@rudderjs/ai 1.17.2 → 1.18.0

package/dist/providers/google.js

@@ -1,903 +0,0 @@
-import { base64ToUtf8 } from '../base64.js';
-import { buildGoogleCacheKey, splitContentsAtCache, durationToGoogleTtl, _internals as _registryInternals, } from './google-cache-registry.js';
-export class GoogleProvider {
-    name = 'google';
-    config;
-    cacheRegistry;
-    constructor(config, cacheRegistry) {
-        this.config = config;
-        if (cacheRegistry)
-            this.cacheRegistry = cacheRegistry;
-    }
-    create(model) {
-        return new GoogleAdapter(this.config, model, this.cacheRegistry);
-    }
-    createEmbedding(model) {
-        return new GoogleEmbeddingAdapter(this.config, model);
-    }
-    createImage(model) {
-        return new GoogleImageAdapter(this.config, model);
-    }
-    createFiles() {
-        return new GoogleFileAdapter(this.config);
-    }
-    createVectorStores() {
-        return new GoogleVectorStoreAdapter(this.config);
-    }
-}
-// ─── Adapter ──────────────────────────────────────────────
-export class GoogleAdapter {
-    config;
-    model;
-    cacheRegistry;
-    client = null;
-    constructor(config, model, cacheRegistry) {
-        this.config = config;
-        this.model = model;
-        this.cacheRegistry = cacheRegistry;
-    }
-    async getClient() {
-        if (this.client)
-            return this.client;
-        const sdk = await import(/* @vite-ignore */ '@google/genai');
-        const GoogleGenAI = sdk.GoogleGenAI ?? sdk.default;
-        this.client = new GoogleGenAI({ apiKey: this.config.apiKey });
-        return this.client;
-    }
-    /**
-     * Build the request payload, consulting the cache registry if `options.cache`
-     * is set. Returns the payload for `generateContent` / `generateContentStream`
-     * plus the cache key (so the caller can `forget()` it on a 404 stale-cache
-     * retry).
-     */
-    async buildPayload(options) {
-        const client = await this.getClient();
-        const { system, contents } = toGeminiContents(options.messages);
-        // `toGeminiTools` returns the already-wrapped top-level array
-        // ({functionDeclarations: [...]} + any native blocks like google_search).
-        const geminiTools = options.tools?.length ? toGeminiTools(options.tools) : undefined;
-        const config = {};
-        if (options.maxTokens)
-            config['maxOutputTokens'] = options.maxTokens;
-        if (options.temperature !== undefined)
-            config['temperature'] = options.temperature;
-        if (options.topP !== undefined)
-            config['topP'] = options.topP;
-        if (options.stop)
-            config['stopSequences'] = options.stop;
-        if (geminiTools && geminiTools.length > 0)
-            config['tools'] = geminiTools;
-        if (options.toolChoice)
-            config['toolConfig'] = toGeminiToolConfig(options.toolChoice);
-        // The Gemini SDK reads abortSignal from the config block.
-        if (options.signal)
-            config['abortSignal'] = options.signal;
-        let cacheName = null;
-        let cacheKey;
-        if (this.cacheRegistry && options.cache) {
-            cacheKey = buildGoogleCacheKey(this.model, options.cache, system, contents, geminiTools);
-            if (cacheKey) {
-                const { cached: cachedSlice } = splitContentsAtCache(contents, options.cache);
-                cacheName = await this.cacheRegistry.resolve({
-                    client,
-                    model: this.model,
-                    cacheKey,
-                    ...(options.cache.instructions && system ? { systemInstruction: { parts: [{ text: system }] } } : {}),
-                    ...(cachedSlice.length > 0 ? { contents: cachedSlice } : {}),
-                    ...(options.cache.tools && geminiTools && geminiTools.length > 0 ? { tools: geminiTools } : {}),
-                    ...(options.cache.ttl ? { ttl: durationToGoogleTtl(options.cache.ttl) } : {}),
-                });
-            }
-        }
-        if (cacheName) {
-            // Drop tools / system from the request — they're inherited from the cache resource.
-            const { fresh } = splitContentsAtCache(contents, options.cache);
-            const configForCachedRequest = { ...config };
-            delete configForCachedRequest['tools'];
-            configForCachedRequest['cachedContent'] = cacheName;
-            return {
-                payload: { model: this.model, contents: fresh, config: configForCachedRequest },
-                cacheKey,
-            };
-        }
-        return {
-            payload: {
-                model: this.model,
-                contents,
-                ...(system ? { systemInstruction: { parts: [{ text: system }] } } : {}),
-                config,
-            },
-            cacheKey,
-        };
-    }
-    async generate(options) {
-        const { payload, cacheKey } = await this.buildPayload(options);
-        const client = await this.getClient();
-        try {
-            const response = await client.models.generateContent(payload);
-            return fromGeminiResponse(response);
-        }
-        catch (err) {
-            if (cacheKey && this.cacheRegistry && _registryInternals.isNotFoundError(err)) {
-                // Stale `cachedContent` resource — drop and retry once with a fresh build.
