Typo Tolerance
How Tachyon matches misspelled queries, and how to tune it.
Tachyon tolerates typos using Damerau-Levenshtein edit distance, with the allowed number of edits scaled to query-token length so short words aren't matched too loosely.
Collection-level configuration
Set at collection creation time, under typo_tolerance in the schema:
{
"typo_tolerance": {
"enabled": true,
"one_typo_min_len": 4,
"two_typo_min_len": 8,
"max_typos": 2
}
}| Field | Default | Meaning |
|---|---|---|
enabled | true | Whether typo tolerance is applied at all |
one_typo_min_len | 4 | Token length at which 1 typo becomes allowed |
two_typo_min_len | 8 | Token length at which 2 typos become allowed |
max_typos | 2 | Hard ceiling on allowed edits, regardless of length |
curl -X POST localhost:8108/collections \
-H 'Content-Type: application/json' \
-d '{
"name": "products",
"fields": [{"name": "title", "type": "text"}],
"typo_tolerance": {
"enabled": true,
"one_typo_min_len": 4,
"two_typo_min_len": 8,
"max_typos": 2
}
}'A token shorter than one_typo_min_len must match exactly. Between
one_typo_min_len and two_typo_min_len, one edit is allowed. At or above
two_typo_min_len, two edits are allowed — capped by max_typos.
Per-query override
A single query can only toggle typo tolerance on or off — the length thresholds are fixed at the collection level and can't be overridden per request:
curl 'localhost:8108/collections/products/search?q=keyboard&typo_tolerance=false'If omitted, the collection's typo_tolerance.enabled setting is used.