@byfriends/kosong 0.2.2 → 0.3.2

package/dist/provider-common-CaxKVTTJ.mjs

@@ -0,0 +1,151 @@
+//#region src/errors.ts
+/**
+* Base error for all chat provider errors.
+*/
+var ChatProviderError = class extends Error {
+	constructor(message) {
+		super(message);
+		this.name = "ChatProviderError";
+	}
+};
+/**
+* Network-level connection failure.
+*/
+var APIConnectionError = class extends ChatProviderError {
+	constructor(message) {
+		super(message);
+		this.name = "APIConnectionError";
+	}
+};
+/**
+* Request timed out.
+*/
+var APITimeoutError = class extends ChatProviderError {
+	constructor(message) {
+		super(message);
+		this.name = "APITimeoutError";
+	}
+};
+/**
+* HTTP status error from the API.
+*/
+var APIStatusError = class extends ChatProviderError {
+	statusCode;
+	requestId;
+	constructor(statusCode, message, requestId) {
+		super(message);
+		this.name = "APIStatusError";
+		this.statusCode = statusCode;
+		this.requestId = requestId ?? null;
+	}
+};
+/**
+* HTTP status error that specifically means the request exceeded the model
+* context window.
+*/
+var APIContextOverflowError = class extends APIStatusError {
+	constructor(statusCode, message, requestId) {
+		super(statusCode, message, requestId);
+		this.name = "APIContextOverflowError";
+	}
+};
+/**
+* The API returned an empty response (no content, no tool calls).
+*/
+var APIEmptyResponseError = class extends ChatProviderError {
+	constructor(message) {
+		super(message);
+		this.name = "APIEmptyResponseError";
+	}
+};
+const CONTEXT_OVERFLOW_MESSAGE_PATTERNS = [
+	/context[ _-]?length/,
+	/(?:context[ _-]?window.*exceed|exceed.*context[ _-]?window)/,
+	/maximum context/,
+	/exceed(?:ed|s|ing)?\s+(?:the\s+)?max(?:imum)?\s+tokens?/,
+	/(?:too many tokens.*(?:prompt|input|context)|(?:prompt|input|context).*too many tokens)/,
+	/prompt is too long.*maximum/,
+	/input token count.*exceeds?.*maximum number of tokens/
+];
+function normalizeAPIStatusError(statusCode, message, requestId) {
+	if (isContextOverflowStatusError(statusCode, message)) return new APIContextOverflowError(statusCode, message, requestId);
+	return new APIStatusError(statusCode, message, requestId);
+}
+function isContextOverflowStatusError(statusCode, message) {
+	if (statusCode !== 400 && statusCode !== 413 && statusCode !== 422) return false;
+	const lowerMessage = message.toLowerCase();
+	return CONTEXT_OVERFLOW_MESSAGE_PATTERNS.some((pattern) => pattern.test(lowerMessage));
+}
+//#endregion
+//#region src/providers/provider-common.ts
+/**
+* Build a finish-reason normalizer from a per-provider raw-string → FinishReason table.
+*
+* Mirrors the shape of the per-adapter `normalizeXxxFinishReason` functions:
+* - `null` / `undefined` raw → `{ finishReason: null, rawFinishReason: null }`
+* - raw present and in the table → mapped FinishReason, raw echoed back
+* - raw present but not in the table → `'other'`, raw echoed back
+*
+* The returned function is stateless and safe to call repeatedly.
+*/
+function makeFinishReasonNormalizer(mapping) {
