@flowrag/provider-local 0.0.1

package/README.md

@@ -0,0 +1,48 @@
+# @flowrag/provider-local
+
+Local AI provider for FlowRAG - ONNX embeddings and future local extraction.
+
+## Installation
+
+```bash
+npm install @flowrag/provider-local
+```
+
+## Usage
+
+### Embedder
+
+```typescript
+import { LocalEmbedder } from '@flowrag/provider-local';
+
+const embedder = new LocalEmbedder({
+  model: 'Xenova/e5-small-v2', // optional, default
+  dtype: 'q8', // optional: 'fp32', 'q8', 'q4'
+  device: 'auto', // optional: 'auto', 'cpu', 'gpu'
+});
+
+// Single embedding
+const embedding = await embedder.embed('Hello world');
+
+// Batch embeddings
+const embeddings = await embedder.embedBatch(['Hello', 'World']);
+```
+
+## Supported Models
+
+- `Xenova/e5-small-v2` (384 dims) - Default, fast
+- `Xenova/e5-base-v2` (768 dims) - Better quality
+- `Xenova/e5-large-v2` (1024 dims) - Best quality
+- `Xenova/all-MiniLM-L6-v2` (384 dims) - Compact
+- `Xenova/all-mpnet-base-v2` (768 dims) - Good balance
+
+## Environment Variables
+
+```bash
+FLOWRAG_EMBEDDING_MODEL=Xenova/e5-small-v2
+FLOWRAG_EMBEDDING_DTYPE=q8
+```
+
+## License
+
+MIT

package/dist/index.d.mts

@@ -0,0 +1,23 @@
+import { Embedder } from "@flowrag/core";
+
+//#region src/embedder.d.ts
+interface LocalEmbedderOptions {
+  model?: string;
+  dtype?: "fp32" | "q8" | "q4";
+  device?: "auto" | "cpu" | "gpu";
+}
+declare class LocalEmbedder implements Embedder {
+  readonly modelName: string;
+  readonly dimensions: number;
+  private readonly dtype;
+  private readonly device;
+  private pipeline;
+  constructor(options?: LocalEmbedderOptions);
+  private getModelDimensions;
+  embed(text: string): Promise<number[]>;
+  embedBatch(texts: string[]): Promise<number[][]>;
+  private getPipeline;
+}
+//#endregion
+export { LocalEmbedder, type LocalEmbedderOptions };
+//# sourceMappingURL=index.d.mts.map

package/dist/index.mjs

@@ -0,0 +1,53 @@
+//#region src/embedder.ts
+var LocalEmbedder = class {
+	modelName;
+	dimensions;
+	dtype;
+	device;
+	pipeline = null;
+	constructor(options = {}) {
+		this.modelName = options.model || "Xenova/e5-small-v2";
+		this.dtype = options.dtype || "q8";
+		this.device = options.device || "auto";
+		this.dimensions = this.getModelDimensions(this.modelName);
+	}
+	getModelDimensions(modelName) {
+		return {
+			"Xenova/e5-small-v2": 384,
+			"Xenova/e5-base-v2": 768,
+			"Xenova/e5-large-v2": 1024,
+			"Xenova/all-MiniLM-L6-v2": 384,
+			"Xenova/all-MiniLM-L12-v2": 384,
+			"Xenova/all-mpnet-base-v2": 768
+		}[modelName] || 384;
+	}
+	async embed(text) {
+		const result = await (await this.getPipeline())(text, {
+			pooling: "mean",
+			normalize: true
+		});
+		return Array.from(result.data);
+	}
+	async embedBatch(texts) {
+		if (texts.length === 0) return [];
+		const pipeline = await this.getPipeline();
+		return (await Promise.all(texts.map((text) => pipeline(text, {
+			pooling: "mean",
+			normalize: true
+		})))).map((result) => Array.from(result.data));
+	}
+	async getPipeline() {
+		if (!this.pipeline) {
+			const { pipeline } = await import("@huggingface/transformers");
+			this.pipeline = await pipeline("feature-extraction", this.modelName, {
+				dtype: this.dtype,
+				device: this.device
+			});
+		}
+		return this.pipeline;
+	}
+};
+
+//#endregion
+export { LocalEmbedder };
+//# sourceMappingURL=index.mjs.map

