@zvec/zvec-grep 0.1.0

package/dist/engine/models/catalog.d.ts

@@ -0,0 +1,47 @@
+import type { EmbeddingCatalogEntry, ModelRef } from "./types.js";
+export declare const MODEL_CATALOG: {
+    readonly openai: {
+        readonly embedding: readonly ["text-embedding-3-small", "text-embedding-3-large"];
+        readonly ranking: readonly [];
+    };
+    readonly qwen: {
+        readonly embedding: readonly ["text-embedding-v4", "qwen3-vl-embedding"];
+        readonly ranking: readonly ["gte-rerank-v2"];
+    };
+    readonly jina: {
+        readonly embedding: readonly ["jina-embeddings-v3"];
+        readonly ranking: readonly ["jina-reranker-v2-base-multilingual"];
+    };
+    readonly local: {
+        readonly embedding: readonly ["embeddinggemma-300m", "qwen3-embedding-0.6b"];
+        readonly ranking: readonly [];
+    };
+};
+export declare const EMBEDDING_MODEL_CATALOG: {
+    readonly "local/embeddinggemma-300m": {
+        readonly id: "local/embeddinggemma-300m";
+        readonly provider: "local";
+        readonly model: "embeddinggemma-300m";
+        readonly uri: "hf:ggml-org/embeddinggemma-300M-GGUF/embeddinggemma-300M-Q8_0.gguf";
+        readonly dimension: 768;
+        readonly metric: "cosine";
+        readonly format: "embeddinggemma";
+        readonly contextSize: 2048;
+        readonly maxBatchSize: 16;
+    };
+    readonly "local/qwen3-embedding-0.6b": {
+        readonly id: "local/qwen3-embedding-0.6b";
+        readonly provider: "local";
+        readonly model: "qwen3-embedding-0.6b";
+        readonly uri: "hf:Qwen/Qwen3-Embedding-0.6B-GGUF/Qwen3-Embedding-0.6B-Q8_0.gguf";
+        readonly dimension: 1024;
+        readonly metric: "cosine";
+        readonly format: "qwen3";
+        readonly contextSize: 8192;
+        readonly maxBatchSize: 8;
+    };
+};
+export type EmbeddingModelCatalogId = keyof typeof EMBEDDING_MODEL_CATALOG;
+export declare function listEmbeddingModels(): EmbeddingCatalogEntry[];
+export declare function getEmbeddingModelCatalogEntry(id: string): EmbeddingCatalogEntry | undefined;
+export declare function getEmbeddingModelCatalogEntryByRef(ref: ModelRef): EmbeddingCatalogEntry | undefined;

package/dist/engine/models/catalog.js

@@ -0,0 +1,68 @@
+export const MODEL_CATALOG = {
+    openai: {
+        embedding: [
+            "text-embedding-3-small",
+            "text-embedding-3-large",
+        ],
+        ranking: [],
+    },
+    qwen: {
+        embedding: [
+            "text-embedding-v4",
+            "qwen3-vl-embedding",
+        ],
+        ranking: [
+            "gte-rerank-v2",
+        ],
+    },
+    jina: {
+        embedding: [
+            "jina-embeddings-v3",
+        ],
+        ranking: [
+            "jina-reranker-v2-base-multilingual",
+        ],
+    },
+    local: {
+        embedding: [
+            "embeddinggemma-300m",
+            "qwen3-embedding-0.6b",
+        ],
+        ranking: [],
+    },
+};
+export const EMBEDDING_MODEL_CATALOG = {
+    "local/embeddinggemma-300m": {
+        id: "local/embeddinggemma-300m",
+        provider: "local",
+        model: "embeddinggemma-300m",
+        uri: "hf:ggml-org/embeddinggemma-300M-GGUF/embeddinggemma-300M-Q8_0.gguf",
+        dimension: 768,
+        metric: "cosine",
+        format: "embeddinggemma",
+        contextSize: 2048,
+        maxBatchSize: 16,
+    },
+    "local/qwen3-embedding-0.6b": {
+        id: "local/qwen3-embedding-0.6b",
+        provider: "local",
+        model: "qwen3-embedding-0.6b",
+        uri: "hf:Qwen/Qwen3-Embedding-0.6B-GGUF/Qwen3-Embedding-0.6B-Q8_0.gguf",
+        dimension: 1024,
+        metric: "cosine",
+        format: "qwen3",
+        contextSize: 8192,
+        maxBatchSize: 8,
+    },
+};
+export function listEmbeddingModels() {
+    return Object.values(EMBEDDING_MODEL_CATALOG);
+}
+export function getEmbeddingModelCatalogEntry(id) {
+    return EMBEDDING_MODEL_CATALOG[id];
+}
+export function getEmbeddingModelCatalogEntryByRef(ref) {
+    return listEmbeddingModels().find((entry) => entry.provider === ref.provider
+        && entry.model === ref.model);
+}
+//# sourceMappingURL=catalog.js.map

