벡터 DB와 검색을 API 하나로. 운영은 Schift가.

import { VectorClient } from "@schift-io/sdk";

const vectors = new VectorClient({
  apiKey: process.env.SCHIFT_API_KEY!,
  dataUrl: "https://gw.schift.io",
});
const where = "custom/catalog/items";
const model = "google/gemini-embedding-2"; // any model; every write asserts it

await vectors.declare(where, {
  embedding: "external",
  vectors: { text: { dimension: 2048, model } },
  metadata: { sku: "string", year: "int64", tags: "string_list" },
});

await vectors.upsert(
  where,
  [{ id: "sku-1", text: "linen shirt, warm beige",
     metadata: { sku: "S1", year: 2025, tags: ["summer"] },
     vectors: { text: embedding } }], // Float32Array goes as base64 f32
  { models: { text: model } },
);

const { hits } = await vectors.search(where, {
  vector: "text",
  queryVector: queryEmbedding,
  query: "linen shirt", // hybrid: cosine + BM25
  filter: { year: { gte: 2024 }, tags: { contains: "summer" } },
  topK: 10,
}); // [{ id, score, scores: { text, keyword }, text, metadata, updatedAt }]

import os
import numpy as np
from schift.vectors import VectorsClient

vectors = VectorsClient(os.environ["SCHIFT_API_KEY"], data_url="https://gw.schift.io")
where = "custom/catalog/items"
model = "google/gemini-embedding-2"  # any model; every write asserts it

vectors.declare(
    where,
    vectors={"text": {"dimension": 2048, "model": model}},
    metadata={"sku": "string", "year": "int64", "tags": "string_list"},
)

vectors.upsert(where, [{
    "id": "sku-1",
    "text": "linen shirt, warm beige",
    "metadata": {"sku": "S1", "year": 2025, "tags": ["summer"]},
    "vectors": {"text": np.asarray(embedding, dtype=np.float32)},
}], models={"text": model})

hits = vectors.search(
    where,
    vector="text",
    queryVector=np.asarray(query_embedding, dtype=np.float32),
    query="linen shirt",
    filter={"year": {"gte": 2024}, "tags": {"contains": "summer"}},
    topK=10,
)["hits"]

curl -X PUT https://gw.schift.io/v1/vectors/custom/catalog/items/schema \
  -H "Authorization: Bearer $SCHIFT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"embedding":"external",
       "vectors":{"text":{"dimension":2048,"model":"google/gemini-embedding-2"}},
       "metadata":{"sku":"string","year":"int64","tags":"string_list"}}'
# 201 {"where":"custom/catalog/items","schema":{...},"created":true}

curl -X POST https://gw.schift.io/v1/vectors/custom/catalog/items/search \
  -H "Authorization: Bearer $SCHIFT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"vector":"text","queryVector":[0.013, -0.021, ...],"topK":10,
       "query":"linen shirt","filter":{"year":{"gte":2024}}}'
# {"hits":[{"id","score","scores":{"text":...,"keyword":...},"text","metadata","updatedAt"}],...}

import { DataClient } from "@schift-io/sdk";

const data = new DataClient({
  apiKey: process.env.SCHIFT_API_KEY!,
  dataUrl: "https://gw.schift.io",
});

await data.declareSchema("custom/support/tickets", [
  { name: "ticket_id", type: "string", nullable: false },
  { name: "body", type: "string", nullable: false },
  { name: "plan", type: "string", nullable: true },
]);

// 202 once the rows are on disk; the body column is embedded right after
await data.put("custom/support/tickets", [
  { ticket_id: "t-81", body: "Refund still pending after 10 days", plan: "pro" },
  { ticket_id: "t-82", body: "Invoice shows the wrong company name", plan: "free" },
], { embed: "body" });

const { rows } = await data.search("custom/support/tickets", {
  near: "late refund",   // the query text is embedded with the same model
  k: 5,
  where: { plan: "pro" },
}); // rows[i]._score = cosine similarity

curl -X POST "https://gw.schift.io/v1/data/custom/support/tickets?embed=body" \
  -H "Authorization: Bearer $SCHIFT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '[{"ticket_id":"t-81","body":"Refund still pending after 10 days","plan":"pro"}]'
# 202 {"accepted":1}

curl "https://gw.schift.io/v1/data/custom/support/tickets?near=late%20refund&k=5&f.plan=pro" \
  -H "Authorization: Bearer $SCHIFT_API_KEY"
# {"rows":[{"ticket_id":"t-81", ..., "_score":...}],"next_cursor":null}