Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/org-quicko/silo/llms.txt

Use this file to discover all available pages before exploring further.

Silo’s search engine lets you find entries by the words they contain, not just by filtering on exact field values. Search works at three levels of reach, each expressed as a distinct URL path — a narrower path can never accidentally widen to a broader one because of a missing parameter.

​
Three search scopes

Collection

Search within a single collection.GET /api/projects/{project}/envs/{env}/collections/{name}/search?q=term

Environment

Search across all collections in one environment.GET /api/projects/{project}/envs/{env}/search?q=term

Instance

Search everything the API key can read.GET /api/search?q=term

​
Making a search request

curl "http://localhost:8090/api/search?q=pricing" \
  -H "Authorization: Bearer $SILO_KEY"
A response names the project, environment, and collection for every hit, and quotes the text that caused the match: 
{
  "data": [
    {
      "project": "acme",
      "env": "prod",
      "collection": "posts",
      "entry": {
        "id": "01J8...",
        "title": "Pricing changes"
      },
      "snippets": [
        {
          "path": "$.data.body",
          "before": "our ",
          "match": "pricing",
          "after": " page"
        }
      ]
    }
  ],
  "total": 1,
  "limit": 50,
  "offset": 0,
  "truncated": false,
  "engine": "fts5"
}
Each snippet is three strings: before + match + after reconstructs the surrounding fragment. The match portion is the run to highlight — it contains no escape sequences, so text with brackets or special characters needs no extra handling.

​
Filtering, sorting, and pagination

Search accepts the same filter, sort, limit, and offset parameters as list queries, plus q for the text term. 
GET /api/search?q=pricing&filter=<url-encoded JSON>&sort=-$.updated_at&limit=20&offset=0
Omit sort to rank results by relevance. Adding a sort overrides relevance ranking and returns results in the specified order instead.
Anonymous callers can search, but they only reach collections that do not set x-silo-auth: true in their schema. Authenticated keys follow their normal claim-based access rules.

​
The engine and truncated fields

The engine field in the response tells you how the query was answered:
ValueMeaning
fts5SQLite’s full-text index answered the query. Results are precise and ranked by relevance.
scanThe portable fallback engine walked entries one by one. Used when the FTS index is unavailable or disabled.
When engine is scan, truncated: true means that Silo hit the scan limit before examining all entries, so total is an approximation of how many matched within the examined window — not the true total.

​
Configuring indexed fields

Search only indexes the fields you explicitly list. Configure which fields are indexed by adding x-silo-search to your collection’s schema: 
{
  "type": "object",
  "x-silo-search": ["$.data.title", "$.data.body"],
  "properties": {
    "title":  { "type": "string" },
    "body":   { "type": "string" },
    "status": { "type": "string" }
  }
}
JSONPath expressions are scoped to the internal entry document. Your own fields live under $.data, so $.data.title indexes the title field of your entry.
x-silo-search is one of the schema keywords that stays editable even when the collection holds entries. Update it via PUT /api/projects/{project}/envs/{env}/collections/{name}/schema, then rebuild the index for the change to take effect.

​
Search configuration in silo.toml

Configure the search engine under the [search] section of silo.toml: 
[search]
enabled              = true
tokenizer            = "unicode61"   # or "trigram" for substring / CJK search
max_entry_bytes      = 65536
scan_limit           = 10000
scan_time_budget_ms  = 2000
​
tokenizer
string
default:"unicode61"
Controls how text is split into searchable tokens.
  • unicode61 — word-based tokenization. Best for most Western-language content. Matches whole words.
  • trigram — splits text into three-character windows. Required for substring search (e.g., matching “price” inside “repriced”) and for CJK languages (Chinese, Japanese, Korean) where word boundaries are not spaces.
​
max_entry_bytes
integer
default:"65536"
Maximum number of bytes extracted from each entry for indexing. Content beyond this limit is not indexed.
​
scan_limit
integer
default:"10000"
Maximum number of entries the portable scan engine will examine before stopping and setting truncated: true.
​
scan_time_budget_ms
integer
default:"2000"
Maximum wall-clock time in milliseconds the portable scan engine may spend on a single query.

​
Rebuilding the search index

If indexed fields change, or if the index gets out of sync for any reason, trigger a full rebuild: 
curl -X POST http://localhost:8090/api/search/reindex \
  -H "Authorization: Bearer $SILO_KEY"
The --check flag on the CLI reports the index status and signals whether a rebuild is needed before performing one.

Build docs developers (and LLMs) love

Get started for freeTalk to us