Every extraction is indexed. Search it by meaning, by keyword, or by field — then total it in the database.
Pick one per request, or let hybrid do both at once.
Exact filters over extracted fields — vendor, document number, type, currency, issue date, total.
Natural-language similarity over document summaries. Finds the right document when you don't know its number.
Keyword and semantic results fused together. The best default for a free-text question.
# Hybrid search: keywords + meaning (best default for free text)
curl -X POST https://api.gemina.co/api/v1/retrieval/query \
-H "X-API-Key: your-api-key" \
-H "Content-Type: application/json" \
-d '{
"mode": "hybrid",
"text": "the credit note about the server outage",
"topK": 10,
"filters": { "issueDateFrom": "2026-01-01" }
}'Sums, averages, and counts — never estimated by a model. The same numbers you would get by querying your own warehouse.
# Exact totals per vendor, computed in the database (never estimated by AI)
curl -X POST https://api.gemina.co/api/v1/retrieval/aggregate \
-H "X-API-Key: your-api-key" \
-H "Content-Type: application/json" \
-d '{
"metrics": [{ "op": "sum", "field": "total_amount" }, { "op": "count" }],
"groupBy": ["vendor_name", "currency"],
"filters": { "issueDateFrom": "2026-01-01", "issueDateTo": "2026-03-31" }
}'Ask for a money total without naming a currency and the server adds currency to the grouping rather than returning one meaningless number.
"meta": {
"currencyGroupingAdded": true
}It tells you it did it, so the number is never quietly wrong.
The same invoice indexed twice is counted once. Duplicates are matched on vendor tax ID, document number and issue date.
"meta": {
"duplicatesIgnored": 3
}Above zero means the same document was uploaded more than once.
Quantities in different units are never summed together, and line items with no unit data are reported rather than silently skipped.
"meta": {
"uomGroupingAdded": true,
"linesWithoutUnitData": 12
}You always know when a total would mislead you.
You never have to hard-code a schema into your search UI. Ask the API which fields your own documents actually carry, and build the filter list from the answer.
Add a new document type and its fields become filterable without a deploy on your side.
Search works across everything you extract. Numeric roll-ups are narrower, and it's worth knowing where the line is.
Anything you extract.
Invoices, forms, contracts, custom templates, FileTag metadata. Any field in your extracted data can be filtered and searched.
Money and quantity.
Totals, net and tax amounts, and measurable quantities — grouped by vendor, currency, type, month, and more.
Custom-template numbers.
Numbers in your own templates are searchable and filterable, but not yet available as aggregate metrics.
Start free. Indexing is opt-in, and search is never metered.