@rudderjs/ai 1.17.2 → 1.18.0

package/dist/providers/openai.js

@@ -1,792 +0,0 @@
-import { cyrb53Hex } from '../util/hash.js';
-import { base64ToUtf8 } from '../base64.js';
-export class OpenAIProvider {
-    name = 'openai';
-    config;
-    constructor(config) {
-        this.config = config;
-    }
-    create(model) {
-        return new OpenAIAdapter(this.config, model);
-    }
-    createEmbedding(model) {
-        return new OpenAIEmbeddingAdapter(this.config, model);
-    }
-    createImage(model) {
-        return new OpenAIImageAdapter(this.config, model);
-    }
-    createTts(model) {
-        return new OpenAITtsAdapter(this.config, model);
-    }
-    createStt(model) {
-        return new OpenAISttAdapter(this.config, model);
-    }
-    createFiles() {
-        return new OpenAIFileAdapter(this.config);
-    }
-    createVectorStores() {
-        return new OpenAIVectorStoreAdapter(this.config);
-    }
-}
-// ─── Adapter (also reused by Ollama) ─────────────────────
-export class OpenAIAdapter {
-    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 */ 'openai');
-        const OpenAI = sdk.default ?? sdk.OpenAI;
-        this.client = new OpenAI({
-            apiKey: this.config.apiKey,
-            ...(this.config.baseUrl ? { baseURL: this.config.baseUrl } : {}),
-            ...(this.config.organization ? { organization: this.config.organization } : {}),
-            ...(this.config.defaultHeaders ? { defaultHeaders: this.config.defaultHeaders } : {}),
-        });
-        return this.client;
-    }
-    async generate(options) {
-        const client = await this.getClient();
-        const messages = toOpenAIMessages(options.messages);
-        const tools = options.tools?.length ? toOpenAITools(options.tools) : undefined;
-        const params = {
-            model: this.model,
-            messages,
-        };
-        if (options.maxTokens)
-            params['max_tokens'] = options.maxTokens;
-        if (options.temperature !== undefined)
-            params['temperature'] = options.temperature;
-        if (options.topP !== undefined)
-            params['top_p'] = options.topP;
-        if (options.stop)
-            params['stop'] = options.stop;
-        if (tools)
-            params['tools'] = tools;
-        if (options.toolChoice)
-            params['tool_choice'] = toOpenAIToolChoice(options.toolChoice);
-        const cacheKey = buildPromptCacheKey(messages, tools, options.cache);
-        if (cacheKey)
-            params['prompt_cache_key'] = cacheKey;
-        const response = await client.chat.completions.create(params, options.signal ? { signal: options.signal } : undefined);
-        return fromOpenAIResponse(response);
-    }
-    async *stream(options) {
-        const client = await this.getClient();
-        const messages = toOpenAIMessages(options.messages);
-        const tools = options.tools?.length ? toOpenAITools(options.tools) : undefined;
-        const params = {
-            model: this.model,
-            messages,
-            stream: true,
-        };
-        if (options.maxTokens)
-            params['max_tokens'] = options.maxTokens;
-        if (options.temperature !== undefined)
-            params['temperature'] = options.temperature;
-        if (options.topP !== undefined)
-            params['top_p'] = options.topP;
-        if (options.stop)
-            params['stop'] = options.stop;
-        if (tools)
-            params['tools'] = tools;
-        if (options.toolChoice)
-            params['tool_choice'] = toOpenAIToolChoice(options.toolChoice);
-        const cacheKey = buildPromptCacheKey(messages, tools, options.cache);
-        if (cacheKey)
-            params['prompt_cache_key'] = cacheKey;
-        const stream = await client.chat.completions.create(params, options.signal ? { signal: options.signal } : undefined);
-        for await (const chunk of stream) {
-            const choice = chunk.choices?.[0];
-            if (!choice)
-                continue;
-            const delta = choice.delta;
-            if (delta?.content) {
-                yield { type: 'text-delta', text: delta.content };
-            }
-            if (delta?.tool_calls?.length) {
-                for (const tc of delta.tool_calls) {
-                    // OpenAI guarantees `index` on every tool_calls delta — it's the
-                    // only stable correlator across the start-delta (carries `id`) and
-                    // subsequent arg-only deltas. We thread it through StreamChunk so
-                    // the agent loop can route arg fragments to the right partial
-                    // under parallel tool calls.