-                await this.cacheRegistry.forget(cacheKey);
-                const { payload: retryPayload } = await this.buildPayload(options);
-                const response = await client.models.generateContent(retryPayload);
-                return fromGeminiResponse(response);
-            }
-            throw err;
-        }
-    }
-    async *stream(options) {
-        let payloadAndKey = await this.buildPayload(options);
-        const client = await this.getClient();
-        let response;
-        try {
-            response = await client.models.generateContentStream(payloadAndKey.payload);
-        }
-        catch (err) {
-            if (payloadAndKey.cacheKey && this.cacheRegistry && _registryInternals.isNotFoundError(err)) {
-                await this.cacheRegistry.forget(payloadAndKey.cacheKey);
-                payloadAndKey = await this.buildPayload(options);
-                response = await client.models.generateContentStream(payloadAndKey.payload);
-            }
-            else {
-                throw err;
-            }
-        }
-        for await (const chunk of response) {
-            const candidate = chunk.candidates?.[0];
-            if (!candidate)
-                continue;
-            for (const part of candidate.content?.parts ?? []) {
-                if (part.text) {
-                    yield { type: 'text-delta', text: part.text };
-                }
-                if (part.functionCall) {
-                    yield {
-                        type: 'tool-call',
-                        toolCall: {
-                            id: `call_${Date.now()}_${Math.random().toString(36).slice(2, 8)}`,
-                            name: part.functionCall.name,
-                            arguments: part.functionCall.args ?? {},
-                        },
-                    };
-                }
-            }
-            if (candidate.finishReason) {
-                yield {
-                    type: 'finish',
-                    finishReason: candidate.finishReason === 'STOP' ? 'stop' : 'tool_calls',
-                    usage: chunk.usageMetadata ? {
-                        promptTokens: chunk.usageMetadata.promptTokenCount ?? 0,
-                        completionTokens: chunk.usageMetadata.candidatesTokenCount ?? 0,
-                        totalTokens: chunk.usageMetadata.totalTokenCount ?? 0,
-                    } : undefined,
-                };
-            }
-        }
-    }
-}
-// ─── Conversion Helpers ──────────────────────────────────
-function contentToString(content) {
-    if (typeof content === 'string')
-        return content;
-    return content.filter(p => p.type === 'text').map(p => p.text).join('');
-}
-function contentToGeminiParts(content) {
-    if (typeof content === 'string')
-        return [{ text: content }];
-    return content.map(p => {
-        if (p.type === 'text')
-            return { text: p.text };
-        if (p.type === 'image')
-            return { inlineData: { mimeType: p.mimeType, data: p.data } };
-        // document — inline data for PDFs, text for text-based
-        if (p.mimeType === 'application/pdf') {
-            return { inlineData: { mimeType: p.mimeType, data: p.data } };
-        }
-        return { text: base64ToUtf8(p.data) };
-    });
-}
-export function toGeminiContents(messages) {
-    const systemMsgs = messages.filter(m => m.role === 'system');
-    const rest = messages.filter(m => m.role !== 'system');
-    const system = systemMsgs.length > 0
-        ? systemMsgs.map(m => contentToString(m.content)).join('\n\n')
-        : undefined;
-    // Gemini's `functionResponse.name` must match the originating `functionCall.name`
-    // (the function name like "search"), not the synthetic call id the adapter
-    // generates per stream. Pre-build a (toolCallId → name) lookup by walking
-    // every prior assistant message's `toolCalls`. Without this the receiving
-    // model sees `name: "call_1234_abc"` and can't pair the result with the call.