package/dist/index.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.mjs","names":[],"sources":["../src/embedder.ts"],"sourcesContent":["import type { Embedder } from '@flowrag/core';\n\nexport interface LocalEmbedderOptions {\n  model?: string;\n  dtype?: 'fp32' | 'q8' | 'q4';\n  device?: 'auto' | 'cpu' | 'gpu';\n}\n\nexport class LocalEmbedder implements Embedder {\n  readonly modelName: string;\n  readonly dimensions: number;\n  private readonly dtype: string;\n  private readonly device: string;\n  private pipeline: unknown = null;\n\n  constructor(options: LocalEmbedderOptions = {}) {\n    this.modelName = options.model || 'Xenova/e5-small-v2';\n    this.dtype = options.dtype || 'q8';\n    this.device = options.device || 'auto';\n\n    // Set dimensions based on model\n    this.dimensions = this.getModelDimensions(this.modelName);\n  }\n\n  private getModelDimensions(modelName: string): number {\n    // Common embedding model dimensions\n    const dimensionMap: Record<string, number> = {\n      'Xenova/e5-small-v2': 384,\n      'Xenova/e5-base-v2': 768,\n      'Xenova/e5-large-v2': 1024,\n      'Xenova/all-MiniLM-L6-v2': 384,\n      'Xenova/all-MiniLM-L12-v2': 384,\n      'Xenova/all-mpnet-base-v2': 768,\n    };\n\n    return dimensionMap[modelName] || 384; // Default to 384\n  }\n\n  async embed(text: string): Promise<number[]> {\n    const pipeline = (await this.getPipeline()) as (\n      text: string,\n      options: { pooling: string; normalize: boolean },\n    ) => Promise<{ data: ArrayLike<number> }>;\n    const result = await pipeline(text, { pooling: 'mean', normalize: true });\n    return Array.from(result.data);\n  }\n\n  async embedBatch(texts: string[]): Promise<number[][]> {\n    if (texts.length === 0) return [];\n\n    const pipeline = (await this.getPipeline()) as (\n      text: string,\n      options: { pooling: string; normalize: boolean },\n    ) => Promise<{ data: ArrayLike<number> }>;\n    const results = await Promise.all(\n      texts.map((text) => pipeline(text, { pooling: 'mean', normalize: true })),\n    );\n    return results.map((result) => Array.from(result.data));\n  }\n\n  private async getPipeline(): Promise<unknown> {\n    if (!this.pipeline) {\n      const { pipeline } = await import('@huggingface/transformers');\n      this.pipeline = await pipeline('feature-extraction', this.modelName, {\n        dtype: this.dtype as 'fp32' | 'q8' | 'q4',\n        device: this.device as 'auto' | 'cpu' | 'gpu',\n      });\n    }\n    return this.pipeline;\n  }\n}\n"],"mappings":";AAQA,IAAa,gBAAb,MAA+C;CAC7C,AAAS;CACT,AAAS;CACT,AAAiB;CACjB,AAAiB;CACjB,AAAQ,WAAoB;CAE5B,YAAY,UAAgC,EAAE,EAAE;AAC9C,OAAK,YAAY,QAAQ,SAAS;AAClC,OAAK,QAAQ,QAAQ,SAAS;AAC9B,OAAK,SAAS,QAAQ,UAAU;AAGhC,OAAK,aAAa,KAAK,mBAAmB,KAAK,UAAU;;CAG3D,AAAQ,mBAAmB,WAA2B;AAWpD,SAT6C;GAC3C,sBAAsB;GACtB,qBAAqB;GACrB,sBAAsB;GACtB,2BAA2B;GAC3B,4BAA4B;GAC5B,4BAA4B;GAC7B,CAEmB,cAAc;;CAGpC,MAAM,MAAM,MAAiC;EAK3C,MAAM,SAAS,OAJG,MAAM,KAAK,aAAa,EAIZ,MAAM;GAAE,SAAS;GAAQ,WAAW;GAAM,CAAC;AACzE,SAAO,MAAM,KAAK,OAAO,KAAK;;CAGhC,MAAM,WAAW,OAAsC;AACrD,MAAI,MAAM,WAAW,EAAG,QAAO,EAAE;EAEjC,MAAM,WAAY,MAAM,KAAK,aAAa;AAO1C,UAHgB,MAAM,QAAQ,IAC5B,MAAM,KAAK,SAAS,SAAS,MAAM;GAAE,SAAS;GAAQ,WAAW;GAAM,CAAC,CAAC,CAC1E,EACc,KAAK,WAAW,MAAM,KAAK,OAAO,KAAK,CAAC;;CAGzD,MAAc,cAAgC;AAC5C,MAAI,CAAC,KAAK,UAAU;GAClB,MAAM,EAAE,aAAa,MAAM,OAAO;AAClC,QAAK,WAAW,MAAM,SAAS,sBAAsB,KAAK,WAAW;IACnE,OAAO,KAAK;IACZ,QAAQ,KAAK;IACd,CAAC;;AAEJ,SAAO,KAAK"}

package/package.json

@@ -0,0 +1,49 @@
+{
+  "name": "@flowrag/provider-local",
+  "version": "0.0.1",
+  "description": "Local AI provider for FlowRAG - ONNX embeddings and future local extraction",
+  "keywords": [
+    "flowrag",
+    "provider",
+    "local",
+    "onnx",
+    "transformers",
+    "embedder",
+    "ai",
+    "rag"
+  ],
+  "homepage": "https://github.com/Zweer/FlowRAG/tree/main/packages/provider-local",
+  "bugs": "https://github.com/Zweer/FlowRAG/issues",
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/Zweer/FlowRAG.git",
+    "directory": "packages/provider-local"
+  },
+  "license": "MIT",
+  "author": {
+    "name": "Zweer",
+    "email": "n.olivieriachille@gmail.com"
+  },
+  "type": "module",
+  "exports": {
+    ".": "./dist/index.mjs",
+    "./package.json": "./package.json"
+  },
+  "main": "./dist/index.mjs",
+  "module": "./dist/index.mjs",
+  "types": "./dist/index.d.mts",
+  "files": [
+    "dist"
+  ],
+  "dependencies": {
+    "@flowrag/core": "1.0.2",
+    "@huggingface/transformers": "^3.8.1"
+  },
+  "engines": {
+    "node": ">=20.3"
+  },
+  "publishConfig": {
+    "access": "public",
+    "provenance": false
+  }
+}