Full-Text Search

Search the words in your content, with typo tolerance and stemming, from SQL or over HTTP. Indexing is automatic for the fields a NodeType marks as searchable.

Search from SQL

FULLTEXT_SEARCH is a table function. It takes the query, the language code, and a required workspaces scope:

SELECT path, score
FROM FULLTEXT_SEARCH('databases', 'en', workspaces => 'blog')
ORDER BY score DESC
LIMIT 10;

{"columns":["path","score"],"rows":[
  {"path":"/posts/post-1","score":0.01639},
  {"path":"/posts/post-2","score":0.01613}
]}

• The language is an ISO 639-1 code ('en', 'de'). It selects the stemmer; the index stores two-letter codes, so 'english' is rejected.
• workspaces is one name, a comma-separated list ('blog, docs'), a glob ('content-*'), or 'ALL READABLE' for every workspace you may read.
• limit => 50 caps the number of hits taken from the index (default 100). LIMIT on the outer query still applies.

Each row also carries node_id, workspace_id, name, node_type, properties, created_at, updated_at, fulltext_rank (1 for the best hit) and, for hybrid queries, vector_rank and vector_distance. score is a rank-based value, 1 / (60 + rank), so it is comparable between queries and between full-text and vector legs.

Filtering search results

Any SQL predicate can follow the search:

SELECT path, properties->>'title' AS title, score
FROM FULLTEXT_SEARCH('databases', 'en', workspaces => 'blog')
WHERE node_type = 'raisin:Page'
  AND properties->>'status'::String = 'published'
  AND PATH_STARTS_WITH(path, '/posts/')
ORDER BY score DESC
LIMIT 10;

Search as a predicate

FULLTEXT_MATCH(query, language) is the same search written as a WHERE condition on the workspace table. It becomes the scan for the query, and other predicates filter its hits:

SELECT path FROM 'blog'
WHERE FULLTEXT_MATCH('databases', 'en')
  AND CHILD_OF('/posts')
  AND properties->>'status'::String = 'published';

Use FULLTEXT_SEARCH when you want the score; FULLTEXT_MATCH when you only need the rows.

Query syntax

The query string uses Tantivy's parser over two fields: name (the node name) and content (every searchable property, concatenated).

Write	Meaning
database query	either word (terms are OR-ed)
database AND query	both words
+database -tutorial	must contain / must not contain
"content management"	the exact phrase
datab*	prefix
name:about	match the node name only

Matching is case-insensitive. Terms tolerate one typo (databses finds databases), and the language's stemmer matches word forms (running finds run). Field names other than name and content are an error; property names are not query fields.

Which fields are searched

A NodeType declares which properties go into the index. In YAML:

name: blog:Article
indexable: true
properties:
  - name: title
    type: String
    required: true
    index: [Fulltext]
  - name: body
    type: String
    index: [Fulltext]
  - name: internal_notes
    type: String        # not searchable

In SQL DDL, the FULLTEXT modifier does the same:

CREATE NODETYPE 'blog:Article' PROPERTIES (
  title String REQUIRED FULLTEXT,
  body String FULLTEXT,
  internal_notes String
) INDEXABLE;

Only properties with the Fulltext index are searchable; a word that appears only in internal_notes finds nothing. The built-in raisin:Page indexes title and content. Nodes are indexed by a background job shortly after each write.

Search over HTTP

curl -X POST http://localhost:8090/api/repository/myrepo/main/fulltext/search \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"query": "databases", "workspace": "blog", "language": "en", "limit": 10}'

[
  {"node_id":"a178…","workspace_id":"blog","name":"post-1","path":"/posts/post-1","node_type":"raisin:Page","score":11.24},
  {"node_id":"d571…","workspace_id":"blog","name":"post-2","path":"/posts/post-2","node_type":"raisin:Page","score":11.24}
]

Field
query	required; same syntax as above
workspace	one workspace; omit to search all workspaces of the repository
language	defaults to the repository's default language
limit	default 20, maximum 100
shape_type	only nodes whose type, archetype or nested element type is this name

This endpoint returns the raw index score (BM25), which is larger for better matches but not comparable across queries.

The combined search endpoint, GET /api/search/{repo}?q=…&workspace=…, runs the same engine as HYBRID_SEARCH and returns rank-fused results. Pass strategy=fulltext for a keyword-only search; the default hybrid strategy needs an embedding configuration for the tenant. See Vector Search.

Next Steps

• Full-Text Search concepts for indexing, stemming and maintenance
• Filtering Data
• SQL Basics