package/dist/engine/models/catalog.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"catalog.js","sourceRoot":"","sources":["../../../src/engine/models/catalog.ts"],"names":[],"mappings":"AAOA,MAAM,CAAC,MAAM,aAAa,GAAG;IAC3B,MAAM,EAAE;QACN,SAAS,EAAE;YACT,wBAAwB;YACxB,wBAAwB;SACzB;QACD,OAAO,EAAE,EAAE;KACZ;IAED,IAAI,EAAE;QACJ,SAAS,EAAE;YACT,mBAAmB;YACnB,oBAAoB;SACrB;QACD,OAAO,EAAE;YACP,eAAe;SAChB;KACF;IAED,IAAI,EAAE;QACJ,SAAS,EAAE;YACT,oBAAoB;SACrB;QACD,OAAO,EAAE;YACP,oCAAoC;SACrC;KACF;IAED,KAAK,EAAE;QACL,SAAS,EAAE;YACT,qBAAqB;YACrB,sBAAsB;SACvB;QACD,OAAO,EAAE,EAAE;KACZ;CAC8B,CAAC;AAGlC,MAAM,CAAC,MAAM,uBAAuB,GAAG;IACrC,2BAA2B,EAAE;QAC3B,EAAE,EAAE,2BAA2B;QAC/B,QAAQ,EAAE,OAAO;QACjB,KAAK,EAAE,qBAAqB;QAC5B,GAAG,EAAE,oEAAoE;QACzE,SAAS,EAAE,GAAG;QACd,MAAM,EAAE,QAAQ;QAChB,MAAM,EAAE,gBAAgB;QACxB,WAAW,EAAE,IAAI;QACjB,YAAY,EAAE,EAAE;KACjB;IAED,4BAA4B,EAAE;QAC5B,EAAE,EAAE,4BAA4B;QAChC,QAAQ,EAAE,OAAO;QACjB,KAAK,EAAE,sBAAsB;QAC7B,GAAG,EAAE,kEAAkE;QACvE,SAAS,EAAE,IAAI;QACf,MAAM,EAAE,QAAQ;QAChB,MAAM,EAAE,OAAO;QACf,WAAW,EAAE,IAAI;QACjB,YAAY,EAAE,CAAC;KAChB;CACuD,CAAC;AAM3D,MAAM,UAAU,mBAAmB;IACjC,OAAO,MAAM,CAAC,MAAM,CAAC,uBAAuB,CAAC,CAAC;AAChD,CAAC;AAGD,MAAM,UAAU,6BAA6B,CAC3C,EAAU;IAEV,OAAO,uBAAuB,CAAC,EAA6B,CAAC,CAAC;AAChE,CAAC;AAGD,MAAM,UAAU,kCAAkC,CAChD,GAAa;IAEb,OAAO,mBAAmB,EAAE,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAC1C,KAAK,CAAC,QAAQ,KAAK,GAAG,CAAC,QAAQ;WAC5B,KAAK,CAAC,KAAK,KAAK,GAAG,CAAC,KAAK,CAC7B,CAAC;AACJ,CAAC"}

package/dist/engine/models/embeddings.d.ts

@@ -0,0 +1,25 @@
+import type { Content, ContentKind } from "../types.js";
+import type { ModelRef, VectorMetric } from "./types.js";
+export type EmbeddingVector = number[];
+export type EmbeddingPurpose = "document" | "query";
+export type EmbeddingOptions = {
+    purpose?: EmbeddingPurpose;
+};
+export type EmbeddingLimits = {
+    maxBatchSize: number;
+    maxInputTokens?: number;
+    maxImageBytes?: number;
+};
+export declare abstract class EmbeddingModel {
+    abstract readonly ref: ModelRef;
+    abstract readonly dimension: number;
+    abstract readonly metric: VectorMetric;
+    abstract readonly supportedContentKinds: readonly ContentKind[];
+    abstract readonly limits: EmbeddingLimits;
+    readonly recommendedIndexConcurrency?: number;
+    embed(contents: readonly Content[], options?: EmbeddingOptions): Promise<EmbeddingVector[]>;
+    dispose(): Promise<void>;
+    protected abstract doEmbed(contents: readonly Content[], options: Required<EmbeddingOptions>): Promise<EmbeddingVector[]>;
+    private validateContents;
+    private validateVectors;
+}