-    const toolNameByCallId = new Map();
-    for (const m of rest) {
-        if (m.role === 'assistant' && m.toolCalls?.length) {
-            for (const tc of m.toolCalls)
-                toolNameByCallId.set(tc.id, tc.name);
-        }
-    }
-    const contents = rest.map(m => {
-        if (m.role === 'assistant' && m.toolCalls?.length) {
-            const text = contentToString(m.content);
-            return {
-                role: 'model',
-                parts: [
-                    ...(text ? [{ text }] : []),
-                    ...m.toolCalls.map(tc => ({
-                        functionCall: { name: tc.name, args: tc.arguments },
-                    })),
-                ],
-            };
-        }
-        if (m.role === 'tool') {
-            const callId = m.toolCallId;
-            const name = (callId && toolNameByCallId.get(callId)) ?? 'unknown';
-            return {
-                role: 'user',
-                parts: [{
-                        functionResponse: {
-                            name,
-                            response: typeof m.content === 'string' ? { result: m.content } : m.content,
-                        },
-                    }],
-            };
-        }
-        return {
-            role: m.role === 'assistant' ? 'model' : 'user',
-            parts: contentToGeminiParts(m.content),
-        };
-    });
-    return { system, contents };
-}
-/**
- * Build Gemini's `tools` array. The Gemini API accepts a mixed array where
- * function declarations live under one wrapper entry and provider-native
- * blocks (e.g. `{ google_search: {} }`) sit as separate top-level entries:
- *
- *   tools: [
- *     { functionDeclarations: [...] },
- *     { google_search: {} },
- *   ]
- *
- * Tools tagged with a recognized `providerHint.type` are emitted as their
- * native top-level block; everything else collects into one
- * `functionDeclarations` entry. Tools with unrecognized hints fall through
- * to the function-declaration shape — the input schema's still there, so
- * the worst case is the model treats it as a regular function-call tool.
- */
-function toGeminiTools(tools) {
-    const fnDecls = [];
-    const blocks = [];
-    for (const t of tools) {
-        if (t.providerHint?.type === 'web-search') {
-            // Gemini's native search tool. The block's `google_search: {}` form is
-            // intentional — Gemini doesn't accept allowed_domains / max_uses on
-            // this block, so the WebSearch.domains() / .maxResults() opts are
-            // ignored on this provider (documented on WebSearch).
-            blocks.push({ google_search: {} });
-            continue;
-        }
-        if (t.providerHint?.type === 'file-search') {
-            // Gemini's native FileSearch tool (#B8.5). The OpenAI-shaped hint
-            // (`vector_store_ids` + typed `filters`) is translated to Gemini's
-            // shape (`fileSearchStoreNames` + `metadataFilter` string). `topK`
-            // mirrors OpenAI's `max_num_results`.
-            const storeNames = t.providerHint['vector_store_ids'] ?? [];
-            const fileSearchConfig = {
-                fileSearchStoreNames: storeNames,
-            };
-            const filters = t.providerHint['filters'];
-            if (filters !== undefined) {
-                fileSearchConfig['metadataFilter'] = filterToGeminiString(filters);
-            }
-            const maxNumResults = t.providerHint['max_num_results'];
-            if (maxNumResults !== undefined) {
-                fileSearchConfig['topK'] = maxNumResults;
-            }
-            blocks.push({ fileSearch: fileSearchConfig });
-            continue;
-        }
-        fnDecls.push({
-            name: t.name,
-            description: t.description,
-            parameters: t.parameters,
-        });
-    }
-    if (fnDecls.length > 0)
-        blocks.unshift({ functionDeclarations: fnDecls });
-    return blocks;
-}
-/**
- * Translate a typed `FileSearchFilter` (OpenAI-shaped) into Gemini's
- * `metadataFilter` string syntax (#B8.5).
- *
- * - `{ type: 'eq',  key, value }` → `key = value`
- * - `{ type: 'ne',  key, value }` → `key != value`
- * - `{ type: 'gt',  key, value }` → `key > value`
- * - `{ type: 'gte', key, value }` → `key >= value`
- * - `{ type: 'lt',  key, value }` → `key < value`
- * - `{ type: 'lte', key, value }` → `key <= value`
- * - `{ type: 'and', filters }`    → `(f1) AND (f2) AND ...`
- * - `{ type: 'or',  filters }`    → `(f1) OR (f2) OR ...`
- *
- * String values are wrapped in double quotes with `"` and `\` escaped.
- * Numbers and booleans render bare.
- *
- * Exported for unit testing — see `google-vector-stores.test.ts`.