+	return (raw) => {
+		if (raw === null || raw === void 0) return {
+			finishReason: null,
+			rawFinishReason: null
+		};
+		return {
+			finishReason: mapping[raw] ?? "other",
+			rawFinishReason: raw
+		};
+	};
+}
+/**
+* Build the four-field `TokenUsage` from already-parsed per-provider numbers,
+* applying the `inputOther = total - cached` formula shared by OpenAI-style and
+* Google providers (which expose only a total prompt count and a cached subset).
+*
+* `inputOther` is clamped to ≥ 0 when `cached` exceeds `total` (defensive — a
+* provider should never report more cached than total, but we never emit a
+* negative usage field). Anthropic is excluded: it reports a real
+* `inputCacheCreation` field that does not fit this formula.
+*/
+function extractCacheUsage(total, cached, output) {
+	return {
+		inputOther: Math.max(0, total - cached),
+		output,
+		inputCacheRead: cached,
+		inputCacheCreation: 0
+	};
+}
+const NETWORK_RE = /network|connection|connect|disconnect/i;
+const TIMEOUT_RE = /timed?\s*out|timeout|deadline/i;
+/**
+* Convert a raw thrown value into a kosong `ChatProviderError` using the
+* shared message-based classification ladder:
+*
+* 1. already a `ChatProviderError` → returned as-is (identity)
+* 2. `status` provided → `normalizeAPIStatusError` (status + message + requestId)
+* 3. message matches `TIMEOUT_RE` → `APITimeoutError`
+* 4. message matches `NETWORK_RE` or any `extraNetworkMatchers`, or the value
+*    is a `TypeError` matching `extraTypeErrorMatch` → `APIConnectionError`
+* 5. otherwise → `ChatProviderError` wrapping the message
+*
+* Provider adapters that recognize SDK-specific error classes (e.g. OpenAI's
+* `APIConnectionTimeoutError`, Google's `GoogleApiError`) should unwrap them
+* into `(message, status?, requestId?)` before calling this function. The
+* SDK-class detection itself is provider-specific and stays in the adapter.
+*/
+function convertProviderError(error, opts = {}) {
+	if (error instanceof ChatProviderError) return error;
+	const message = error instanceof Error ? error.message : String(error);
+	if (typeof opts.status === "number") return normalizeAPIStatusError(opts.status, message, opts.requestId);
+	if (TIMEOUT_RE.test(message)) return new APITimeoutError(message);
+	if (NETWORK_RE.test(message)) return new APIConnectionError(message);
+	if (opts.extraNetworkMatchers?.some((re) => re.test(message))) return new APIConnectionError(message);
+	if (opts.extraTypeErrorMatch !== void 0 && error instanceof TypeError && message.includes(opts.extraTypeErrorMatch)) return new APIConnectionError(message);
+	if (error instanceof Error) return new ChatProviderError(`Error: ${message}`);
+	return new ChatProviderError(`Error: ${String(error)}`);
+}
+//#endregion
+export { APIContextOverflowError as a, APITimeoutError as c, APIConnectionError as i, ChatProviderError as l, extractCacheUsage as n, APIEmptyResponseError as o, makeFinishReasonNormalizer as r, APIStatusError as s, convertProviderError as t, normalizeAPIStatusError as u };