package/dist/engine/models/embeddings.js

@@ -0,0 +1,100 @@
+import { EngineError } from "../errors/index.js";
+export class EmbeddingModel {
+    recommendedIndexConcurrency;
+    async embed(contents, options = {}) {
+        this.validateContents(contents);
+        const vectors = await this.doEmbed(contents, {
+            purpose: options.purpose ?? "document",
+        });
+        this.validateVectors(contents, vectors);
+        return vectors;
+    }
+    async dispose() {
+        // Most embedding providers do not hold local resources.
+    }
+    validateContents(contents) {
+        if (contents.length === 0) {
+            throw new EngineError("Embedding requires at least one content item", {
+                code: "ZVEC_GREP.ENGINE.MODELS.EMBEDDING_EMPTY_INPUT"
+            });
+        }
+        if (contents.length > this.limits.maxBatchSize) {
+            throw new EngineError("Embedding batch size exceeds model limit", {
+                code: "ZVEC_GREP.ENGINE.MODELS.EMBEDDING_BATCH_TOO_LARGE",
+                context: `model=${this.ref.model} batchSize=${contents.length} maxBatchSize=${this.limits.maxBatchSize}`,
+            });
+        }
+        for (const [index, content] of contents.entries()) {
+            if (!this.supportedContentKinds.includes(content.kind)) {
+                throw new EngineError("Embedding model does not support content kind", {
+                    code: "ZVEC_GREP.ENGINE.MODELS.EMBEDDING_UNSUPPORTED_CONTENT",
+                    context: `model=${this.ref.model} index=${index} kind=${content.kind}`,
+                });
+            }
+            if (content.kind === "text" && content.text.trim().length === 0) {
+                throw new EngineError("Embedding text content must not be empty", {
+                    code: "ZVEC_GREP.ENGINE.MODELS.EMBEDDING_EMPTY_TEXT",
+                    context: `model=${this.ref.model} index=${index}`,
+                });
+            }
+            if (content.kind === "image") {
+                if (content.data.byteLength === 0) {
+                    throw new EngineError("Embedding image content must not be empty", {
+                        code: "ZVEC_GREP.ENGINE.MODELS.EMBEDDING_EMPTY_IMAGE",
+                        context: `model=${this.ref.model} index=${index}`,
+                    });
+                }
+                if (content.format.trim().length === 0) {
+                    throw new EngineError("Embedding image content must include a format", {
+                        code: "ZVEC_GREP.ENGINE.MODELS.EMBEDDING_MISSING_IMAGE_FORMAT",
+                        context: `model=${this.ref.model} index=${index}`,
+                    });
+                }
+                if (this.limits.maxImageBytes !== undefined
+                    && content.data.byteLength > this.limits.maxImageBytes) {
+                    throw new EngineError("Embedding image content exceeds model limit", {
+                        code: "ZVEC_GREP.ENGINE.MODELS.EMBEDDING_IMAGE_TOO_LARGE",
+                        context: `model=${this.ref.model} index=${index} imageBytes=${content.data.byteLength} maxImageBytes=${this.limits.maxImageBytes}`,
+                    });
+                }
+            }
+        }
+    }
+    validateVectors(contents, vectors) {
+        if (!Array.isArray(vectors)) {
+            throw new EngineError("Embedding model returned a non-array response", {
+                code: "ZVEC_GREP.ENGINE.MODELS.EMBEDDING_INVALID_RESPONSE",
+                context: `model=${this.ref.model}`,
+            });
+        }
+        if (vectors.length !== contents.length) {
+            throw new EngineError("Embedding model returned the wrong number of vectors", {
+                code: "ZVEC_GREP.ENGINE.MODELS.EMBEDDING_VECTOR_COUNT_MISMATCH",
+                context: `model=${this.ref.model} contentCount=${contents.length} vectorCount=${vectors.length}`,
+            });
+        }
+        for (const [vectorIndex, vector] of vectors.entries()) {
+            if (!Array.isArray(vector)) {
+                throw new EngineError("Embedding model returned a non-array vector", {
+                    code: "ZVEC_GREP.ENGINE.MODELS.EMBEDDING_INVALID_VECTOR",
+                    context: `model=${this.ref.model} vectorIndex=${vectorIndex}`,
+                });
+            }
+            if (vector.length !== this.dimension) {
+                throw new EngineError("Embedding model returned a vector with the wrong dimension", {
+                    code: "ZVEC_GREP.ENGINE.MODELS.EMBEDDING_DIMENSION_MISMATCH",
+                    context: `model=${this.ref.model} vectorIndex=${vectorIndex} expectedDimension=${this.dimension} actualDimension=${vector.length}`,
+                });
+            }
+            for (const [valueIndex, value] of vector.entries()) {
+                if (!Number.isFinite(value)) {
+                    throw new EngineError("Embedding model returned a non-finite vector value", {
+                        code: "ZVEC_GREP.ENGINE.MODELS.EMBEDDING_NON_FINITE_VECTOR_VALUE",
+                        context: `model=${this.ref.model} vectorIndex=${vectorIndex} valueIndex=${valueIndex}`,
+                    });
+                }
+            }
+        }
+    }
+}
+//# sourceMappingURL=embeddings.js.map

