@lucas-bur/pix 0.7.0 → 0.9.0

package/dist/index.mjs

@@ -23,7 +23,7 @@ var GetStatus = class extends Effect.Service()("GetStatus", {
 		const configStore = yield* ConfigStore;
 		const getStatus = () => Effect.gen(function* () {
 			const status = yield* store.getStatus();
-			const configModel = yield* configStore.readConfig().pipe(Effect.map((c) => c.model), Effect.catchAll(() => Effect.succeed(status.model)));
+			const configModel = yield* configStore.readConfig().pipe(Effect.map((c) => c.embedder.model), Effect.catchAll(() => Effect.succeed(status.model)));
 			return {
 				...status,
 				model: configModel
@@ -33,6 +33,21 @@ var GetStatus = class extends Effect.Service()("GetStatus", {
 	})
 }) {};
 //#endregion
+//#region src/domain/config.ts
+var ConfigError = class extends Data.TaggedError("ConfigError") {};
+const DEFAULT_CONFIG = {
+	schema: "1",
+	chunkLines: 60,
+	overlapLines: 10,
+	chunkConcurrency: 8,
+	files: {},
+	embedder: {
+		model: "Xenova/all-MiniLM-L6-v2",
+		device: "auto",
+		dtype: "fp32"
+	}
+};
+//#endregion
 //#region src/application/index-project.ts
 /**
 * Use case: index project files. Pipeline: scan → chunk → embed → store. Depends on ConfigStore,
@@ -47,6 +62,7 @@ var IndexProject = class extends Effect.Service()("IndexProject", {
 		const embedder = yield* Embedder;
 		const vectorStore = yield* VectorStore;
 		const index = () => Effect.gen(function* () {
+			if (!(yield* configStore.configExists())) yield* configStore.writeConfig(DEFAULT_CONFIG);
 			const config = yield* configStore.readConfig();
 			const extensions = Object.keys(config.files).length > 0 ? Object.keys(config.files) : [
 				".ts",
@@ -89,18 +105,6 @@ var IndexProject = class extends Effect.Service()("IndexProject", {
 	})
 }) {};
 //#endregion
-//#region src/domain/config.ts
-var ConfigError = class extends Data.TaggedError("ConfigError") {};
-const DEFAULT_CONFIG = {
-	schema: "1",
-	model: "Xenova/all-MiniLM-L6-v2",
-	dims: 384,
-	chunkLines: 60,
-	overlapLines: 10,
-	chunkConcurrency: 8,
-	files: {}
-};
-//#endregion
 //#region src/application/init-project.ts
 /**
 * Use case: initialize a pix project by writing default config. Depends on ConfigStore via Effect
@@ -167,11 +171,17 @@ const codeFromError = (error) => {
 	if (error && typeof error === "object" && "_tag" in error) return errorCodes[String(error._tag)] ?? "UNKNOWN";
 	return "UNKNOWN";
 };
+const causeFromError = (error) => {
+	if (typeof error === "string") return error;
+	if (error && typeof error === "object" && "cause" in error) return String(error.cause);
+	return "Unknown cause";
+};
 /** Format an error as spec-mandated JSON: `{ error: true, code: "...", message: "..." }`. */
 const formatError = (error) => JSON.stringify({
 	error: true,
 	code: codeFromError(error),
-	message: messageFromError(error)
+	message: messageFromError(error),
+	cause: causeFromError(error)
 });
 /** Log the error as JSON to stdout, then re-fail to preserve non-zero exit code. */
 const reportError = (error) => Console.log(formatError(error)).pipe(Effect.flatMap(() => Effect.fail(error)));
@@ -474,9 +484,34 @@ const make$3 = Effect.gen(function* () {
 });
 const ConfigStoreLive = Layer.effect(ConfigStore, make$3);
 //#endregion
+//#region src/domain/models.ts
+/** Registry of supported embedding models. */
+const MODEL_REGISTRY = {
+	"Xenova/all-MiniLM-L6-v2": {
+		id: "Xenova/all-MiniLM-L6-v2",
+		dims: 384,
+		dtypes: [
+			"fp32",
+			"fp16",
+			"q8",
+			"q4"
+		],
+		description: "General-purpose sentence embeddings, 23MB q8"
+	},
+	"Xenova/bge-small-en-v1.5": {
+		id: "Xenova/bge-small-en-v1.5",
+		dims: 384,
+		dtypes: [
+			"fp32",
+			"fp16",
+			"q8",
+			"q4"
+		],
+		description: "BGE retrieval-optimized embeddings, 34MB q8"
+	}
+};
+//#endregion
 //#region src/services/embedder.ts
-const MODEL_NAME = "Xenova/all-MiniLM-L6-v2";
-const DIMS = 384;