- *
- * @internal
- */
-export function filterToGeminiString(filter) {
-    switch (filter.type) {
-        case 'eq':
-        case 'ne':
-        case 'gt':
-        case 'gte':
-        case 'lt':
-        case 'lte': {
-            const op = GEMINI_FILTER_OP[filter.type];
-            return `${filter.key} ${op} ${formatGeminiValue(filter.value)}`;
-        }
-        case 'and':
-        case 'or': {
-            if (filter.filters.length === 0) {
-                throw new Error(`[Rudder AI] Gemini metadataFilter: ${filter.type.toUpperCase()} requires at least one sub-filter.`);
-            }
-            const joiner = filter.type === 'and' ? ' AND ' : ' OR ';
-            return filter.filters.map(f => `(${filterToGeminiString(f)})`).join(joiner);
-        }
-    }
-}
-const GEMINI_FILTER_OP = {
-    eq: '=',
-    ne: '!=',
-    gt: '>',
-    gte: '>=',
-    lt: '<',
-    lte: '<=',
-};
-function formatGeminiValue(value) {
-    if (typeof value === 'string') {
-        return `"${value.replace(/\\/g, '\\\\').replace(/"/g, '\\"')}"`;
-    }
-    return String(value);
-}
-function toGeminiToolConfig(choice) {
-    if (choice === 'auto')
-        return { functionCallingConfig: { mode: 'AUTO' } };
-    if (choice === 'required')
-        return { functionCallingConfig: { mode: 'ANY' } };
-    if (choice === 'none')
-        return { functionCallingConfig: { mode: 'NONE' } };
-    if (typeof choice === 'object' && 'name' in choice) {
-        return { functionCallingConfig: { mode: 'ANY', allowedFunctionNames: [choice.name] } };
-    }
-    return { functionCallingConfig: { mode: 'AUTO' } };
-}
-function fromGeminiResponse(response) {
-    const candidate = response.candidates?.[0];
-    const toolCalls = [];
-    let text = '';
-    for (const part of candidate?.content?.parts ?? []) {
-        if (part.text)
-            text += part.text;
-        if (part.functionCall) {
-            toolCalls.push({
-                id: `call_${Date.now()}_${Math.random().toString(36).slice(2, 8)}`,
-                name: part.functionCall.name,
-                arguments: part.functionCall.args ?? {},
-            });
-        }
-    }
-    return {
-        message: {
-            role: 'assistant',
-            content: text,
-            ...(toolCalls.length > 0 ? { toolCalls } : {}),
-        },
-        usage: {
-            promptTokens: response.usageMetadata?.promptTokenCount ?? 0,
-            completionTokens: response.usageMetadata?.candidatesTokenCount ?? 0,
-            totalTokens: response.usageMetadata?.totalTokenCount ?? 0,
-        },
-        finishReason: toolCalls.length > 0 ? 'tool_calls' : 'stop',
-    };
-}
-// ─── Embedding Adapter ──────────────────────────────────
-class GoogleEmbeddingAdapter {
-    config;
-    model;
-    client = null;
-    constructor(config, model) {
-        this.config = config;
-        this.model = model;
-    }
-    async getClient() {
-        if (this.client)
-            return this.client;
-        const sdk = await import(/* @vite-ignore */ '@google/genai');
-        const GoogleGenAI = sdk.GoogleGenAI ?? sdk.default;
-        this.client = new GoogleGenAI({ apiKey: this.config.apiKey });
-        return this.client;
-    }
-    async embed(input) {
-        const client = await this.getClient();
-        const inputs = Array.isArray(input) ? input : [input];
-        const results = await Promise.all(inputs.map(text => client.models.embedContent({
-            model: this.model,
-            content: { parts: [{ text }] },
-        })));
-        const embeddings = results.map((r) => r.embedding?.values ?? []);
-        return {
-            embeddings,
-            usage: { promptTokens: 0, totalTokens: 0 },
-        };
-    }
-}
-// ─── Image Generation Adapter (Imagen) ──────────────────
-const GOOGLE_IMAGE_SIZE_MAP = {
-    square: '1024x1024',
-    landscape: '1792x1024',
-    portrait: '1024x1792',
-};
-class GoogleImageAdapter {
-    config;
-    model;
-    constructor(config, model) {
-        this.config = config;
-        this.model = model;
-    }
-    async generate(options) {
-        const size = options.size
-            ? (GOOGLE_IMAGE_SIZE_MAP[options.size] ?? options.size)
-            : '1024x1024';
-        const [width, height] = size.split('x').map(Number);
-        const body = {
-            instances: [{ prompt: options.prompt }],
-            parameters: {
-                sampleCount: options.n ?? 1,
-                ...(width && height ? { aspectRatio: `${width}:${height}` } : {}),
-            },
-        };
-        const endpoint = `https://generativelanguage.googleapis.com/v1beta/models/${this.model}:predict?key=${this.config.apiKey}`;
-        const res = await fetch(endpoint, {
-            method: 'POST',
-            headers: { 'Content-Type': 'application/json' },
-            body: JSON.stringify(body),
-        });
-        if (!res.ok) {
-            throw new Error(`[Rudder AI] Google image generation error: ${res.status} ${await res.text()}`);
-        }
-        const data = await res.json();
-        return {
-            images: (data.predictions ?? []).map((p) => ({
-                ...(p.bytesBase64Encoded ? { base64: p.bytesBase64Encoded } : {}),