-                    const index = typeof tc.index === 'number' ? tc.index : undefined;
-                    if (tc.id) {
-                        yield {
-                            type: 'tool-call-delta',
-                            toolCall: { id: tc.id, name: tc.function?.name },
-                            ...(index !== undefined ? { toolCallIndex: index } : {}),
-                        };
-                    }
-                    if (tc.function?.arguments) {
-                        yield {
-                            type: 'tool-call-delta',
-                            text: tc.function.arguments,
-                            ...(index !== undefined ? { toolCallIndex: index } : {}),
-                        };
-                    }
-                }
-            }
-            if (choice.finish_reason) {
-                yield {
-                    type: 'finish',
-                    finishReason: choice.finish_reason === 'tool_calls' ? 'tool_calls' : 'stop',
-                    usage: chunk.usage ? {
-                        promptTokens: chunk.usage.prompt_tokens ?? 0,
-                        completionTokens: chunk.usage.completion_tokens ?? 0,
-                        totalTokens: chunk.usage.total_tokens ?? 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 contentToOpenAIParts(content) {
-    if (typeof content === 'string')
-        return content;
-    return content.map(p => {
-        if (p.type === 'text')
-            return { type: 'text', text: p.text };
-        if (p.type === 'image')
-            return { type: 'image_url', image_url: { url: `data:${p.mimeType};base64,${p.data}` } };
-        // document — for text-based docs, decode to text; for PDFs, send as image_url (GPT-4o supports)
-        if (p.mimeType === 'application/pdf') {
-            return { type: 'file', file: { data: p.data, mime_type: p.mimeType } };
-        }
-        return { type: 'text', text: base64ToUtf8(p.data) };
-    });
-}
-/**
- * Repair tool-call ↔ tool-result adjacency before serializing for an
- * OpenAI-compatible provider.
- *
- * Anthropic carries tool results as content blocks inside user turns, so a
- * loosely-ordered transcript round-trips fine. The OpenAI wire protocol (and
- * strict implementers like DeepSeek) enforce two hard rules:
- *
- *   1. every `role:'tool'` message must immediately follow the `assistant`
- *      message whose `tool_calls` declares its `tool_call_id`, and
- *   2. every `tool_calls` entry on an assistant message must be answered by a
- *      following `role:'tool'` message before the next assistant/user turn.
- *
- * A persist→resume cycle (client-tool pause, approval round-trip, or an app
- * that re-stores assistant turns without their `toolCalls`) can violate
- * either rule, yielding `400 Messages with role 'tool' must be a response to
- * a preceding message with 'tool_calls'` — or its mirror, an unanswered
- * `tool_calls`. See `docs/plans/2026-06-11-deepseek-tool-transcript-400.md`.
- *
- * This pass enforces BOTH directions:
- *   - **Detached / out-of-order results** are pulled up to sit immediately
- *     after their parent assistant, in `tool_calls` order.
- *   - **Unanswered `tool_calls`** get a synthesized stub result so the
- *     request is well-formed (mirrors the placeholder strategy in
- *     `resumePendingToolCalls`).
- *   - **Orphan tool results** — whose `tool_call_id` is declared by no
- *     assistant message anywhere — are dropped; they can never be valid on
- *     the wire. (Lossy only when the app already discarded the parent's
- *     `toolCalls`; the framework can't reconstruct a deleted call.)
- *
- * Transcripts that already satisfy the invariant pass through unchanged
- * (same message object references), so the common single-run path pays only
- * a linear scan.
- */
-export function normalizeToolTranscript(messages) {
-    // Index tool results by the call id they answer (a FIFO queue per id
-    // tolerates pathological duplicate ids without dropping a message), and
-    // collect every call id any assistant message declares.
