The docs graph

The relate documentation is also published as a graph. Every page, heading, code example, glossary term, CLI command and flag, cookbook recipe, capability and problem is a nodenode An entity in the graph: it carries zero or more labels and a map of properties, and relationships connect it to other nodes. Read more in the glossary, and the links between pages, the prerequisites, the terms a page uses and the flags it documents are relationshipsrelationship A directed, typed connection between two nodes, with properties of its own; every relationship has exactly one type and a start and an end node. Read more in the glossary. The build writes the graph from the same files as the pages, so it says what the pages say, at the version in the page header. It ships as a Cypher seed script, as JSON and, from launch, as a relate storestore A relate database: a directory, a single .rl file, or :memory:, opened with --database. Read more in the glossary; the cookbook recipe Query the relate docs searches it. The view on this page draws its pages, grouped by section, below the tables of its labels and relationship types.

Synopsis

Node labels

A property whose value would be null is left out, as Cypher stores it. Section text and Example code are quoted from the pages: reference text, not instructions.

embedding is a VECTOR<FLOAT32>(384) from the bge-small-en-v1.5 model. It is set in the seed script and the store, not in the JSON.

Relationship types

The graph, drawn

The view draws each published page as a node, one cluster and one color per section, with the links and prerequisites between pages. Glossary terms, capabilities and cookbook recipes sit on the ring nearest the pages; CLI commands, flags and surfaces on the middle ring; the problems capabilities solve on the outermost ring. Each ring is off until its toggle is on. Sections and examples are not drawn: there are more than 3,000 of them, past the roughly 1,000 nodes a canvas draws comfortably, and each belongs to a page that is drawn. The build places every page from the list of pages alone, not from the links between them, so the live view draws each page where the picture puts it, and nothing moves when it loads. The links, and the counts of every kind, change with nearly every edit to any page; the view reads them from the data the site is built with.

Files

Seed script. /docs/graph.cypher is Cypher for an empty store: a comment header with the verified version, the node and relationship counts and the schema; a uniqueness constraint per label; UNWIND statements of up to 500 rows that create the nodes, then the relationships; the statements that set the section vectors; then the three indexes. Its header names the command that builds a new store from it, relate query --database ./relate-docs.rl < graph.cypher.

JSON. /docs/graph.json holds format (1), kind (documentation), verified, seed (the seed script's URL), embeddings (the label, property, index, dimension, similarity and model of the vectors, which it does not carry), then nodes, each with id (Label:id), labels and properties, and relationships, each with type, start, end and properties. It carries every section's text and every example's code, so it runs to several megabytes.

relate store. /docs/relate-docs.rl is the seed script built into a single-file storesingle-file store A store kept in one .rl file that holds the snapshot and the WAL together; the layout for a --database path ending in .rl. Read more in the glossary by the relate CLI, with its SHA-256 beside it. Both URLs are launch-day targets: the store is built and served once the deploy workflows install relate. --load reads it into an in-memory storein-memory store A store that lives only in the process that opened it, named :memory:; the default --database, gone when the process exits. Read more in the glossary, which is read-only.

Per page. Every published page has /docs/<page>.graph.md, a markdown list of its neighborhood under the headings Prerequisites, Prerequisite of, Related, Links to, Linked from, Terms, Sections, Examples, and CLI commands and flags.

The view's data. /docs/graph-view.json is what the view on this page draws: the pages, terms, capabilities, problems, recipes, commands, flags and surfaces, each with its title, description, link and position, and the relationships between them. It leaves out sections and examples, and it is shaped for this page; the seed script and graph.json are the complete forms.

What runs on it

The local form serves the docs graph to an MCP host over stdio. The URL is a launch-day target:

relate serve --mcp stdio --database :memory: --load https://relatedb.ai/docs/relate-docs.rl --read-only

A store opened read-only lists read_cypher and get_schema, not write_cypher (read-only policy), and --load into :memory: opens it read-only. --safe refuses a remote --load source, so the checked path is to download the store, compare its SHA-256 with the published value, and load the local copy. In this block, <sha256> is the value in relate-docs.rl.sha256, and the search runs only when it matches; on macOS, shasum -a 256 --strict -c - stands in for sha256sum --strict -c -:

curl -sSfLO https://relatedb.ai/docs/relate-docs.rl
echo "<sha256>  relate-docs.rl" | sha256sum --strict -c - && relate --load ./relate-docs.rl --mode envelope "MATCH (s:Section) SEARCH s IN (FULLTEXT INDEX docs FOR 'exit code' LIMIT 5) SCORE AS score RETURN s.page AS page, s.heading AS heading, s.url AS url, score"

The downloaded store served to an MCP host. A relative path resolves against the directory the host starts the server in:

claude mcp add relate-docs-local -- relate serve --mcp stdio --database :memory: --load <absolute-path>/relate-docs.rl --read-only

Examples

Over the store, or the seed script replayed into an empty one, these statements read the graph. Their results change as the pages do.

The pages that link to the MCP server reference:

MATCH (p:Page)-[:LINKS_TO]->(:Page {id: 'reference/server/mcp'})
RETURN DISTINCT p.id AS page, p.section AS section ORDER BY page

The glossary terms the most pages use:

MATCH (:Page)-[:USES_TERM]->(t:Term)
RETURN t.term AS term, count(*) AS pages ORDER BY pages DESC, term LIMIT 5

The chains of prerequisites that lead to a page:

MATCH path = (:Page)-[:PREREQUISITE_OF*1..]->(:Page {id: 'guides/agents/connect-over-mcp'})
RETURN [n IN nodes(path) | n.id] AS chain ORDER BY length(path) DESC

See also

• Query the relate docs: vector search and graph traversals over a part of this graph, in the browser.
• MCP server: relate's own server, which the local form runs.
• The capability catalog: the source of the Capability and Problem nodes.