package/dist/engine/models/embeddings.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"embeddings.js","sourceRoot":"","sources":["../../../src/engine/models/embeddings.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,oBAAoB,CAAC;AAyBjD,MAAM,OAAgB,cAAc;IAMzB,2BAA2B,CAAU;IAE9C,KAAK,CAAC,KAAK,CACT,QAA4B,EAC5B,UAA4B,EAAE;QAE9B,IAAI,CAAC,gBAAgB,CAAC,QAAQ,CAAC,CAAC;QAChC,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,QAAQ,EAAE;YAC3C,OAAO,EAAE,OAAO,CAAC,OAAO,IAAI,UAAU;SACvC,CAAC,CAAC;QACH,IAAI,CAAC,eAAe,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;QAExC,OAAO,OAAO,CAAC;IACjB,CAAC;IAGD,KAAK,CAAC,OAAO;QACX,wDAAwD;IAC1D,CAAC;IASO,gBAAgB,CAAC,QAA4B;QACnD,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC1B,MAAM,IAAI,WAAW,CAAC,8CAA8C,EAAE;gBACpE,IAAI,EAAE,+CAA+C;aACtD,CAAC,CAAC;QACL,CAAC;QAED,IAAI,QAAQ,CAAC,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,YAAY,EAAE,CAAC;YAC/C,MAAM,IAAI,WAAW,CAAC,0CAA0C,EAAE;gBAChE,IAAI,EAAE,mDAAmD;gBACzD,OAAO,EAAE,SAAS,IAAI,CAAC,GAAG,CAAC,KAAK,cAAc,QAAQ,CAAC,MAAM,iBAAiB,IAAI,CAAC,MAAM,CAAC,YAAY,EAAE;aACzG,CAAC,CAAC;QACL,CAAC;QAED,KAAK,MAAM,CAAC,KAAK,EAAE,OAAO,CAAC,IAAI,QAAQ,CAAC,OAAO,EAAE,EAAE,CAAC;YAClD,IAAI,CAAC,IAAI,CAAC,qBAAqB,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,CAAC;gBACvD,MAAM,IAAI,WAAW,CAAC,+CAA+C,EAAE;oBACrE,IAAI,EAAE,uDAAuD;oBAC7D,OAAO,EAAE,SAAS,IAAI,CAAC,GAAG,CAAC,KAAK,UAAU,KAAK,SAAS,OAAO,CAAC,IAAI,EAAE;iBACvE,CAAC,CAAC;YACL,CAAC;YAED,IAAI,OAAO,CAAC,IAAI,KAAK,MAAM,IAAI,OAAO,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBAChE,MAAM,IAAI,WAAW,CAAC,0CAA0C,EAAE;oBAChE,IAAI,EAAE,8CAA8C;oBACpD,OAAO,EAAE,SAAS,IAAI,CAAC,GAAG,CAAC,KAAK,UAAU,KAAK,EAAE;iBAClD,CAAC,CAAC;YACL,CAAC;YAED,IAAI,OAAO,CAAC,IAAI,KAAK,OAAO,EAAE,CAAC;gBAC7B,IAAI,OAAO,CAAC,IAAI,CAAC,UAAU,KAAK,CAAC,EAAE,CAAC;oBAClC,MAAM,IAAI,WAAW,CAAC,2CAA2C,EAAE;wBACjE,IAAI,EAAE,+CAA+C;wBACrD,OAAO,EAAE,SAAS,IAAI,CAAC,GAAG,CAAC,KAAK,UAAU,KAAK,EAAE;qBAClD,CAAC,CAAC;gBACL,CAAC;gBAED,IAAI,OAAO,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;oBACvC,MAAM,IAAI,WAAW,CAAC,+CAA+C,EAAE;wBACrE,IAAI,EAAE,wDAAwD;wBAC9D,OAAO,EAAE,SAAS,IAAI,CAAC,GAAG,CAAC,KAAK,UAAU,KAAK,EAAE;qBAClD,CAAC,CAAC;gBACL,CAAC;gBAED,IACE,IAAI,CAAC,MAAM,CAAC,aAAa,KAAK,SAAS;uBACpC,OAAO,CAAC,IAAI,CAAC,UAAU,GAAG,IAAI,CAAC,MAAM,CAAC,aAAa,EACtD,CAAC;oBACD,MAAM,IAAI,WAAW,CAAC,6CAA6C,EAAE;wBACnE,IAAI,EAAE,mDAAmD;wBACzD,OAAO,EAAE,SAAS,IAAI,CAAC,GAAG,CAAC,KAAK,UAAU,KAAK,eAAe,OAAO,CAAC,IAAI,CAAC,UAAU,kBAAkB,IAAI,CAAC,MAAM,CAAC,aAAa,EAAE;qBACnI,CAAC,CAAC;gBACL,CAAC;YACH,CAAC;QACH,CAAC;IACH,CAAC;IAGO,eAAe,CAAC,QAA4B,EAAE,OAA0B;QAC9E,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;YAC5B,MAAM,IAAI,WAAW,CAAC,+CAA+C,EAAE;gBACrE,IAAI,EAAE,oDAAoD;gBAC1D,OAAO,EAAE,SAAS,IAAI,CAAC,GAAG,CAAC,KAAK,EAAE;aACnC,CAAC,CAAC;QACL,CAAC;QAED,IAAI,OAAO,CAAC,MAAM,KAAK,QAAQ,CAAC,MAAM,EAAE,CAAC;YACvC,MAAM,IAAI,WAAW,CAAC,sDAAsD,EAAE;gBAC5E,IAAI,EAAE,yDAAyD;gBAC/D,OAAO,EAAE,SAAS,IAAI,CAAC,GAAG,CAAC,KAAK,iBAAiB,QAAQ,CAAC,MAAM,gBAAgB,OAAO,CAAC,MAAM,EAAE;aACjG,CAAC,CAAC;QACL,CAAC;QAED,KAAK,MAAM,CAAC,WAAW,EAAE,MAAM,CAAC,IAAI,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC;YACtD,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC;gBAC3B,MAAM,IAAI,WAAW,CAAC,6CAA6C,EAAE;oBACnE,IAAI,EAAE,kDAAkD;oBACxD,OAAO,EAAE,SAAS,IAAI,CAAC,GAAG,CAAC,KAAK,gBAAgB,WAAW,EAAE;iBAC9D,CAAC,CAAC;YACL,CAAC;YAED,IAAI,MAAM,CAAC,MAAM,KAAK,IAAI,CAAC,SAAS,EAAE,CAAC;gBACrC,MAAM,IAAI,WAAW,CAAC,4DAA4D,EAAE;oBAClF,IAAI,EAAE,sDAAsD;oBAC5D,OAAO,EAAE,SAAS,IAAI,CAAC,GAAG,CAAC,KAAK,gBAAgB,WAAW,sBAAsB,IAAI,CAAC,SAAS,oBAAoB,MAAM,CAAC,MAAM,EAAE;iBACnI,CAAC,CAAC;YACL,CAAC;YAED,KAAK,MAAM,CAAC,UAAU,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,EAAE,EAAE,CAAC;gBACnD,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;oBAC5B,MAAM,IAAI,WAAW,CAAC,oDAAoD,EAAE;wBAC1E,IAAI,EAAE,2DAA2D;wBACjE,OAAO,EAAE,SAAS,IAAI,CAAC,GAAG,CAAC,KAAK,gBAAgB,WAAW,eAAe,UAAU,EAAE;qBACvF,CAAC,CAAC;gBACL,CAAC;YACH,CAAC;QACH,CAAC;IACH,CAAC;CACF"}

