메뉴
Use when adding semantic, similarity, embedding, or KNN vector search to a Go app with gosqlite — the typed vec.Table API over sqlite-vec, distance metrics, encodings, quantization, and metadata filters.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
Use when the task needs a SQL function or virtual table from the gosqlite ext/ catalog — regexp, uuid, hash, ipaddr, zorder, stats, unicode, encode, text, fuzzy, decimal, money, time, eval, csv, lines, statement, closure, pivot, rtree, series, array, bloom, spellfix1, fileio, blobio.
Use when debugging surprising behaviour with gosqlite, or as a pre-ship checklist — the gotchas around per-conn hooks, the connection pool, vec quirks, bind limits, and what is not supported.
Use when the task needs encryption at rest, storing a database compressed at rest, page-checksum corruption detection, an in-memory database with isolation, or serving a database from an embed.FS / byte buffer in a Go app with gosqlite.
Use when using gorm.io/gorm with gosqlite — the gosqlite.org/gorm dialector, a CGo-free drop-in for glebarez/sqlite and gorm.io/driver/sqlite, shipped as its own module.
Use when switching an existing Go project to gosqlite from mattn/go-sqlite3, modernc.org/sqlite, glebarez/sqlite, gorm.io/driver/sqlite, or ncruces/go-sqlite3.
Use when opening a SQLite database in Go with gosqlite.org, choosing the "sqlite" vs "sqlite3" driver name, or deciding which sub-package to reach for. The starting point for any task using this driver.
SOC 직업 분류 기준
| name | vector-search |
| description | Use when adding semantic, similarity, embedding, or KNN vector search to a Go app with gosqlite — the typed vec.Table API over sqlite-vec, distance metrics, encodings, quantization, and metadata filters. |
Blank-importing vec auto-registers the sqlite-vec extension on every connection.
import (
_ "gosqlite.org"
"gosqlite.org/vec"
)
tbl, _ := vec.Create(ctx, db, "docs", 8, vec.Options{Metric: vec.Cosine})
tbl.BatchInsert(ctx, []vec.Row{{Rowid: 1, Embedding: []float32{...}}})
for m, err := range tbl.KNN(ctx, query, 5) { // streaming iter.Seq2
if err != nil { return err }
fmt.Println(m.Rowid, m.Distance)
}
hits, _ := tbl.KNNSlice(ctx, query, 5) // non-streaming
vec.L2, vec.Cosine, vec.Dot (alias L1), vec.Hamming.vec.Options{Encoding: vec.Binary} or vec.Int8 / vec.Bit for compact storage. int8 needs values in [-1, 1]; bit[N] needs N%8==0.tbl.KNNSlice(ctx, q, k, vec.WithFilter("lang = ?", "en")).vec.CreateKeyed[string](...) → KeyedTable[string].vec.Options{Columns: []vec.Column{{Name: "lang", Type: "text", Kind: vec.Metadata}}}, then insert with tbl.InsertRow(ctx, vec.Row{..., Values: map[string]any{"lang": "en"}}).INSERT OR REPLACE — vec0 rejects it. Use tbl.Update(...) to replace a row.tbl.Insert only works on tables with no extra columns or aux-only.gorm.Transaction is not supported — run KNN outside the transaction.For gorm models, use the tag-driven sidecar (vec:"dim=N;metric=cosine") — see the gorm skill. To combine with keyword search, see hybrid-search. Full reference: docs/guides/vector-search.md.