-    const resultsByCallId = new Map();
-    const declaredCallIds = new Set();
-    for (const m of messages) {
-        if (m.role === 'tool' && m.toolCallId) {
-            const queue = resultsByCallId.get(m.toolCallId);
-            if (queue)
-                queue.push(m);
-            else
-                resultsByCallId.set(m.toolCallId, [m]);
-        }
-        else if (m.role === 'assistant' && m.toolCalls?.length) {
-            for (const tc of m.toolCalls)
-                declaredCallIds.add(tc.id);
-        }
-    }
-    const out = [];
-    for (const m of messages) {
-        if (m.role === 'assistant' && m.toolCalls?.length) {
-            out.push(m);
-            // Emit each declared call's answer adjacent + in declaration order,
-            // claiming the real result wherever it sat or synthesizing a stub.
-            for (const tc of m.toolCalls) {
-                const real = resultsByCallId.get(tc.id)?.shift();
-                if (real) {
-                    out.push(real);
-                }
-                else {
-                    out.push({
-                        role: 'tool',
-                        toolCallId: tc.id,
-                        content: '[Rudder] tool result missing — synthesized to satisfy the OpenAI tool-call/tool-result protocol.',
-                    });
-                }
-            }
-            continue;
-        }
-        // A tool message is emitted only by its parent assistant block above —
-        // here it is either already-claimed (skip) or an orphan with no declaring
-        // assistant (drop). Either way, never emit it standalone.
-        if (m.role === 'tool')
-            continue;
-        out.push(m);
-    }
-    return out;
-}
-export function toOpenAIMessages(messages) {
-    return normalizeToolTranscript(messages).map(m => {
-        if (m.role === 'assistant' && m.toolCalls?.length) {
-            return {
-                role: 'assistant',
-                content: contentToString(m.content) || null,
-                tool_calls: m.toolCalls.map(tc => ({
-                    id: tc.id,
-                    type: 'function',
-                    function: { name: tc.name, arguments: JSON.stringify(tc.arguments) },
-                })),
-            };
-        }
-        if (m.role === 'tool') {
-            return {
-                role: 'tool',
-                tool_call_id: m.toolCallId,
-                content: typeof m.content === 'string' ? m.content : JSON.stringify(m.content),
-            };
-        }
-        // User messages with attachments → content array
-        if (Array.isArray(m.content)) {
-            return { role: m.role, content: contentToOpenAIParts(m.content) };
-        }
-        return { role: m.role, content: m.content };
-    });
-}
-export function toOpenAITools(tools) {
-    return tools.map(t => {
-        // Provider-native tool blocks: when a tool carries a recognized
-        // `providerHint`, emit OpenAI's native shape instead of the standard
-        // function-call schema. Currently:
-        //   - 'file-search' → { type: 'file_search', vector_store_ids, filters?,
-        //                       max_num_results? }. The model is trained on the
-        //                       native tool — quality is dramatically better
-        //                       than wrapping it as a function call, and the
-        //                       provider runs the search server-side so no
-        //                       client-side execute is needed.
-        if (t.providerHint?.type === 'file-search') {
-            const vectorStoreIds = t.providerHint['vector_store_ids'] ?? [];
-            const block = {
-                type: 'file_search',
-                vector_store_ids: vectorStoreIds,
-            };
-            if (t.providerHint['filters'] !== undefined)
-                block['filters'] = t.providerHint['filters'];
-            if (t.providerHint['max_num_results'] !== undefined)
-                block['max_num_results'] = t.providerHint['max_num_results'];
-            return block;
-        }
-        return {
-            type: 'function',
-            function: { name: t.name, description: t.description, parameters: t.parameters },
-        };
-    });
-}
-function toOpenAIToolChoice(choice) {
-    if (choice === 'auto')
-        return 'auto';
-    if (choice === 'required')
-        return 'required';
-    if (choice === 'none')
-        return 'none';
-    if (typeof choice === 'object' && 'name' in choice)
-        return { type: 'function', function: { name: choice.name } };
-    return 'auto';
-}
-// ─── Prompt-cache key ────────────────────────────────────
-//
-// OpenAI caches prompts automatically once they exceed 1024 tokens. The only
-// SDK knob is `prompt_cache_key`: an opaque string that gives OpenAI a routing
-// hint so requests with the same cacheable prefix land on the same backend
-// (which has the prefix already cached). Stable hashing is the goal — not
-// cryptographic strength — so we use cyrb53 over canonical JSON of the
-// regions the agent declared as `cacheable()`.