package/dist/engine/models/factory.d.ts

@@ -0,0 +1,5 @@
+import type { EmbeddingModel } from "./embeddings.js";
+import type { ModelProviderOptions, ModelRef } from "./types.js";
+export declare function createEmbeddingModel(ref: ModelRef, options: ModelProviderOptions): EmbeddingModel;
+export declare function createEmbeddingModelFromCatalog(id: string, options: ModelProviderOptions): EmbeddingModel;
+export declare function createEmbeddingModelFromReference(reference: string, options: ModelProviderOptions): EmbeddingModel;

package/dist/engine/models/factory.js

@@ -0,0 +1,66 @@
+import { EngineError } from "../errors/index.js";
+import { getEmbeddingModelCatalogEntry, getEmbeddingModelCatalogEntryByRef, } from "./catalog.js";
+import { LlamaCppEmbeddingModel, Qwen3VlEmbeddingModel, QwenTextEmbeddingV4Model, } from "./providers/index.js";
+export function createEmbeddingModel(ref, options) {
+    if (ref.provider === "local") {
+        const entry = getEmbeddingModelCatalogEntryByRef(ref);
+        if (!entry || entry.provider !== "local") {
+            throw new EngineError("Local embedding model is not in the zvec-grep catalog", {
+                code: "ZVEC_GREP.ENGINE.MODELS.LOCAL_MODEL_NOT_IN_CATALOG",
+                context: `provider=${ref.provider} model=${ref.model}`,
+            });
+        }
+        return new LlamaCppEmbeddingModel(entry, options);
+    }
+    if (ref.provider === "qwen" && ref.model === "text-embedding-v4") {
+        return new QwenTextEmbeddingV4Model(options);
+    }
+    if (ref.provider === "qwen" && ref.model === "qwen3-vl-embedding") {
+        return new Qwen3VlEmbeddingModel(options);
+    }
+    throw new EngineError("Embedding model is not implemented", {
+        code: "ZVEC_GREP.ENGINE.MODELS.EMBEDDING_MODEL_NOT_IMPLEMENTED",
+        context: `provider=${ref.provider} model=${ref.model}`,
+    });
+}
+export function createEmbeddingModelFromCatalog(id, options) {
+    const entry = getEmbeddingModelCatalogEntry(id);
+    if (!entry) {
+        throw new EngineError("Embedding model is not in the zvec-grep catalog", {
+            code: "ZVEC_GREP.ENGINE.MODELS.EMBEDDING_CATALOG_MODEL_NOT_FOUND",
+            context: `embedding=${id}`,
+        });
+    }
+    return createEmbeddingModel({
+        provider: entry.provider,
+        model: entry.model,
+    }, options);
+}
+export function createEmbeddingModelFromReference(reference, options) {
+    const catalogEntry = getEmbeddingModelCatalogEntry(reference);
+    if (catalogEntry) {
+        return createEmbeddingModel({
+            provider: catalogEntry.provider,
+            model: catalogEntry.model,
+        }, options);
+    }
+    const ref = parseModelReference(reference);
+    if (!ref) {
+        throw new EngineError("Embedding model is not in the zvec-grep catalog", {
+            code: "ZVEC_GREP.ENGINE.MODELS.EMBEDDING_CATALOG_MODEL_NOT_FOUND",