package/dist/providers/anthropic.d.mts

@@ -1,2 +1,2 @@
-import { i as resolveDefaultMaxTokens, n as AnthropicOptions, r as convertAnthropicError, t as AnthropicChatProvider } from "../anthropic-0CVG5rBE.mjs";
+import { i as resolveDefaultMaxTokens, n as AnthropicOptions, r as convertAnthropicError, t as AnthropicChatProvider } from "../anthropic-BYlmhwZs.mjs";
 export { AnthropicChatProvider, AnthropicOptions, convertAnthropicError, resolveDefaultMaxTokens };

package/dist/providers/anthropic.mjs

@@ -1,5 +1,5 @@
-import { a as APITimeoutError, o as ChatProviderError, s as normalizeAPIStatusError, t as APIConnectionError$1 } from "../errors-WFxxzL1B.mjs";
-import { i as getAnthropicModelCapability, n as requireProviderApiKey, r as resolveAuthBackedClient, t as mergeRequestHeaders } from "../request-auth-BMXt8jRu.mjs";
+import { c as APITimeoutError, i as APIConnectionError$1, l as ChatProviderError, r as makeFinishReasonNormalizer, u as normalizeAPIStatusError } from "../provider-common-CaxKVTTJ.mjs";
+import { n as BaseChatProvider, o as getAnthropicModelCapability, r as mergeRequestHeaders, t as BaseStreamedMessage } from "../base-streamed-message-Kok6Drxr.mjs";
 import Anthropic, { APIConnectionError, APIConnectionTimeoutError, APIError, AnthropicError } from "@anthropic-ai/sdk";
 //#region src/providers/anthropic.ts
 /**
@@ -9,39 +9,14 @@ import Anthropic, { APIConnectionError, APIConnectionTimeoutError, APIError, Ant
 * Source: `message.stop_reason` (non-stream) or the last `message_delta`
 * event's `delta.stop_reason` (stream).
 */
-function normalizeAnthropicStopReason(raw) {
-	if (raw === null || raw === void 0) return {
-		finishReason: null,
-		rawFinishReason: null
-	};
-	switch (raw) {
-		case "end_turn":
-		case "stop_sequence": return {
-			finishReason: "completed",
-			rawFinishReason: raw
-		};
-		case "max_tokens": return {
-			finishReason: "truncated",
-			rawFinishReason: raw
-		};
-		case "tool_use": return {
-			finishReason: "tool_calls",
-			rawFinishReason: raw
-		};
-		case "pause_turn": return {
-			finishReason: "paused",
-			rawFinishReason: raw
-		};
-		case "refusal": return {
-			finishReason: "filtered",
-			rawFinishReason: raw
-		};
-		default: return {
-			finishReason: "other",
-			rawFinishReason: raw
-		};
-	}
-}
+const normalizeAnthropicStopReason = makeFinishReasonNormalizer({
+	end_turn: "completed",
+	stop_sequence: "completed",
+	max_tokens: "truncated",
+	tool_use: "tool_calls",
+	pause_turn: "paused",
+	refusal: "filtered"
+});
 const INTERLEAVED_THINKING_BETA = "interleaved-thinking-2025-05-14";
 const FAMILY_VERSION_RE = /(?:opus|sonnet|haiku)[.-](\d+)[.-](\d{1,2})(?!\d)/;
 const OPUS_VERSION_RE = /opus[.-](\d+)[.-](\d{1,2})(?!\d)/;
@@ -355,35 +330,23 @@ function convertAnthropicError(error) {
 	if (error instanceof Error) return new ChatProviderError(`Error: ${error.message}`);
 	return new ChatProviderError(`Error: ${String(error)}`);
 }
-var AnthropicStreamedMessage = class {
-	_id = null;
-	_usage = {
-		inputOther: 0,
-		output: 0,
-		inputCacheRead: 0,
-		inputCacheCreation: 0
-	};
-	_finishReason = null;
-	_rawFinishReason = null;
-	_iter;
+var AnthropicStreamedMessage = class extends BaseStreamedMessage {
+	_response;
+	_isStream;
 	constructor(response, isStream) {
-		if (isStream) this._iter = this._convertStreamResponse(response);
