges explicitly marked as undirected.
	include_entity_content	true to embed normalized mental-model envelope content in every node model_refs entry. JSON format only.
	include_edge_content	true to embed normalized mental-model envelope content in grounded co-occurrence edge model_refs entries. JSON format only.

When any node selector is used (labels, id_prefix, or ids), the export trims the edge list to edges whose source and target are both in the returned node set, so the download stays a coherent subgraph.

Example JSON output with content enabled

{
  "meta": {
    "server_id": 1,
    "bank_id": "Mozart-API",
    "format": "json",
    "include_entity_content": true,
    "include_edge_content": true,
    "node_count": 2,
    "edge_count": 1
  },
  "nodes": [
    {
      "id": "svc:SVC-005",
      "labels": ["service"],
      "properties": {
        "provenance": {
          "source": "hindsight",
          "model_refs": [
            {
              "role": "sys_entity_summary",
              "ext_id": "summary-svc-005",
              "content": {
                "narratives": [{ "narrative_name": "Summary", "narrative": "..." }],
                "graph": { "name": "...", "nodes": [], "edges": [] },
                "tables": [],
                "diagrams": []
              }
            }
          ]
        }
      }
    }
  ],
  "edges": [
    {
      "id": "e1",
      "source_id": "svc:SVC-005",
      "target_id": "com:COM-001",
      "type": null,
      "properties": {
        "directed": false,
        "labels": ["grounded", "active"],
        "provenance": {
          "source": "contextual-graph",
          "model_refs": [
            {
              "role": "sys_edge_context",
              "ext_id": "edge-context-e1",
              "content": {
                "narratives": [{ "narrative_name": "Context", "narrative": "..." }],
                "graph": { "name": "...", "nodes": [], "edges": [] },
                "tables": [],
                "diagrams": []
              }
            }
          ]
        }
      }
    }
  ]
}

How the Graph tab relates to the others

The Graph tab contains the grounded items: things already imported and promoted from Hindsight. Candidate nodes and edges that have not been promoted live in the Candidates tab, and the contextual mental models attached to graph items are managed in the Mental Models tab. The background work that imports and refreshes these items is shown in Sync Jobs.