-            })),
-            model: this.model,
-        };
-    }
-}
-// ─── Files ──────────────────────────────────────────────
-class GoogleFileAdapter {
-    config;
-    client = null;
-    constructor(config) {
-        this.config = config;
-    }
-    async getClient() {
-        if (this.client)
-            return this.client;
-        const sdk = await import(/* @vite-ignore */ '@google/genai');
-        const GoogleGenAI = sdk.GoogleGenAI ?? sdk.default;
-        this.client = new GoogleGenAI({ apiKey: this.config.apiKey });
-        return this.client;
-    }
-    async upload(options) {
-        const client = await this.getClient();
-        const { readFile, stat } = await import(/* @vite-ignore */ 'node:fs/promises');
-        const { basename } = await import(/* @vite-ignore */ 'node:path');
-        const data = await readFile(options.filePath);
-        const stats = await stat(options.filePath);
-        const filename = basename(options.filePath);
-        const response = await client.files.upload({
-            file: new Blob([data]),
-            config: { displayName: filename },
-        });
-        return {
-            id: response.name ?? response.uri,
-            filename,
-            bytes: stats.size,
-        };
-    }
-    async list() {
-        const client = await this.getClient();
-        const response = await client.files.list();
-        const files = [];
-        for (const f of response.files ?? response ?? []) {
-            files.push({
-                id: f.name ?? f.uri,
-                filename: f.displayName ?? f.name ?? '',
-                bytes: Number(f.sizeBytes ?? 0),
-            });
-        }
-        return { files };
-    }
-    async delete(fileId) {
-        const client = await this.getClient();
-        await client.files.delete({ name: fileId });
-    }
-}
-// ─── Vector Stores (Gemini FileSearchStores, #B8.5) ──────
-//
-// Gemini's hosted RAG surface is `ai.fileSearchStores.*` — a direct
-// OpenAI-equivalent that handles ingestion, chunking, embedding, and
-// retrieval server-side. NOT available on Vertex AI; the underlying SDK
-// methods throw for Vertex clients.
-//
-// Mapping decisions:
-// - `VectorStoreInfo.id` is the full Gemini resource name
-//   (`fileSearchStores/foo-bar`). Apps pass it back verbatim to `get` /
-//   `delete` / `addFile`.
-// - `VectorStoreInfo.name` is `displayName`. The OpenAI adapter populates
-//   it from the store's name field; we use the user-supplied display name
-//   to keep `create('Knowledge Base')` round-trip-able.
-// - `createdAt` is parsed from ISO 8601 to Unix seconds for parity with
-//   OpenAI's `created_at`.
-// - `fileCount` sums `activeDocumentsCount + pendingDocumentsCount` (both
-//   string-encoded). `failedDocumentsCount` is dropped — it's surfaced
-//   per-file via `addFile`'s status when polling.
-// - `bytesUsed` is parsed from `sizeBytes` (string-encoded).
-// - Store-level `metadata` and `expiresAfter` are NOT supported by Gemini.
-//   Passing them throws fail-loud so apps don't silently lose data.
-//
-// `addFile` paths:
-// - `{ fileId }` → `importFile` (re-uses an existing Files API file).
-// - `{ filePath | fileBuffer }` → `uploadToFileSearchStore` (single-shot
-//   upload). Both paths return long-running operations; default
-//   `wait: true` polls `client.operations.get` until `done`.
-// - `attributes` (Record<string, primitive>) → Gemini's `customMetadata`
-//   array shape; booleans coerce to `stringValue: 'true' | 'false'`.
-class GoogleVectorStoreAdapter {
-    config;
-    client = null;
-    constructor(config) {
-        this.config = config;
-    }
-    async getClient() {
-        if (this.client)
-            return this.client;
-        const sdk = await import(/* @vite-ignore */ '@google/genai');
-        const GoogleGenAI = sdk.GoogleGenAI ?? sdk.default;
-        this.client = new GoogleGenAI({ apiKey: this.config.apiKey });
-        return this.client;
-    }
-    async create(opts) {
-        if (opts.metadata) {
-            throw new Error('[Rudder AI] Gemini FileSearchStores does not support store-level metadata. ' +
-                'Attach searchable metadata per-document via addFile({ attributes }).');
-        }
-        if (opts.expiresAfter) {
-            throw new Error('[Rudder AI] Gemini FileSearchStores does not support expiresAfter. ' +
-                'Stores persist until explicitly deleted via VectorStores.delete().');
-        }
-        const client = await this.getClient();
-        const response = await client.fileSearchStores.create({
-            config: { displayName: opts.name },
-        });
-        return fromGeminiFileSearchStore(response, opts.name);
-    }
-    async list(opts) {
-        const client = await this.getClient();
-        const config = {};
-        if (opts?.limit !== undefined)
-            config['pageSize'] = opts.limit;
-        if (opts?.after !== undefined)
-            config['pageToken'] = opts.after;
-        // Gemini paginates forward via pageToken only — `before` has no
-        // equivalent. Drop it silently (matches OpenAI when `before` is unset).