+            context: `embedding=${reference}`,
+        });
+    }
+    return createEmbeddingModel(ref, options);
+}
+function parseModelReference(reference) {
+    const separator = reference.indexOf("/");
+    if (separator <= 0 || separator === reference.length - 1) {
+        return undefined;
+    }
+    return {
+        provider: reference.slice(0, separator),
+        model: reference.slice(separator + 1),
+    };
+}
+//# sourceMappingURL=factory.js.map

package/dist/engine/models/factory.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"factory.js","sourceRoot":"","sources":["../../../src/engine/models/factory.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,oBAAoB,CAAC;AACjD,OAAO,EACL,6BAA6B,EAC7B,kCAAkC,GACnC,MAAM,cAAc,CAAC;AAEtB,OAAO,EACL,sBAAsB,EACtB,qBAAqB,EACrB,wBAAwB,GACzB,MAAM,sBAAsB,CAAC;AAI9B,MAAM,UAAU,oBAAoB,CAClC,GAAa,EACb,OAA6B;IAE7B,IAAI,GAAG,CAAC,QAAQ,KAAK,OAAO,EAAE,CAAC;QAC7B,MAAM,KAAK,GAAG,kCAAkC,CAAC,GAAG,CAAC,CAAC;QACtD,IAAI,CAAC,KAAK,IAAI,KAAK,CAAC,QAAQ,KAAK,OAAO,EAAE,CAAC;YACzC,MAAM,IAAI,WAAW,CAAC,uDAAuD,EAAE;gBAC7E,IAAI,EAAE,oDAAoD;gBAC1D,OAAO,EAAE,YAAY,GAAG,CAAC,QAAQ,UAAU,GAAG,CAAC,KAAK,EAAE;aACvD,CAAC,CAAC;QACL,CAAC;QAED,OAAO,IAAI,sBAAsB,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;IACpD,CAAC;IAED,IAAI,GAAG,CAAC,QAAQ,KAAK,MAAM,IAAI,GAAG,CAAC,KAAK,KAAK,mBAAmB,EAAE,CAAC;QACjE,OAAO,IAAI,wBAAwB,CAAC,OAAO,CAAC,CAAC;IAC/C,CAAC;IAED,IAAI,GAAG,CAAC,QAAQ,KAAK,MAAM,IAAI,GAAG,CAAC,KAAK,KAAK,oBAAoB,EAAE,CAAC;QAClE,OAAO,IAAI,qBAAqB,CAAC,OAAO,CAAC,CAAC;IAC5C,CAAC;IAED,MAAM,IAAI,WAAW,CAAC,oCAAoC,EAAE;QAC1D,IAAI,EAAE,yDAAyD;QAC/D,OAAO,EAAE,YAAY,GAAG,CAAC,QAAQ,UAAU,GAAG,CAAC,KAAK,EAAE;KACvD,CAAC,CAAC;AACL,CAAC;AAGD,MAAM,UAAU,+BAA+B,CAC7C,EAAU,EACV,OAA6B;IAE7B,MAAM,KAAK,GAAG,6BAA6B,CAAC,EAAE,CAAC,CAAC;IAChD,IAAI,CAAC,KAAK,EAAE,CAAC;QACX,MAAM,IAAI,WAAW,CAAC,iDAAiD,EAAE;YACvE,IAAI,EAAE,2DAA2D;YACjE,OAAO,EAAE,aAAa,EAAE,EAAE;SAC3B,CAAC,CAAC;IACL,CAAC;IAED,OAAO,oBAAoB,CAAC;QAC1B,QAAQ,EAAE,KAAK,CAAC,QAAQ;QACxB,KAAK,EAAE,KAAK,CAAC,KAAK;KACnB,EAAE,OAAO,CAAC,CAAC;AACd,CAAC;AAGD,MAAM,UAAU,iCAAiC,CAC/C,SAAiB,EACjB,OAA6B;IAE7B,MAAM,YAAY,GAAG,6BAA6B,CAAC,SAAS,CAAC,CAAC;IAC9D,IAAI,YAAY,EAAE,CAAC;QACjB,OAAO,oBAAoB,CAAC;YAC1B,QAAQ,EAAE,YAAY,CAAC,QAAQ;YAC/B,KAAK,EAAE,YAAY,CAAC,KAAK;SAC1B,EAAE,OAAO,CAAC,CAAC;IACd,CAAC;IAED,MAAM,GAAG,GAAG,mBAAmB,CAAC,SAAS,CAAC,CAAC;IAC3C,IAAI,CAAC,GAAG,EAAE,CAAC;QACT,MAAM,IAAI,WAAW,CAAC,iDAAiD,EAAE;YACvE,IAAI,EAAE,2DAA2D;YACjE,OAAO,EAAE,aAAa,SAAS,EAAE;SAClC,CAAC,CAAC;IACL,CAAC;IAED,OAAO,oBAAoB,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC;AAC5C,CAAC;AAGD,SAAS,mBAAmB,CAAC,SAAiB;IAC5C,MAAM,SAAS,GAAG,SAAS,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;IACzC,IAAI,SAAS,IAAI,CAAC,IAAI,SAAS,KAAK,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACzD,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,OAAO;QACL,QAAQ,EAAE,SAAS,CAAC,KAAK,CAAC,CAAC,EAAE,SAAS,CAAC;QACvC,KAAK,EAAE,SAAS,CAAC,KAAK,CAAC,SAAS,GAAG,CAAC,CAAC;KACtC,CAAC;AACJ,CAAC"}