-		else this._iter = this._convertNonStreamResponse(response);
-	}
-	get id() {
-		return this._id;
-	}
-	get usage() {
-		return this._usage;
-	}
-	get finishReason() {
-		return this._finishReason;
-	}
-	get rawFinishReason() {
-		return this._rawFinishReason;
+		super();
+		this._response = response;
+		this._isStream = isStream;
+		this._usage = {
+			inputOther: 0,
+			output: 0,
+			inputCacheRead: 0,
+			inputCacheCreation: 0
+		};
 	}
-	async *[Symbol.asyncIterator]() {
-		yield* this._iter;
+	_buildIter() {
+		if (this._isStream) return this._convertStreamResponse(this._response);
+		return this._convertNonStreamResponse(this._response);
 	}
 	_captureStopReason(raw) {
 		const normalized = normalizeAnthropicStopReason(raw);
@@ -391,11 +354,14 @@ var AnthropicStreamedMessage = class {
 		this._rawFinishReason = normalized.rawFinishReason;
 	}
 	_extractUsage(usage) {
+		const inputTokens = usage.input_tokens ?? 0;
+		const cacheRead = usage.cache_read_input_tokens ?? 0;
+		const cacheCreation = usage.cache_creation_input_tokens ?? 0;
 		this._usage = {
-			inputOther: usage.input_tokens ?? 0,
+			inputOther: Math.max(0, inputTokens - cacheRead - cacheCreation),
 			output: usage.output_tokens ?? 0,
-			inputCacheRead: usage.cache_read_input_tokens ?? 0,
-			inputCacheCreation: usage.cache_creation_input_tokens ?? 0
+			inputCacheRead: cacheRead,
+			inputCacheCreation: cacheCreation
 		};
 	}
 	async *_convertNonStreamResponse(response) {
@@ -520,9 +486,16 @@ var AnthropicStreamedMessage = class {
 					const deltaUsage = evt.usage;
 					if (deltaUsage !== void 0) {
 						if (typeof deltaUsage["output_tokens"] === "number") this._usage.output = deltaUsage["output_tokens"];
+						const prevInputOther = this._usage.inputOther;
+						const prevCacheRead = this._usage.inputCacheRead;
+						const prevCacheCreation = this._usage.inputCacheCreation;
 						if (typeof deltaUsage["cache_read_input_tokens"] === "number") this._usage.inputCacheRead = deltaUsage["cache_read_input_tokens"];
 						if (typeof deltaUsage["cache_creation_input_tokens"] === "number") this._usage.inputCacheCreation = deltaUsage["cache_creation_input_tokens"];
-						if (typeof deltaUsage["input_tokens"] === "number") this._usage.inputOther = deltaUsage["input_tokens"];
+						if (typeof deltaUsage["input_tokens"] === "number") this._usage.inputOther = Math.max(0, deltaUsage["input_tokens"] - this._usage.inputCacheRead - this._usage.inputCacheCreation);
+						else {
+							const totalInput = prevInputOther + prevCacheRead + prevCacheCreation;
+							this._usage.inputOther = Math.max(0, totalInput - this._usage.inputCacheRead - this._usage.inputCacheCreation);
+						}
 					}
 					const messageDeltaPayload = evt.delta;
 					if (messageDeltaPayload !== void 0 && "stop_reason" in messageDeltaPayload) this._captureStopReason(messageDeltaPayload["stop_reason"]);
@@ -533,31 +506,22 @@ var AnthropicStreamedMessage = class {
 		}
 	}
 };
-var AnthropicChatProvider = class {
+var AnthropicChatProvider = class AnthropicChatProvider extends BaseChatProvider {
 	name = "anthropic";
-	_model;
 	_stream;
-	_client;
-	_generationKwargs;
 	_metadata;
-	_apiKey;
-	_baseUrl;
-	_defaultHeaders;
-	_clientFactory;
 	constructor(options) {
-		this._model = options.model;
-		this._stream = options.stream ?? true;
-		this._metadata = options.metadata;