-        const pager = await client.fileSearchStores.list({ config });
-        const items = Array.isArray(pager?.page) ? pager.page : [];
-        return { stores: items.map(item => fromGeminiFileSearchStore(item)) };
-    }
-    async get(id) {
-        const client = await this.getClient();
-        const response = await client.fileSearchStores.get({ name: id });
-        return fromGeminiFileSearchStore(response);
-    }
-    async delete(id) {
-        const client = await this.getClient();
-        // `force: true` mirrors OpenAI's behavior — deleting a store also
-        // drops attached documents. Without `force`, Gemini returns
-        // FAILED_PRECONDITION when the store has any documents.
-        await client.fileSearchStores.delete({ name: id, config: { force: true } });
-    }
-    async addFile(storeId, opts) {
-        const client = await this.getClient();
-        const customMetadata = opts.attributes ? attributesToCustomMetadata(opts.attributes) : undefined;
-        // Path 1: re-use an existing Files API file.
-        if (opts.fileId) {
-            const importConfig = {};
-            if (customMetadata)
-                importConfig['customMetadata'] = customMetadata;
-            if (opts.chunkingStrategy)
-                importConfig['chunkingConfig'] = opts.chunkingStrategy;
-            const op = await client.fileSearchStores.importFile({
-                fileSearchStoreName: storeId,
-                fileName: opts.fileId,
-                config: importConfig,
-            });
-            return finishVectorStoreOperation(client, op, storeId, opts);
-        }
-        // Path 2: upload a local file directly. Either `filePath` or
-        // `fileBuffer` is required — Gemini's SDK accepts a path string OR a
-        // Blob. For `filePath`, the SDK infers mimeType from the extension;
-        // for `fileBuffer`, it reads `blob.type` which is empty on a
-        // untyped `new Blob([data])`, so we forward an explicit `mimeType`
-        // derived from `filename` to avoid `Can not determine mimeType`.
-        if (opts.filePath || opts.fileBuffer) {
-            const uploadConfig = {};
-            if (customMetadata)
-                uploadConfig['customMetadata'] = customMetadata;
-            if (opts.chunkingStrategy)
-                uploadConfig['chunkingConfig'] = opts.chunkingStrategy;
-            if (opts.fileBuffer?.filename)
-                uploadConfig['displayName'] = opts.fileBuffer.filename;
-            if (opts.fileBuffer?.filename) {
-                const mimeType = mimeTypeFromFilename(opts.fileBuffer.filename);
-                if (mimeType)
-                    uploadConfig['mimeType'] = mimeType;
-            }
-            const file = opts.filePath ?? new Blob([opts.fileBuffer.data]);
-            const op = await client.fileSearchStores.uploadToFileSearchStore({
-                fileSearchStoreName: storeId,
-                file,
-                config: uploadConfig,
-            });
-            return finishVectorStoreOperation(client, op, storeId, opts);
-        }
-        throw new Error('[Rudder AI] addFile requires fileId, filePath, or fileBuffer. ' +
-            'Pass an existing Gemini Files API id via { fileId } (e.g. `files/abc-123`) or ' +
-            'a local source via { filePath } / { fileBuffer }.');
-    }
-    async removeFile(storeId, fileId) {
-        const client = await this.getClient();
-        // Document resource names are `fileSearchStores/<store>/documents/<doc>`.
-        // Apps that pass the full path use it verbatim; apps that pass only
-        // the document id get the store prefix joined for them.