-//
-// Spec: https://platform.openai.com/docs/guides/prompt-caching
-/**
- * Build a stable `prompt_cache_key` from the regions the agent marked as
- * cacheable. Returns `undefined` if no markers apply (request goes out
- * without a cache key — OpenAI still caches automatically above 1024
- * tokens, just without routing affinity).
- *
- * Exported for unit testing.
- */
-export function buildPromptCacheKey(messages, tools, cache) {
-    if (!cache)
-        return undefined;
-    const parts = [];
-    if (cache.instructions) {
-        const sys = messages.find(m => m.role === 'system');
-        if (sys)
-            parts.push({ s: sys.content });
-    }
-    if (cache.tools && tools && tools.length > 0) {
-        parts.push({ t: tools });
-    }
-    if (cache.messages && cache.messages > 0) {
-        const conv = messages.filter(m => m.role !== 'system');
-        const sliced = conv.slice(0, cache.messages);
-        if (sliced.length > 0)
-            parts.push({ m: sliced });
-    }
-    if (parts.length === 0)
-        return undefined;
-    return cyrb53Hex(JSON.stringify(parts));
-}
-function fromOpenAIResponse(response) {
-    const choice = response.choices?.[0];
-    const message = choice?.message;
-    const toolCalls = (message?.tool_calls ?? []).map((tc) => ({
-        id: tc.id,
-        name: tc.function.name,
-        arguments: JSON.parse(tc.function.arguments),
-    }));
-    return {
-        message: {
-            role: 'assistant',
-            content: message?.content ?? '',
-            ...(toolCalls.length > 0 ? { toolCalls } : {}),
-        },
-        usage: {
-            promptTokens: response.usage?.prompt_tokens ?? 0,
-            completionTokens: response.usage?.completion_tokens ?? 0,
-            totalTokens: response.usage?.total_tokens ?? 0,
-        },
-        finishReason: choice?.finish_reason === 'tool_calls' ? 'tool_calls' : 'stop',
-    };
-}
-// ─── Embedding Adapter ────────────────────────────────────
-class OpenAIEmbeddingAdapter {
-    config;
-    model;
-    constructor(config, model) {
-        this.config = config;
-        this.model = model;
-    }
-    async embed(input) {
-        const baseUrl = this.config.baseUrl ?? 'https://api.openai.com/v1';
-        const inputs = Array.isArray(input) ? input : [input];
-        const res = await fetch(`${baseUrl}/embeddings`, {
-            method: 'POST',
-            headers: {
-                'Authorization': `Bearer ${this.config.apiKey}`,
-                'Content-Type': 'application/json',
-                ...(this.config.organization ? { 'OpenAI-Organization': this.config.organization } : {}),
-                ...(this.config.defaultHeaders ?? {}),
-            },
-            body: JSON.stringify({ model: this.model, input: inputs }),
-        });
-        if (!res.ok)
-            throw new Error(`[Rudder AI] OpenAI embeddings error: ${res.status} ${await res.text()}`);
-        const data = await res.json();
-        return {
-            embeddings: data.data.map(d => d.embedding),
-            usage: { promptTokens: data.usage.prompt_tokens, totalTokens: data.usage.total_tokens },
-        };
-    }
-}
-// ─── Image Generation Adapter ────────────────────────────
-const IMAGE_SIZE_MAP = {
-    square: '1024x1024',
-    landscape: '1792x1024',
-    portrait: '1024x1792',
-};
-class OpenAIImageAdapter {
-    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 */ 'openai');
-        const OpenAI = sdk.default ?? sdk.OpenAI;
-        this.client = new OpenAI({
-            apiKey: this.config.apiKey,
-            ...(this.config.baseUrl ? { baseURL: this.config.baseUrl } : {}),
-            ...(this.config.organization ? { organization: this.config.organization } : {}),
-            ...(this.config.defaultHeaders ? { defaultHeaders: this.config.defaultHeaders } : {}),
-        });
-        return this.client;
-    }
-    async generate(options) {
-        const client = await this.getClient();
-        const size = options.size
-            ? (IMAGE_SIZE_MAP[options.size] ?? options.size)
-            : '1024x1024';
-        const params = {
-            model: this.model,
-            prompt: options.prompt,
-            size,
-            response_format: 'b64_json',
-        };
-        if (options.n !== undefined)
-            params['n'] = options.n;
-        if (options.quality)
-            params['quality'] = options.quality;
-        if (options.style)
-            params['style'] = options.style;
-        const response = await client.images.generate(params);
-        return {
-            images: (response.data ?? []).map((img) => ({
-                ...(img.b64_json ? { base64: img.b64_json } : {}),
-                ...(img.url ? { url: img.url } : {}),
-                ...(img.revised_prompt ? { revisedPrompt: img.revised_prompt } : {}),
-            })),
-            model: this.model,
-        };
-    }
-}
-// ─── TTS Adapter ─────────────────────────────────────────