package/dist/engine/models/index.d.ts

@@ -0,0 +1,6 @@
+export { EMBEDDING_MODEL_CATALOG, MODEL_CATALOG, getEmbeddingModelCatalogEntry, getEmbeddingModelCatalogEntryByRef, listEmbeddingModels, } from "./catalog.js";
+export { EmbeddingModel } from "./embeddings.js";
+export { LlamaCppEmbeddingModel, Qwen3VlEmbeddingModel, QwenTextEmbeddingV4Model, setLlamaCppRuntimeForTesting, } from "./providers/index.js";
+export type { RankingCandidate, RankingModel, RankingScore } from "./ranking.js";
+export { createEmbeddingModel, createEmbeddingModelFromCatalog, createEmbeddingModelFromReference, } from "./factory.js";
+export type { EmbeddingCatalogEntry, LlamaCppEmbeddingCatalogEntry, LlamaGpuMode, LocalEmbeddingFormat, ModelProviderOptions, } from "./types.js";

package/dist/engine/models/index.js

@@ -0,0 +1,5 @@
+export { EMBEDDING_MODEL_CATALOG, MODEL_CATALOG, getEmbeddingModelCatalogEntry, getEmbeddingModelCatalogEntryByRef, listEmbeddingModels, } from "./catalog.js";
+export { EmbeddingModel } from "./embeddings.js";
+export { LlamaCppEmbeddingModel, Qwen3VlEmbeddingModel, QwenTextEmbeddingV4Model, setLlamaCppRuntimeForTesting, } from "./providers/index.js";
+export { createEmbeddingModel, createEmbeddingModelFromCatalog, createEmbeddingModelFromReference, } from "./factory.js";
+//# sourceMappingURL=index.js.map

package/dist/engine/models/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/engine/models/index.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,uBAAuB,EACvB,aAAa,EACb,6BAA6B,EAC7B,kCAAkC,EAClC,mBAAmB,GACpB,MAAM,cAAc,CAAC;AAGtB,OAAO,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAC;AAGjD,OAAO,EACL,sBAAsB,EACtB,qBAAqB,EACrB,wBAAwB,EACxB,4BAA4B,GAC7B,MAAM,sBAAsB,CAAC;AAU9B,OAAO,EACL,oBAAoB,EACpB,+BAA+B,EAC/B,iCAAiC,GAClC,MAAM,cAAc,CAAC"}

