> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.cloudraker.com/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.cloudraker.com/_mcp/server.

# Search and the knowledge graph

Once a space's files finish processing, two tabs let you work with what's inside them: **Search** finds passages, and **Graph** shows the entities and relationships extracted from your documents. In both, clicking a result opens its exact source.

## Grounding — click a result, see the source

Every result in Search, the Graph, the Data tab, and action results is **grounded**: clicking it opens the cited value **in its original source**.

* For a **PDF**, the viewer navigates to the cited page and highlights the bounding box of the citation.
* For **audio**, the player seeks to the mention's timecode and the transcript scrolls to it.

Every extracted value, search hit, and graph entity points back to where it came from, so you can verify it before you approve it.

## Search

The **Search** tab runs semantic and keyword search across the space's documents and transcripts. Type into **Search this space's documents…** to begin. The empty state reads **Search this space** — "Find passages across this space's documents and transcripts."

Results are **grouped by file**, each with a type icon and filename. Every hit shows:

* A **location badge** — **p. N** for documents, or **Speaker N · m:ss–m:ss** for audio.
* A snippet with the matching words highlighted.

Use **Show N more** to expand a file's remaining hits. Click any hit to open it in its source through the grounding dialog above. If nothing matches, you'll see "No results" — "Try different words or a broader query."

## Graph — the knowledge graph

The **Graph** tab shows a knowledge graph automatically built from the space's processed documents. An overlay panel titled **Space graph** shows a count line ("`{entities}` entities, `{relations}` relations"), a **Refetch** button, and an entity **search box**.

The graph is empty until files finish processing — you'll see **No entities yet**, "Processed files will appear here after ontology extraction finishes." Once populated, the canvas shows entities as nodes and relationships as edges, with zoom and fit controls.

* **Click an entity** to open a detail rail: its name, a **type badge**, an optional key, and its **mentions** — each with the source file name and a location badge (page or timecode). Click a mention to open it in its source.
* **Click an edge** to see the relation, its source → target, and the source file. Click through to the source.

How the graph is shaped — the extraction guidance, the entity types it recognizes, and whether the model gets the current date — is configured on the space's **space type**, under its Ontology tab. See [Space types](/spaces/space-types).