-class OpenAITtsAdapter {
-    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 */ 'openai');
-        const OpenAI = sdk.default ?? sdk.OpenAI;
-        this.client = new OpenAI({
-            apiKey: this.config.apiKey,
-            ...(this.config.baseUrl ? { baseURL: this.config.baseUrl } : {}),
-            ...(this.config.organization ? { organization: this.config.organization } : {}),
-            ...(this.config.defaultHeaders ? { defaultHeaders: this.config.defaultHeaders } : {}),
-        });
-        return this.client;
-    }
-    async generate(options) {
-        const client = await this.getClient();
-        const format = options.format ?? 'mp3';
-        const params = {
-            model: this.model,
-            input: options.text,
-            voice: options.voice ?? 'alloy',
-        };
-        if (options.speed !== undefined)
-            params['speed'] = options.speed;
-        if (options.format)
-            params['response_format'] = options.format;
-        const response = await client.audio.speech.create(params);
-        const arrayBuffer = await response.arrayBuffer();
-        return {
-            audio: Buffer.from(arrayBuffer),
-            format,
-            model: this.model,
-        };
-    }
-}
-// ─── STT Adapter ─────────────────────────────────────────
-class OpenAISttAdapter {
-    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 */ 'openai');
-        const OpenAI = sdk.default ?? sdk.OpenAI;
-        this.client = new OpenAI({
-            apiKey: this.config.apiKey,
-            ...(this.config.baseUrl ? { baseURL: this.config.baseUrl } : {}),
-            ...(this.config.organization ? { organization: this.config.organization } : {}),
-            ...(this.config.defaultHeaders ? { defaultHeaders: this.config.defaultHeaders } : {}),
-        });
-        return this.client;
-    }
-    async transcribe(options) {
-        const client = await this.getClient();
-        const file = new File([options.audio], 'audio.mp3', { type: 'audio/mpeg' });
-        const params = {
-            model: this.model,
-            file,
-            response_format: 'verbose_json',
-        };
-        if (options.language)
-            params['language'] = options.language;
-        if (options.prompt)
-            params['prompt'] = options.prompt;
-        const response = await client.audio.transcriptions.create(params);
-        return {
-            text: response.text,
-            language: response.language,
-            duration: response.duration,
-            model: this.model,
-        };
-    }
-}
-// ─── Files ──────────────────────────────────────────────
-class OpenAIFileAdapter {
-    config;
-    client = null;
-    constructor(config) {
-        this.config = config;
-    }
-    async getClient() {
-        if (this.client)
-            return this.client;
-        const sdk = await import(/* @vite-ignore */ 'openai');
-        const OpenAI = sdk.default ?? sdk.OpenAI;
-        this.client = new OpenAI({
-            apiKey: this.config.apiKey,
-            ...(this.config.baseUrl ? { baseURL: this.config.baseUrl } : {}),
-            ...(this.config.organization ? { organization: this.config.organization } : {}),
-            ...(this.config.defaultHeaders ? { defaultHeaders: this.config.defaultHeaders } : {}),
-        });
-        return this.client;
-    }
-    async upload(options) {
-        const client = await this.getClient();
-        const { createReadStream } = await import(/* @vite-ignore */ 'node:fs');
-        const file = createReadStream(options.filePath);
-        const response = await client.files.create({
-            file,
-            purpose: options.purpose ?? 'assistants',
-        });
-        return {
-            id: response.id,
-            filename: response.filename,
-            bytes: response.bytes,
-            purpose: response.purpose,
-        };
-    }
-    async list() {
-        const client = await this.getClient();
-        const response = await client.files.list();
-        const files = [];
-        for await (const f of response) {
-            files.push({
-                id: f.id,
-                filename: f.filename,
-                bytes: f.bytes,
-                purpose: f.purpose,
-            });
-        }
-        return { files };
-    }
-    async delete(fileId) {
-        const client = await this.getClient();
-        await client.files.del(fileId);
-    }
-    async retrieve(fileId) {
-        const client = await this.getClient();
-        const response = await client.files.content(fileId);
-        const buffer = Buffer.from(await response.arrayBuffer());
-        return { data: buffer, mimeType: 'application/octet-stream' };
-    }
-}
-// ─── OpenAI Vector Stores (#B8 Phase 1) ──────────────────
-/**
- * OpenAI hosted vector store adapter. Wraps `client.vectorStores.*` and
- * `client.vectorStores.files.*` from the v4+ SDK. Lazy SDK load mirrors
- * the rest of the OpenAI provider.