-		const apiKey = options.apiKey ?? process.env["ANTHROPIC_API_KEY"];
-		this._apiKey = apiKey === void 0 || apiKey.length === 0 ? void 0 : apiKey;
-		this._baseUrl = options.baseUrl;
-		this._defaultHeaders = options.defaultHeaders;
-		this._clientFactory = options.clientFactory;
-		this._client = this._apiKey === void 0 ? void 0 : this._buildClient(this._apiKey);
-		this._generationKwargs = {
+		const apiKey = options.apiKey === void 0 || options.apiKey.length === 0 ? process.env["ANTHROPIC_API_KEY"] ?? void 0 : options.apiKey;
+		const apiKeyResolved = apiKey === void 0 || apiKey.length === 0 ? void 0 : apiKey;
+		const baseUrl = options.baseUrl;
+		const client = apiKeyResolved === void 0 ? void 0 : AnthropicChatProvider.buildClient(apiKeyResolved, baseUrl, options.defaultHeaders);
+		const generationKwargs = {
 			max_tokens: resolveDefaultMaxTokens(options.model, options.defaultMaxTokens),
 			betaFeatures: options.betaFeatures ?? [INTERLEAVED_THINKING_BETA]
 		};
+		super(options.model, generationKwargs, apiKeyResolved, baseUrl ?? "", options.defaultHeaders, client, options.clientFactory);
+		this._stream = options.stream ?? true;
+		this._metadata = options.metadata;
 	}
 	get modelName() {
 		return this._model;
@@ -650,24 +614,25 @@ var AnthropicChatProvider = class {
 			throw convertAnthropicError(error);
 		}
 	}
-	_createClient(auth) {
-		return resolveAuthBackedClient({
-			cachedClient: this._client,
-			clientFactory: this._clientFactory
-		}, auth, (a) => this._buildClient(requireProviderApiKey("AnthropicChatProvider", a, this._apiKey)));
-	}
-	_buildClient(apiKey) {
+	static buildClient(apiKey, baseUrl, defaultHeaders) {
 		return new Anthropic({
 			apiKey,
+			baseURL: baseUrl,
+			defaultHeaders
+		});
+	}
+	createRawClient(auth, defaultHeaders) {
+		return new Anthropic({
+			apiKey: auth.apiKey,
 			baseURL: this._baseUrl,
-			defaultHeaders: this._defaultHeaders
+			defaultHeaders
 		});
 	}
 	withThinking(effort) {
 		if (effort === "off") {
 			let newBetas = [...this._generationKwargs.betaFeatures ?? []];
 			newBetas = newBetas.filter((b) => b !== INTERLEAVED_THINKING_BETA);
-			const clone = this._withGenerationKwargs({
+			const clone = this.withGenerationKwargs({
 				thinking: { type: "disabled" },
 				betaFeatures: newBetas
 			});
@@ -678,7 +643,7 @@ var AnthropicChatProvider = class {
 		if (effectiveEffort === "off") throw new Error("Non-off thinking effort unexpectedly clamped to off.");
 		let newBetas = [...this._generationKwargs.betaFeatures ?? []];
 		newBetas = newBetas.filter((b) => b !== INTERLEAVED_THINKING_BETA);
-		return this._withGenerationKwargs({
+		return this.withGenerationKwargs({
 			thinking: {
 				type: "adaptive",
 				display: "summarized"
@@ -687,22 +652,6 @@ var AnthropicChatProvider = class {
 			betaFeatures: newBetas
 		});
 	}
-	withGenerationKwargs(kwargs) {
-		return this._withGenerationKwargs(kwargs);
-	}
-	_withGenerationKwargs(kwargs) {
-		const clone = this._clone();
-		clone._generationKwargs = {
-			...clone._generationKwargs,
-			...kwargs
-		};
-		return clone;
-	}
-	_clone() {
-		const clone = Object.assign(Object.create(Object.getPrototypeOf(this)), this);
-		clone._generationKwargs = { ...this._generationKwargs };
-		return clone;
-	}
 };
 //#endregion
 export { AnthropicChatProvider, convertAnthropicError, resolveDefaultMaxTokens };