package/dist/engine/models/providers/index.d.ts

@@ -0,0 +1,2 @@
+export { Qwen3VlEmbeddingModel, QwenTextEmbeddingV4Model } from "./qwen/index.js";
+export { LlamaCppEmbeddingModel, setLlamaCppRuntimeForTesting, } from "./llama-cpp/index.js";

package/dist/engine/models/providers/index.js

@@ -0,0 +1,3 @@
+export { Qwen3VlEmbeddingModel, QwenTextEmbeddingV4Model } from "./qwen/index.js";
+export { LlamaCppEmbeddingModel, setLlamaCppRuntimeForTesting, } from "./llama-cpp/index.js";
+//# sourceMappingURL=index.js.map

package/dist/engine/models/providers/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../../src/engine/models/providers/index.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,qBAAqB,EACrB,wBAAwB,EACzB,MAAM,iBAAiB,CAAC;AAEzB,OAAO,EACL,sBAAsB,EACtB,4BAA4B,GAC7B,MAAM,sBAAsB,CAAC"}

package/dist/engine/models/providers/llama-cpp/embedding.d.ts

@@ -0,0 +1,94 @@
+import type { Content } from "../../../types.js";
+import { EmbeddingModel, type EmbeddingOptions, type EmbeddingVector } from "../../embeddings.js";
+import type { LlamaCppEmbeddingCatalogEntry, ModelProviderOptions } from "../../types.js";
+type LlamaEmbedding = {
+    vector: ArrayLike<number>;
+};
+type LlamaEmbeddingContext = {
+    getEmbeddingFor(text: string): Promise<LlamaEmbedding>;
+    dispose?(): Promise<void> | void;
+};
+type LlamaModel = {
+    trainContextSize?: number;
+    tokenize?(text: string): readonly unknown[];
+    detokenize?(tokens: readonly unknown[]): string;
+    createEmbeddingContext(options: Record<string, unknown>): Promise<LlamaEmbeddingContext>;
+    dispose?(): Promise<void> | void;
+};
+type Llama = {
+    gpu: string | false;
+    cpuMathCores?: number;
+    supportsGpuOffloading?: boolean;
+    getVramState?(): Promise<{
+        total: number;
+        used: number;
+        free: number;
+    }>;
+    loadModel(options: {
+        modelPath: string;
+        gpuLayers?: number;
+    }): Promise<LlamaModel>;
+    dispose?(): Promise<void> | void;
+};
+type NodeLlamaCppModule = {
+    getLlama(options: Record<string, unknown>): Promise<Llama>;
+    resolveModelFile(model: string, options: {
+        directory: string;
+        cli?: boolean;
+    }): Promise<string>;
+    LlamaLogLevel?: {
+        error?: unknown;
+    };
+};
+type NodeLlamaCppLoader = () => Promise<NodeLlamaCppModule>;
+export declare function setLlamaCppRuntimeForTesting(loader: NodeLlamaCppLoader | null): void;
+export declare class LlamaCppEmbeddingModel extends EmbeddingModel {
+    private readonly entry;
+    readonly ref: {
+        readonly provider: "local";
+        readonly model: string;
+    };
+    readonly dimension: number;
+    readonly metric: import("../../types.js").VectorMetric;
+    readonly supportedContentKinds: readonly ["text"];
+    readonly limits: {
+        readonly maxBatchSize: number;
+        readonly maxInputTokens: number;
+    };
+    readonly recommendedIndexConcurrency = 1;
+    private readonly modelCacheDir;
+    private readonly gpu;
+    private readonly embeddingParallelism?;
+    private llama;
+    private model;
+    private contexts;
+    private llamaLoadPromise;
+    private modelLoadPromise;
+    private contextsCreatePromise;
+    private usingCpuFallback;
+    private disposed;
+    constructor(entry: LlamaCppEmbeddingCatalogEntry, options: ModelProviderOptions);
+    protected doEmbed(contents: readonly Content[], options: Required<EmbeddingOptions>): Promise<EmbeddingVector[]>;
+    dispose(): Promise<void>;
+    private embedTexts;
+    private ensureLlama;
+    private loadLlamaWithFallback;
+    private loadCpuCompatibleLlama;
+    private ensureModel;
+    private loadModelWithFallback;
+    private loadModel;
+    private modelLoadOptions;
+    private ensureEmbeddingContexts;
+    private createEmbeddingContextsWithFallback;
+    private createEmbeddingContexts;
+    private canRetryGpuOperationOnCpu;
+    private shouldDisableModelGpuOffload;
+    private disposeLoadedRuntime;
+    private resolveModelPath;
+    private resolveParallelism;
+    private resolveEffectiveParallelism;
+    private resolveThreadsPerContext;
+    private truncateToContextSize;
+    private ensureNotDisposed;
+}
+export {};