-        const name = fileId.includes('/documents/') ? fileId : `${storeId}/documents/${fileId}`;
-        await client.fileSearchStores.documents.delete({ name });
-    }
-    async listFiles(storeId, opts) {
-        const client = await this.getClient();
-        const config = {};
-        if (opts?.limit !== undefined)
-            config['pageSize'] = opts.limit;
-        if (opts?.after !== undefined)
-            config['pageToken'] = opts.after;
-        const pager = await client.fileSearchStores.documents.list({ parent: storeId, config });
-        const items = Array.isArray(pager?.page) ? pager.page : [];
-        return { files: items.map(doc => fromGeminiDocument(doc, storeId)) };
-    }
-}
-/**
- * Wait for a long-running file ingestion operation to finish and map the
- * result into `VectorStoreFileInfo`. Honors `wait`/`pollInterval`/
- * `pollTimeout` from `VectorStoreAddOptions` (defaults: wait=true,
- * interval=1000ms, timeout=120_000ms).
- *
- * The terminal state of a Gemini ingestion op is exposed two ways:
- * - `op.error?: { code, message }` when ingestion failed.
- * - `op.response?: { documentName: 'fileSearchStores/.../documents/...' }`
- *   when successful.
- *
- * On success we follow up with a single `documents.get` to fetch
- * `state` / `sizeBytes` / `createTime`. On failure we surface the error
- * message via `lastError` and the status flips to `'failed'`.
- */
-async function finishVectorStoreOperation(client, initialOp, storeId, opts) {
-    if (opts.wait === false) {
-        return {
-            id: initialOp?.name ?? `${storeId}/documents/pending-${Date.now()}`,
-            vectorStoreId: storeId,
-            status: 'in_progress',
-            createdAt: Math.floor(Date.now() / 1000),
-            ...(opts.attributes ? { attributes: opts.attributes } : {}),
-        };
-    }
-    const pollInterval = opts.pollInterval ?? 1000;
-    const pollTimeout = opts.pollTimeout ?? 120_000;
-    const deadline = Date.now() + pollTimeout;
-    let current = initialOp;
-    while (!current?.done) {
-        if (Date.now() > deadline) {
-            throw new Error(`[Rudder AI] Gemini FileSearchStore ingestion timed out after ${pollTimeout}ms ` +
-                `(store=${storeId}). Increase pollTimeout or set wait: false for fire-and-forget.`);
-        }
-        await sleep(pollInterval);
-        current = await client.operations.get({ operation: current });
-    }
-    if (current.error) {
-        const errMessage = current.error.message ?? 'unknown error';
-        return {
-            id: current.name ?? `${storeId}/documents/failed-${Date.now()}`,
-            vectorStoreId: storeId,
-            status: 'failed',
-            createdAt: Math.floor(Date.now() / 1000),
-            lastError: errMessage,
-            ...(opts.attributes ? { attributes: opts.attributes } : {}),
-        };
-    }
-    const documentName = current.response?.documentName;
-    if (!documentName) {
-        // Op done, no error, no documentName — surface as completed without
-        // follow-up details rather than failing.
-        return {
-            id: current.name ?? `${storeId}/documents/unknown-${Date.now()}`,
-            vectorStoreId: storeId,
-            status: 'completed',
-            createdAt: Math.floor(Date.now() / 1000),
-            ...(opts.attributes ? { attributes: opts.attributes } : {}),
-        };
-    }
-    // Follow up with documents.get to surface real createdAt + sizeBytes.
-    // Best-effort: if the get fails (rare race), fall back to the op data.
-    try {
-        const doc = await client.fileSearchStores.documents.get({ name: documentName });
-        const info = fromGeminiDocument(doc, storeId);
-        if (opts.attributes && !info.attributes)
-            info.attributes = opts.attributes;
-        return info;
-    }
-    catch {
-        return {
-            id: documentName,
-            vectorStoreId: storeId,
-            status: 'completed',
-            createdAt: Math.floor(Date.now() / 1000),
-            ...(opts.attributes ? { attributes: opts.attributes } : {}),
-        };
-    }
-}
-/**
- * Map a Gemini `FileSearchStore` resource into the framework's
- * `VectorStoreInfo` shape. `displayNameOverride` lets `create()` populate
- * the human-friendly name from the user-supplied value when the API
- * response omits it (some response variants do).
- *
- * @internal
- */
-export function fromGeminiFileSearchStore(raw, displayNameOverride) {
-    const r = raw;
-    const id = r.name ?? '';
-    const active = Number(r.activeDocumentsCount ?? 0) || 0;
-    const pending = Number(r.pendingDocumentsCount ?? 0) || 0;
-    const result = {
-        id,
-        name: r.displayName ?? displayNameOverride ?? id,
-        createdAt: r.createTime ? Math.floor(Date.parse(r.createTime) / 1000) : Math.floor(Date.now() / 1000),
-        fileCount: active + pending,
-    };
-    if (r.sizeBytes !== undefined) {
-        const bytes = Number(r.sizeBytes);
-        if (Number.isFinite(bytes))
-            result.bytesUsed = bytes;
-    }
-    return result;
-}
-/**
- * Map a Gemini `Document` resource into the framework's
- * `VectorStoreFileInfo` shape. `DocumentState` enum values flatten to the
- * shared `'in_progress' | 'completed' | 'failed' | 'cancelled'` union.
