@infino-ai/infino 0.1.0 → 0.1.1

package/README.md

@@ -144,18 +144,22 @@ vecs.vectorSearch("emb", queryVector, 10, {
 
 ## Hybrid search
 
-Combine BM25 and vector search in **one query** with the `hybrid_search` table
-function — a single pass over both indexes, fused inside the engine (no separate
+Combine BM25 and vector search in **one call** — a single pass over both
+indexes, fused inside the engine with reciprocal-rank fusion (no separate
 reranker service, no two round-trips). Keyword-only search misses paraphrases;
 vector-only search misses exact terms — hybrid gets both. Results come back
-best-first with a fused `score`.
+best-first with a fused `score` (higher is better).
 
 ```javascript
 const spec = new IndexSpec().fts("body").vector("emb", 384, 256, "cosine");
 const docs = db.createTable("docs", { body: "large_utf8", emb: { vector: 384 } }, spec);
 docs.append([{ body: "To cancel a subscription, open Settings then Billing.", emb: embed(/* … */) }]);
 
-// hybrid_search(table, text_col, query_text, vec_col, query_vec, k)
+// hybridSearch(textColumn, textQuery, vectorColumn, vectorQuery, k, opts?)
+docs.hybridSearch("body", "cancel subscription", "emb", embed("how do I stop my plan?"), 10);
+// opts: { mode } tunes the BM25 side, { nprobe } the vector side.
+
+// The same fusion is also a SQL table function, so it composes in a query:
 const qvec = embed("how do I stop my plan?").join(",");
 db.querySql(
   `SELECT _id, score FROM hybrid_search('docs', 'body', 'cancel subscription', 'emb', '${qvec}', 10)`,
@@ -226,24 +230,51 @@ docs.optimize({ targetSuperfileSizeMb: 256, minFillPercent: 50 });
 
 `connect` selects the backend from the URI:
 
-| URI                   | Backend                                  |
-| --------------------- | ---------------------------------------- |
-| `./data`, `/abs/path` | Local filesystem                         |
-| `s3://bucket/prefix`  | Amazon S3 / S3-compatible object storage |
-| `memory://`           | In-process, ephemeral (testing)          |
+| URI                      | Backend                                  |
+| ------------------------ | ---------------------------------------- |
+| `./data`, `/abs/path`    | Local filesystem                         |
+| `s3://bucket/prefix`     | Amazon S3 / S3-compatible object storage |
+| `az://container/prefix`  | Azure Blob Storage                       |
+| `memory://`              | In-process, ephemeral (testing)          |
 
-For S3-compatible stores that need an explicit endpoint and static credentials,
-pass them in `options` (omit to use ambient AWS credentials):
+Credentials go in `storageOptions`, keyed by the standard `object_store` config
+strings (`aws_*` / `azure_*` — the same names the AWS and Azure SDKs use). Omit
+them to use ambient cloud identity (IAM instance role / managed identity);
+infino reads no credentials from the environment.
 