package/dist/providers/google-genai.d.mts

@@ -1,2 +1,2 @@
-import { a as convertGoogleGenAIError, i as GoogleGenAIStreamedMessage, n as GoogleGenAIGenerationKwargs, o as messagesToGoogleGenAIContents, r as GoogleGenAIOptions, t as GoogleGenAIChatProvider } from "../google-genai-xKK8lI_R.mjs";
+import { a as convertGoogleGenAIError, i as GoogleGenAIStreamedMessage, n as GoogleGenAIGenerationKwargs, o as messagesToGoogleGenAIContents, r as GoogleGenAIOptions, t as GoogleGenAIChatProvider } from "../google-genai-CSE4ZD3A.mjs";
 export { GoogleGenAIChatProvider, GoogleGenAIGenerationKwargs, GoogleGenAIOptions, GoogleGenAIStreamedMessage, convertGoogleGenAIError, messagesToGoogleGenAIContents };

package/dist/providers/google-genai.mjs

@@ -1,5 +1,5 @@
-import { a as APITimeoutError, o as ChatProviderError, s as normalizeAPIStatusError, t as APIConnectionError } from "../errors-WFxxzL1B.mjs";
-import { a as getGoogleGenAIModelCapability, n as requireProviderApiKey, r as resolveAuthBackedClient } from "../request-auth-BMXt8jRu.mjs";
+import { l as ChatProviderError, n as extractCacheUsage, t as convertProviderError, u as normalizeAPIStatusError } from "../provider-common-CaxKVTTJ.mjs";
+import { a as resolveAuthBackedClient, i as requireProviderApiKey, n as BaseChatProvider, s as getGoogleGenAIModelCapability, t as BaseStreamedMessage } from "../base-streamed-message-Kok6Drxr.mjs";
 import { ApiError, GoogleGenAI } from "@google/genai";
 //#region src/providers/google-genai.ts
 /**
@@ -269,30 +269,19 @@ function messagesToGoogleGenAIContents(messages) {
 	}
 	return contents;
 }
-var GoogleGenAIStreamedMessage = class {
-	_id = null;
-	_usage = null;
-	_finishReason = null;
-	_rawFinishReason = null;
-	_iter;
+var GoogleGenAIStreamedMessage = class extends BaseStreamedMessage {
+	_response;
+	_isStream;
+	_signal;
 	constructor(response, isStream, signal) {
-		if (isStream) this._iter = this._convertStreamResponse(response, signal);
-		else this._iter = this._convertNonStreamResponse(response, signal);
+		super();
+		this._response = response;
+		this._isStream = isStream;
+		this._signal = signal;
 	}
-	get id() {
-		return this._id;
-	}
-	get usage() {
-		return this._usage;
-	}
-	get finishReason() {
-		return this._finishReason;
-	}
-	get rawFinishReason() {
-		return this._rawFinishReason;
-	}
-	async *[Symbol.asyncIterator]() {
-		yield* this._iter;
+	_buildIter() {
+		if (this._isStream) return this._convertStreamResponse(this._response, this._signal);
+		return this._convertNonStreamResponse(this._response, this._signal);
 	}
 	_captureFinishReason(response) {
 		const candidates = response["candidates"];
@@ -348,12 +337,8 @@ var GoogleGenAIStreamedMessage = class {
 		if (usageMetadata) {
 			const promptTokenCount = typeof usageMetadata["promptTokenCount"] === "number" ? usageMetadata["promptTokenCount"] : 0;
 			const cachedContentTokenCount = typeof usageMetadata["cachedContentTokenCount"] === "number" ? usageMetadata["cachedContentTokenCount"] : 0;
-			this._usage = {
-				inputOther: Math.max(promptTokenCount - cachedContentTokenCount, 0),
-				output: usageMetadata["candidatesTokenCount"] ?? 0,
-				inputCacheRead: cachedContentTokenCount,
-				inputCacheCreation: 0
-			};
+			const output = usageMetadata["candidatesTokenCount"] ?? 0;
+			this._usage = extractCacheUsage(promptTokenCount, cachedContentTokenCount, output);
 		}
 	}
 	/** Extract response ID from a response chunk. */
@@ -391,59 +376,44 @@ var GoogleGenAIStreamedMessage = class {
 		}
 	}
 };
-const NETWORK_RE = /network|connection|connect|disconnect|fetch failed/i;
-const TIMEOUT_RE = /timed?\s*out|timeout|deadline/i;
 /**
 * Convert a Google GenAI SDK error (or raw Error) to a kosong `ChatProviderError`.
 */
 function convertGoogleGenAIError(error) {
 	if (error instanceof ApiError) return normalizeAPIStatusError(error.status, error.message);
-	if (error instanceof Error) {
-		const msg = error.message;
-		if (TIMEOUT_RE.test(msg)) return new APITimeoutError(msg);
-		if (NETWORK_RE.test(msg) || error instanceof TypeError && msg.includes("fetch")) return new APIConnectionError(msg);
-		const statusCode = error.code;