 const CACHE_DIR = ".pix/cache";
 const BATCH_SIZE = 16;
 env.cacheDir = CACHE_DIR;
@@ -489,18 +524,51 @@ const normalize = (arr) => {
 	for (let i = 0; i < arr.length; i++) result[i] = arr[i] / norm;
 	return result;
 };
+const resolveEmbedderConfig = (configStore) => Effect.gen(function* () {
+	const config = yield* configStore.readConfig().pipe(Effect.catchAll(() => Effect.succeed(void 0)));
+	const model = config?.embedder.model ?? "Xenova/all-MiniLM-L6-v2";
+	const device = config?.embedder.device ?? "auto";
+	const dtype = config?.embedder.dtype ?? "fp32";
+	const modelInfo = MODEL_REGISTRY[model];
+	if (!modelInfo) return yield* new ModelLoadError({
+		message: `Unknown embedding model "${model}". Available: ${Object.keys(MODEL_REGISTRY).join(", ")}`,
+		model
+	});
+	if (!modelInfo.dtypes.includes(dtype)) return yield* new ModelLoadError({
+		message: `Unsupported dtype "${dtype}" for model "${model}". Supported: ${modelInfo.dtypes.join(", ")}`,
+		model
+	});
+	return {
+		model,
+		device,
+		dtype,
+		dims: modelInfo.dims
+	};
+});
+const createExtractor = (opts) => Effect.tryPromise(async () => {
+	const { pipeline } = await import("@huggingface/transformers");
+	return pipeline("feature-extraction", opts.model, {
+		device: opts.device,
+		dtype: opts.dtype
+	});
+}).pipe(Effect.mapError((cause) => new ModelLoadError({
+	message: `Failed to load embedding model with device "${opts.device}"`,
+	model: opts.model,
+	cause
+})));
+const createExtractorWithFallback = (opts) => {
+	if (opts.device === "cpu") return createExtractor(opts);
+	return createExtractor(opts).pipe(Effect.catchAll((originalError) => Effect.gen(function* () {
+		yield* Effect.logWarning(`Embedding device "${opts.device}" failed, falling back to "cpu": ${originalError.message}`);
+		return yield* createExtractor({
+			...opts,
+			device: "cpu"
+		}).pipe(Effect.catchAll(() => Effect.fail(originalError)));
+	})));
+};
 const make$2 = Effect.gen(function* () {
-	const getExtractor = yield* Effect.cached(Effect.tryPromise(async () => {
-		const { pipeline } = await import("@huggingface/transformers");
-		return pipeline("feature-extraction", MODEL_NAME, {
-			device: "cpu",
-			dtype: "q8"
-		});
-	}).pipe(Effect.mapError((cause) => new ModelLoadError({
-		message: "Failed to load embedding model",
-		model: MODEL_NAME,
-		cause
-	}))));
+	const cfg = yield* resolveEmbedderConfig(yield* ConfigStore);
+	const getExtractor = yield* Effect.cached(createExtractorWithFallback(cfg));
 	const embed = (text) => Effect.gen(function* () {
 		const extractor = yield* getExtractor;
 		const data = (yield* Effect.tryPromise(() => extractor(text, {
@@ -512,7 +580,7 @@ const make$2 = Effect.gen(function* () {
 		})))).data;
 		return {
 			vector: normalize(data),
-			dims: DIMS
+			dims: cfg.dims
 		};
 	});
 	const batch = (texts) => Effect.gen(function* () {
@@ -530,13 +598,13 @@ const make$2 = Effect.gen(function* () {
 			const data = tensor.data;
 			const n = tensor.dims[0];
 			for (let j = 0; j < n; j++) {
-				const offset = j * DIMS;
-				results.push(normalize(data.slice(offset, offset + DIMS)));
+				const offset = j * cfg.dims;
+				results.push(normalize(data.slice(offset, offset + cfg.dims)));
 			}
 		}
 		return results.map((vector) => ({
 			vector,
-			dims: DIMS
+			dims: cfg.dims
 		}));
 	});
 	return {
@@ -544,7 +612,7 @@ const make$2 = Effect.gen(function* () {
 		batch
 	};
 });
-const OnnxEmbedderLive = Layer.effect(Embedder, make$2);
+const OnnxEmbedderLive = Layer.provideMerge(Layer.effect(Embedder, make$2), ConfigStoreLive);
 //#endregion
 //#region src/services/scanner.ts
 const ALWAYS_IGNORE = new Set([

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lucas-bur/pix",
-  "version": "0.7.0",
+  "version": "0.9.0",
   "description": "Lightweight local semantic project indexer",
   "keywords": [
     "cli",