 ```javascript
+// S3
 const db = connect("s3://bucket/prefix", {
-  endpoint: "https://s3.example.com",
-  region: "us-east-1",
-  accessKey: "…",
-  secretKey: "…",
+  storageOptions: {
+    aws_access_key_id: "…",
+    aws_secret_access_key: "…",
+    aws_region: "us-east-1",
+  },
+});
+
+// Azure
+const db = connect("az://container/prefix", {
+  storageOptions: {
+    azure_storage_account_name: "…",
+    azure_storage_account_key: "…",
+  },
 });
 ```
 
+Common keys:
+
+| Backend | Keys |
+| ------- | ---- |
+| S3      | `aws_access_key_id`, `aws_secret_access_key`, `aws_region`, `aws_session_token`, `aws_endpoint` |
+| Azure   | `azure_storage_account_name`, `azure_storage_account_key`, `azure_storage_sas_key`, `azure_storage_client_id`, `azure_storage_client_secret`, `azure_storage_tenant_id` |
+
+The full set is whatever `object_store` accepts for the backend; an unknown key
+is rejected at `connect`. Pass `validate: true` to probe the backend at
+`connect`, so wrong credentials or an unreachable bucket throw there instead of
+on the first query. For an S3-compatible endpoint (MinIO / R2 / Ceph), set
+`aws_endpoint` (with `aws_allow_http: "true"` for plain HTTP) alongside the
+credentials.
+
 ### Local disk cache
 
 For object-storage-backed catalogs, a local disk cache keeps hot data on fast
@@ -273,9 +304,9 @@ const db = connect("s3://bucket/prefix", {
 ## API reference
 
 - `connect(uri, options?)` — backend from the URI scheme. `options`:
-  S3-compatible credentials (`endpoint`, `region`, `accessKey`, `secretKey` —
-  `endpoint` requires the other three) and, for remote-backed tables, a local
-  disk cache (`cacheDir`, `cacheBudgetBytes`, `coldFetchMode`).
+  `storageOptions` (credentials, keyed by `object_store`'s `aws_*` / `azure_*`
+  strings), `validate`, and a local disk cache (`cacheDir`, `cacheBudgetBytes`,
+  `coldFetchMode`).
 - `Connection`
   - `createTable(name, schema, IndexSpec)` / `openTable(name)` /
     `dropTable(name, purge?)` (`purge = true` also deletes the data) /

package/infino/index.d.ts

@@ -17,17 +17,23 @@ export type SchemaDescriptor = Record<string, string | {
 export type AppendData = RowRecord[] | arrow.Table | arrow.RecordBatch | Buffer | Uint8Array;
 /** Storage and cache config the `connect` URI can't carry. All optional. */
 export interface ConnectOptions {
-    /** S3-compatible endpoint; requires `region`, `accessKey`, `secretKey`. */
-    endpoint?: string;
-    region?: string;
-    accessKey?: string;
-    secretKey?: string;
+    /**
+     * Credentials/tuning for the URI-selected backend, keyed by `object_store`
+     * config strings (`aws_*` / `azure_*`). An unknown key is rejected at
+     * `connect`.
+     */
+    storageOptions?: Record<string, string>;
     /** Local disk-cache directory for remote-backed tables. */
     cacheDir?: string;
     /** Disk-cache budget in bytes. */
     cacheBudgetBytes?: number;
     /** How cold misses are serviced. */
     coldFetchMode?: "hybrid_with_prefetch" | "range_only" | "lazy_foreground_with_background_fill";
+    /**
+     * Probe the object store at `connect` (default `false`). `true` fails fast
+     * on bad credentials instead of on the first table operation.
+     */
+    validate?: boolean;
 }
 /** Row counts returned by `update` / `delete`. */
 export interface MutationStats {
@@ -74,6 +80,16 @@ export interface VectorSearchOptions {
     /** Restrict the kNN to rows matching a text predicate (pushdown pre-filter). */
     filter?: VectorFilter;
 }
+/** Options for `hybridSearch`. `mode` applies to the BM25 side; `nprobe` to
+ * the vector side. */
+export interface HybridSearchOptions {
+    /** BM25 boolean mode: `"or"` (default) or `"and"`. */
+    mode?: BoolMode;
+    /** IVF partitions to probe on the vector side (higher = better recall). */
+    nprobe?: number;
+    projection?: string[];
+    arrow?: boolean;
+}
 export interface TokenMatchOptions {
     mode?: BoolMode;
     projection?: string[];
@@ -107,6 +123,13 @@ export declare class Table {
         arrow: true;
     }): arrow.Table;
     vectorSearch(column: string, query: number[] | Float32Array, k: number, opts?: VectorSearchOptions): RowRecord[];
+    /** Hybrid BM25 + vector search, fused with reciprocal-rank fusion; rows as
+     * records (or an Arrow `Table`). `score` is the fused RRF score (higher is
+     * better). */
+    hybridSearch(textColumn: string, textQuery: string, vectorColumn: string, vectorQuery: number[] | Float32Array, k: number, opts: HybridSearchOptions & {
+        arrow: true;
+    }): arrow.Table;