-		if (typeof statusCode === "number") return normalizeAPIStatusError(statusCode, msg);
-		return new ChatProviderError(`GoogleGenAI error: ${msg}`);
-	}
-	return new ChatProviderError(`GoogleGenAI error: ${String(error)}`);
+	const statusCode = error.code;
+	if (error instanceof Error && typeof statusCode === "number") return normalizeAPIStatusError(statusCode, error.message);
+	return convertProviderError(error, {
+		extraNetworkMatchers: [/^fetch failed$/i],
+		extraTypeErrorMatch: "fetch"
+	});
 }
-var GoogleGenAIChatProvider = class {
+var GoogleGenAIChatProvider = class GoogleGenAIChatProvider extends BaseChatProvider {
 	name = "google_genai";
-	_model;
-	_client;
-	_generationKwargs;
-	_vertexai;
 	_stream;
-	_apiKey;
+	_vertexai;
 	_project;
 	_location;
-	_clientFactory;
 	constructor(options) {
-		this._model = options.model;
-		this._vertexai = options.vertexai ?? false;
-		this._stream = options.stream ?? true;
-		this._generationKwargs = {};
 		const apiKey = options.apiKey ?? process.env["GOOGLE_API_KEY"];
-		this._apiKey = apiKey === void 0 || apiKey.length === 0 ? void 0 : apiKey;
+		const apiKeyResolved = apiKey === void 0 || apiKey.length === 0 ? void 0 : apiKey;
+		const client = (options.vertexai ?? false) || apiKeyResolved !== void 0 ? GoogleGenAIChatProvider.buildClient(apiKeyResolved, options.vertexai ?? false, options.project, options.location) : void 0;
+		super(options.model, {}, apiKeyResolved, "", void 0, client, options.clientFactory);
+		this._stream = options.stream ?? true;
+		this._vertexai = options.vertexai ?? false;
 		this._project = options.project;
 		this._location = options.location;
-		this._clientFactory = options.clientFactory;
-		this._client = this._vertexai || this._apiKey !== void 0 ? this._buildClient(this._apiKey) : void 0;
 	}
-	_buildClient(apiKey) {
+	static buildClient(apiKey, vertexai, project, location) {
 		return new GoogleGenAI({
 			apiKey,
-			...this._vertexai ? {
+			...vertexai ? {
 				vertexai: true,
-				project: this._project,
-				location: this._location
+				project,
+				location
 			} : {}
 		});
 	}
-	get modelName() {
-		return this._model;
-	}
 	get thinkingEffort() {
 		const thinkingConfig = this._generationKwargs.thinking_config;
 		if (thinkingConfig === void 0) return null;
@@ -493,15 +463,26 @@ var GoogleGenAIChatProvider = class {
 			throw convertGoogleGenAIError(error);
 		}
 	}
+	/**
+	* Override the base auth-resolution path to preserve the Vertex AI
+	* short-circuit: Vertex uses service credentials, not request-scoped keys
+	* or headers, so `requireProviderApiKey` must not be enforced there.
+	*/
 	_createClient(auth) {
 		return resolveAuthBackedClient({
 			cachedClient: this._client,
 			clientFactory: this._clientFactory
 		}, auth, (a) => {
-			if (this._vertexai) return this._buildClient(this._apiKey);
-			return this._buildClient(requireProviderApiKey("GoogleGenAIChatProvider", a, this._apiKey));
+			if (this._vertexai) return GoogleGenAIChatProvider.buildClient(this._apiKey, this._vertexai, this._project, this._location);
+			return this.createRawClient({
+				apiKey: requireProviderApiKey("GoogleGenAIChatProvider", a, this._apiKey),
+				headers: void 0
+			}, void 0);
 		});
 	}
+	createRawClient(auth, _defaultHeaders) {
+		return GoogleGenAIChatProvider.buildClient(auth.apiKey, this._vertexai, this._project, this._location);
+	}
 	withThinking(effort) {
 		const thinkingConfig = { include_thoughts: true };
 		if (this._model.includes("gemini-3")) switch (effort) {
@@ -543,19 +524,6 @@ var GoogleGenAIChatProvider = class {
 		}
 		return this.withGenerationKwargs({ thinking_config: thinkingConfig });
 	}
-	withGenerationKwargs(kwargs) {
-		const clone = this._clone();
-		clone._generationKwargs = {
-			...clone._generationKwargs,
-			...kwargs
-		};
-		return clone;
-	}
-	_clone() {
-		const clone = Object.assign(Object.create(Object.getPrototypeOf(this)), this);
-		clone._generationKwargs = { ...this._generationKwargs };
-		return clone;
-	}
 };
 //#endregion
 export { GoogleGenAIChatProvider, GoogleGenAIStreamedMessage, convertGoogleGenAIError, messagesToGoogleGenAIContents };