- *
- * `addFile` defaults to polling until the file is fully indexed
- * (`status === 'completed'`). Pass `{ wait: false }` to fire-and-forget.
- *
- * Local file paths route through OpenAI's Files API first
- * (`client.files.create({ purpose: 'assistants' })`); the resulting
- * `file_id` then attaches to the vector store. Apps that already have
- * a file id pass `{ fileId }` directly.
- */
-class OpenAIVectorStoreAdapter {
-    config;
-    client = null;
-    constructor(config) {
-        this.config = config;
-    }
-    async getClient() {
-        if (this.client)
-            return this.client;
-        const sdk = await import(/* @vite-ignore */ 'openai');
-        const OpenAI = sdk.default ?? sdk.OpenAI;
-        this.client = new OpenAI({
-            apiKey: this.config.apiKey,
-            ...(this.config.baseUrl ? { baseURL: this.config.baseUrl } : {}),
-            ...(this.config.organization ? { organization: this.config.organization } : {}),
-            ...(this.config.defaultHeaders ? { defaultHeaders: this.config.defaultHeaders } : {}),
-        });
-        return this.client;
-    }
-    async create(opts) {
-        const client = await this.getClient();
-        const params = { name: opts.name };
-        if (opts.metadata)
-            params['metadata'] = opts.metadata;
-        if (opts.expiresAfter)
-            params['expires_after'] = opts.expiresAfter;
-        const response = await client.vectorStores.create(params);
-        return fromOpenAIVectorStore(response);
-    }
-    async list(opts) {
-        const client = await this.getClient();
-        const params = {};
-        if (opts?.limit !== undefined)
-            params['limit'] = opts.limit;
-        if (opts?.after !== undefined)
-            params['after'] = opts.after;
-        if (opts?.before !== undefined)
-            params['before'] = opts.before;
-        const response = await client.vectorStores.list(params);
-        const data = (response.data ?? []);
-        return { stores: data.map(d => fromOpenAIVectorStore(d)) };
-    }
-    async get(id) {
-        const client = await this.getClient();
-        const response = await client.vectorStores.retrieve(id);
-        return fromOpenAIVectorStore(response);
-    }
-    async delete(id) {
-        const client = await this.getClient();
-        await client.vectorStores.del(id);
-    }
-    async addFile(storeId, opts) {
-        const client = await this.getClient();
-        // Step 1: resolve the file id. If the user passed an existing one we
-        // skip the upload; otherwise upload via the standard Files API and
-        // reuse the resulting id.
-        const fileId = opts.fileId ?? await this.uploadAndGetId(client, opts);
-        // Step 2: attach to the store. OpenAI splits attribute + chunking
-        // config from the file payload so we pass them as a sibling object.
-        const attachParams = { file_id: fileId };
-        if (opts.attributes)
-            attachParams['attributes'] = opts.attributes;
-        if (opts.chunkingStrategy)
-            attachParams['chunking_strategy'] = opts.chunkingStrategy;
-        const attached = await client.vectorStores.files.create(storeId, attachParams);
-        if (opts.wait === false) {
-            return fromOpenAIVectorStoreFile(attached, storeId);
-        }
-        // Step 3: poll until `completed` / `failed` / timeout. Default 2-min
-        // budget — enough for typical PDFs but small enough that runaway
-        // batch uploads surface a clear error fast.