- *
- * @internal
- */
-export function fromGeminiDocument(raw, storeId) {
-    const r = raw;
-    const status = mapGeminiDocumentState(r.state);
-    const result = {
-        id: r.name ?? `${storeId}/documents/unknown`,
-        vectorStoreId: storeId,
-        status,
-        createdAt: r.createTime ? Math.floor(Date.parse(r.createTime) / 1000) : Math.floor(Date.now() / 1000),
-    };
-    if (r.sizeBytes !== undefined) {
-        const bytes = Number(r.sizeBytes);
-        if (Number.isFinite(bytes))
-            result.bytes = bytes;
-    }
-    if (r.customMetadata && r.customMetadata.length > 0) {
-        result.attributes = customMetadataToAttributes(r.customMetadata);
-    }
-    return result;
-}
-function mapGeminiDocumentState(state) {
-    switch (state) {
-        case 'STATE_ACTIVE': return 'completed';
-        case 'STATE_FAILED': return 'failed';
-        case 'STATE_PENDING': return 'in_progress';
-        default: return 'in_progress';
-    }
-}
-/**
- * Convert the framework's flat attribute map to Gemini's `CustomMetadata`
- * array shape. Strings → `stringValue`, numbers → `numericValue`,
- * booleans → `stringValue: 'true' | 'false'` (Gemini has no boolean
- * variant — string is the safe lossless choice; filter-builders can
- * still match on `key = "true"`).
- *
- * @internal
- */
-export function attributesToCustomMetadata(attrs) {
-    return Object.entries(attrs).map(([key, value]) => {
-        if (typeof value === 'number')
-            return { key, numericValue: value };
-        if (typeof value === 'boolean')
-            return { key, stringValue: value ? 'true' : 'false' };
-        return { key, stringValue: value };
-    });
-}
-/**
- * Inverse of {@link attributesToCustomMetadata}. Drops `stringListValue`
- * variants (no flat-attribute representation; apps that need lists
- * should read the raw Document via the SDK).
- *
- * @internal
- */
-export function customMetadataToAttributes(metadata) {
-    const out = {};
-    for (const entry of metadata) {
-        if (!entry.key)
-            continue;
-        if (entry.numericValue !== undefined)
-            out[entry.key] = entry.numericValue;
-        else if (entry.stringValue !== undefined) {
-            // Round-trip booleans encoded by attributesToCustomMetadata.
-            if (entry.stringValue === 'true')
-                out[entry.key] = true;
-            else if (entry.stringValue === 'false')
-                out[entry.key] = false;
-            else
-                out[entry.key] = entry.stringValue;
-        }
-        // stringListValue intentionally dropped.
-    }
-    return out;
-}
-function sleep(ms) {
-    return new Promise(resolve => setTimeout(resolve, ms));
-}
-/**
- * Best-effort MIME type from a filename extension. Gemini's
- * `uploadToFileSearchStore` requires a mimeType on Blob uploads (it
- * reads `blob.type`, which is empty on untyped `new Blob([data])`).
- *
- * Coverage matches Gemini's supported FileSearchStore document formats.
- * Unknown extensions return `''` — the caller drops the field so the
- * Gemini SDK's own error fires loudly rather than silently picking a
- * wrong type.
- *
- * @internal
- */
-export function mimeTypeFromFilename(filename) {
-    const ext = filename.toLowerCase().split('.').pop() ?? '';
-    switch (ext) {
-        case 'txt': return 'text/plain';
-        case 'md': return 'text/markdown';
-        case 'pdf': return 'application/pdf';
-        case 'html':
-        case 'htm': return 'text/html';
-        case 'json': return 'application/json';
-        case 'csv': return 'text/csv';
-        case 'tsv': return 'text/tab-separated-values';
-        case 'xml': return 'application/xml';
-        case 'rtf': return 'application/rtf';
-        case 'doc': return 'application/msword';
-        case 'docx': return 'application/vnd.openxmlformats-officedocument.wordprocessingml.document';
-        case 'js': return 'text/javascript';
-        case 'ts': return 'text/x-typescript';
-        case 'py': return 'text/x-python';
-        default: return '';
-    }
-}
-//# sourceMappingURL=google.js.map