package/dist/providers/openai-common.d.mts

@@ -1,2 +1,2 @@
-import { a as convertContentPart, c as extractUsage, d as reasoningEffortToThinkingEffort, f as thinkingEffortToReasoningEffort, i as ToolMessageConversion, l as isFunctionToolCall, n as OpenAIContentPart, o as convertOpenAIError, p as toolToOpenAI, r as OpenAIToolParam, s as convertToolMessageContent, t as FunctionToolCallShape, u as normalizeOpenAIFinishReason } from "../openai-common-DwkxUSyI.mjs";
+import { a as convertContentPart, c as extractUsage, d as reasoningEffortToThinkingEffort, f as thinkingEffortToReasoningEffort, i as ToolMessageConversion, l as isFunctionToolCall, n as OpenAIContentPart, o as convertOpenAIError, p as toolToOpenAI, r as OpenAIToolParam, s as convertToolMessageContent, t as FunctionToolCallShape, u as normalizeOpenAIFinishReason } from "../openai-common-7D2LXUK2.mjs";
 export { FunctionToolCallShape, OpenAIContentPart, OpenAIToolParam, ToolMessageConversion, convertContentPart, convertOpenAIError, convertToolMessageContent, extractUsage, isFunctionToolCall, normalizeOpenAIFinishReason, reasoningEffortToThinkingEffort, thinkingEffortToReasoningEffort, toolToOpenAI };

package/dist/providers/openai-common.mjs

@@ -1,2 +1,2 @@
-import { a as isFunctionToolCall, c as thinkingEffortToReasoningEffort, i as extractUsage, l as toolToOpenAI, n as convertOpenAIError, o as normalizeOpenAIFinishReason, r as convertToolMessageContent, s as reasoningEffortToThinkingEffort, t as convertContentPart } from "../openai-common-Dl42y_vn.mjs";
+import { a as isFunctionToolCall, c as thinkingEffortToReasoningEffort, i as extractUsage, l as toolToOpenAI, n as convertOpenAIError, o as normalizeOpenAIFinishReason, r as convertToolMessageContent, s as reasoningEffortToThinkingEffort, t as convertContentPart } from "../openai-common-B7Ex3mNh.mjs";
 export { convertContentPart, convertOpenAIError, convertToolMessageContent, extractUsage, isFunctionToolCall, normalizeOpenAIFinishReason, reasoningEffortToThinkingEffort, thinkingEffortToReasoningEffort, toolToOpenAI };

package/dist/providers/openai-responses.d.mts

@@ -1,2 +1,2 @@
-import { i as OpenAIResponsesStreamedMessage, n as OpenAIResponsesGenerationKwargs, r as OpenAIResponsesOptions, t as OpenAIResponsesChatProvider } from "../openai-responses-DZ9mQ5RA.mjs";
+import { i as OpenAIResponsesStreamedMessage, n as OpenAIResponsesGenerationKwargs, r as OpenAIResponsesOptions, t as OpenAIResponsesChatProvider } from "../openai-responses-Bj6eg-S0.mjs";
 export { OpenAIResponsesChatProvider, OpenAIResponsesGenerationKwargs, OpenAIResponsesOptions, OpenAIResponsesStreamedMessage };