-        const pollInterval = opts.pollInterval ?? 1000;
-        const pollTimeout = opts.pollTimeout ?? 120_000;
-        const deadline = Date.now() + pollTimeout;
-        let current = attached;
-        while (true) {
-            const info = fromOpenAIVectorStoreFile(current, storeId);
-            if (info.status === 'completed' || info.status === 'failed' || info.status === 'cancelled') {
-                return info;
-            }
-            if (Date.now() > deadline) {
-                throw new Error(`[Rudder AI] vector-store file ingestion timed out after ${pollTimeout}ms ` +
-                    `(store=${storeId}, file=${fileId}, status=${info.status}). ` +
-                    'Increase pollTimeout or set wait: false for fire-and-forget.');
-            }
-            await sleep(pollInterval);
-            current = await client.vectorStores.files.retrieve(storeId, fileId);
-        }
-    }
-    async removeFile(storeId, fileId) {
-        const client = await this.getClient();
-        await client.vectorStores.files.del(storeId, fileId);
-    }
-    async listFiles(storeId, opts) {
-        const client = await this.getClient();
-        const params = {};
-        if (opts?.limit !== undefined)
-            params['limit'] = opts.limit;
-        if (opts?.after !== undefined)
-            params['after'] = opts.after;
-        if (opts?.before !== undefined)
-            params['before'] = opts.before;
-        const response = await client.vectorStores.files.list(storeId, params);
-        const data = (response.data ?? []);
-        return { files: data.map(d => fromOpenAIVectorStoreFile(d, storeId)) };
-    }
-    /** @internal — upload a local file via the Files API and return the
-     *  provider's file id. Used when the user passes `filePath` or
-     *  `fileBuffer` to `addFile` instead of an existing `fileId`. */
-    async uploadAndGetId(client, opts) {
-        if (opts.filePath) {
-            const { createReadStream } = await import(/* @vite-ignore */ 'node:fs');
-            const file = createReadStream(opts.filePath);
-            const uploaded = await client.files.create({ file, purpose: 'assistants' });
-            return uploaded.id;
-        }
-        if (opts.fileBuffer) {
-            const { toFile } = await import(/* @vite-ignore */ 'openai/uploads');
-            const file = await toFile(opts.fileBuffer.data, opts.fileBuffer.filename);
-            const uploaded = await client.files.create({ file, purpose: 'assistants' });
-            return uploaded.id;
-        }
-        throw new Error('[Rudder AI] addFile requires fileId, filePath, or fileBuffer. ' +
-            'Pass an existing OpenAI file id via { fileId } or a local source via { filePath }.');
-    }
-}
-function fromOpenAIVectorStore(raw) {
-    const r = raw;
-    const fileCount = r.file_counts?.total ??
-        (r.file_counts ? (r.file_counts.in_progress ?? 0) + (r.file_counts.completed ?? 0) + (r.file_counts.failed ?? 0) + (r.file_counts.cancelled ?? 0) : 0);
-    const result = {
-        id: r.id,
-        name: r.name,
-        createdAt: r.created_at,
-        fileCount,
-    };
-    if (r.usage_bytes !== undefined)
-        result.bytesUsed = r.usage_bytes;
-    if (r.metadata !== undefined)
-        result.metadata = r.metadata;
-    return result;
-}
-function fromOpenAIVectorStoreFile(raw, storeId) {
-    const r = raw;
-    const status = r.status === 'completed' || r.status === 'failed' || r.status === 'cancelled' || r.status === 'in_progress'
-        ? r.status
-        : 'in_progress';
-    const result = {
-        id: r.id,
-        vectorStoreId: storeId,
-        status,
-        createdAt: r.created_at,
-    };
-    if (r.usage_bytes !== undefined)
-        result.bytes = r.usage_bytes;
-    if (r.attributes !== undefined)
-        result.attributes = r.attributes;
-    if (r.last_error?.message !== undefined)
-        result.lastError = r.last_error.message;
-    return result;
-}
-function sleep(ms) {
-    return new Promise(resolve => setTimeout(resolve, ms));
-}
-//# sourceMappingURL=openai.js.map