+    hybridSearch(textColumn: string, textQuery: string, vectorColumn: string, vectorQuery: number[] | Float32Array, k: number, opts?: HybridSearchOptions): RowRecord[];
     /** Unranked token match; matching rows as records (or an Arrow `Table`). */
     tokenMatch(column: string, query: string, opts: TokenMatchOptions & {
         arrow: true;

package/infino/index.js

@@ -216,6 +216,11 @@ class Table {
         const buf = this.inner.vectorSearch(column, q, k, opts.nprobe, opts.rerankMult, opts.projection, opts.filter);
         return decode(buf, opts.arrow);
     }
+    hybridSearch(textColumn, textQuery, vectorColumn, vectorQuery, k, opts = {}) {
+        const q = vectorQuery instanceof Float32Array ? vectorQuery : Float32Array.from(vectorQuery);
+        const buf = this.inner.hybridSearch(textColumn, textQuery, vectorColumn, q, k, opts.mode, opts.nprobe, opts.projection);
+        return decode(buf, opts.arrow);
+    }
     tokenMatch(column, query, opts = {}) {
         const buf = this.inner.tokenMatch(column, query, opts.mode, opts.projection);
         return decode(buf, opts.arrow);

package/infino/native.d.ts

@@ -9,11 +9,12 @@
  * disk cache.
  */
 export interface ConnectOptions {
-  /** S3-compatible endpoint; requires `region`, `accessKey`, `secretKey`. */
-  endpoint?: string
-  region?: string
-  accessKey?: string
-  secretKey?: string
+  /**
+   * Credentials/tuning for the URI-selected backend, keyed by
+   * `object_store` config strings (`aws_*` / `azure_*`). An unknown key
+   * is rejected at `connect`.
+   */
+  storageOptions?: Record<string, string>
   /** Local disk-cache directory for remote-backed tables. */
   cacheDir?: string
   /** Disk-cache budget in bytes (a JS number; up to 2^53). */
@@ -23,6 +24,11 @@ export interface ConnectOptions {
    * `"lazy_foreground_with_background_fill"`.
    */
   coldFetchMode?: string
+  /**
+   * Probe the object store at `connect` (default `false`). `true` fails
+   * fast on bad credentials instead of on first use.
+   */
+  validate?: boolean
 }
 /** Tuning for `optimize`; all fields optional (omitted ⇒ engine default). */
 export interface OptimizeOptions {
@@ -57,8 +63,10 @@ export interface VectorFilter {
 }
 /**
  * Open (or create) a catalog rooted at `uri` (local dir, `memory://`, or
- * object-store prefix). S3-compatible static credentials are passed via
- * `options` (the JS-idiomatic form of the Rust `ConnectOptions`).
+ * object-store prefix). Credentials are passed via `options.storageOptions`
+ * (the JS-idiomatic form of the Rust `ConnectOptions`). Pass `validate: true`
+ * to probe object stores at connect (off by default) so bad credentials fail
+ * there rather than on the first table operation.
  */
 export declare function connect(uri: string, options?: ConnectOptions | undefined | null): Connection
 /**
@@ -141,6 +149,14 @@ export declare class Table {
    * columns (omit for full rows).
    */
   exactMatch(column: string, value: string, projection?: Array<string> | undefined | null): Buffer
+  /**
+   * Hybrid BM25 + vector search fused with reciprocal-rank fusion.
+   * `text_column`/`text_query` (under `mode`) drive BM25; `vector_column`/
+   * `vector_query` (a `Float32Array`, with optional `nprobe`) drive vector
+   * kNN. Returns Arrow rows like [`Table::bm25_search`], with `score` the
+   * fused RRF score (higher is better); `projection` selects columns.
+   */
+  hybridSearch(textColumn: string, textQuery: string, vectorColumn: string, vectorQuery: Float32Array, k: number, mode?: string | undefined | null, nprobe?: number | undefined | null, projection?: Array<string> | undefined | null): Buffer
   /**
    * Delete every row matching a SQL `predicate` (e.g. `"status = 'spam'"`),
    * returning the mutation counts. Requires durable storage — a `memory://`

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@infino-ai/infino",
-  "version": "0.1.0",
+  "version": "0.1.1",
   "description": "Fast search on object storage — SQL, full-text, and vector search.",
   "license": "Apache-2.0",
   "publishConfig": {
@@ -82,12 +82,12 @@
     "apache-arrow": "^17"
   },
   "optionalDependencies": {
-    "infx-darwin-x64": "0.1.0",
-    "infx-darwin-arm64": "0.1.0",
-    "infx-linux-x64-gnu": "0.1.0",
-    "infx-linux-arm64-gnu": "0.1.0",
-    "infx-linux-x64-musl": "0.1.0",
-    "infx-linux-arm64-musl": "0.1.0"
+    "infx-darwin-x64": "0.1.1",
+    "infx-darwin-arm64": "0.1.1",
+    "infx-linux-x64-gnu": "0.1.1",
+    "infx-linux-arm64-gnu": "0.1.1",
+    "infx-linux-x64-musl": "0.1.1",
+    "infx-linux-arm64-musl": "0.1.1"
   },
   "devDependencies": {
     "@napi-rs/cli": "^2.18.0",
@@ -99,4 +99,4 @@
     "infino/native.js",
     "infino/native.d.ts"
   ]
-}
+}