@warlock.js/ai-google 5.31.0 → 5.33.0

package/CHANGELOG.md

@@ -4,6 +4,26 @@ All notable changes to `@warlock.js/ai-google` are documented in this file.
 
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/). `@warlock.js/*` packages are released in lockstep — every package shares the same version number, so a version below may list only the changes that affected this package.
 
+## 5.33.0 - 2026-10-19
+
+### Changed
+
+- Lockstep release; package APIs are unchanged.
+
+## 5.32.0 - 2026-10-18
+
+### Changed
+
+- Pin the package development compiler to TypeScript 7.0.2 and use explicit Node types and source roots with supported module resolution.
+
+### Fixed
+
+- Gemini Developer API token counting sends the full `generateContentRequest`, including system instructions, tools and generation settings, through the existing SDK transport with cancellation and error mapping. Vertex counting is unchanged; unsupported cached-content counts carry the typed signal used by `countInputTokens` to estimate.
+
+### Added
+
+- Preserve raw candidate grounding metadata and normalize unique nonempty search queries in completions and streams. Export opt-in Gemini 3+ request prices as `GOOGLE_SERVER_TOOL_PRICING`; document older models’ per-prompt billing.
+
 ## 5.31.0 - 2026-10-18
 
 ### Fixed

package/cjs/index.cjs

@@ -499,7 +499,7 @@ const DEFAULT_RESPONSE_MODALITIES = ["TEXT", "IMAGE"];
 * or image generation being "stopped" for safety, prohibited content,
 * recitation, blocklist, or SPII.
 */
-const FILTERED_FINISH_REASONS = new Set([
+const FILTERED_FINISH_REASONS = /* @__PURE__ */ new Set([
 	"SAFETY",
 	"IMAGE_SAFETY",
 	"PROHIBITED_CONTENT",
@@ -832,10 +832,42 @@ const GOOGLE_MODEL_PRICING = { "gemini-3.5-flash-lite": {
 function googlePricingFor(modelName) {
 	return GOOGLE_MODEL_PRICING[modelName];
 }
+/** USD per nonempty unique query for Gemini 3+, before free monthly allowances.
+* Gemini 2.5 and older bill per grounded prompt; do not use this rate for them.
+* https://ai.google.dev/gemini-api/docs/google-search#pricing — verified 2026-10-18. */
+const GOOGLE_SERVER_TOOL_PRICING = { webSearch: .014 };
+
+//#endregion
+//#region ../../packages/ai-google/src/utils/grounding.ts
+/** Candidate snapshots are replaced, not added: streamed metadata may repeat.
+* Installed @google/genai 2.19.0 GroundingMetadata.webSearchQueries; REST schema:
+* https://ai.google.dev/api/generate-content#GroundingMetadata
+* Billing ignores empty queries and counts unique queries for Gemini 3+:
+* https://ai.google.dev/gemini-api/docs/google-search#pricing
+* Gemini 2.5 and older bill per grounded prompt; this counter is NOT that unit. */
+function applyGrounding(usage, snapshots, response) {
+	for (const [position, candidate] of (response.candidates ?? []).entries()) if (candidate.groundingMetadata) snapshots.set(candidate.index ?? position, candidate.groundingMetadata);
+	const reported = [...snapshots.values()].filter((metadata) => metadata.webSearchQueries !== void 0);
+	if (reported.length > 0) {
+		const queries = reported.flatMap((metadata) => metadata.webSearchQueries ?? []);
+		usage.serverToolUse = {
+			...usage.serverToolUse,
+			webSearchRequests: new Set(queries.filter((query) => query !== "")).size,
+			googleWebSearchQueries: queries.length
+		};
+	}
+}
+/** Retain chunks, supports, search entry point and queries without flattening them. */
+function groundingMetadata(snapshots) {
+	return snapshots.size ? { google: { groundingMetadata: [...snapshots].map(([index, metadata]) => ({
+		index,
+		...metadata
+	})) } } : void 0;
+}
 
 //#endregion
 //#region ../../packages/ai-google/src/provider-options.ts
-const RESERVED_FIELDS = new Set([
+const RESERVED_FIELDS = /* @__PURE__ */ new Set([
 	"systemInstruction",
 	"tools",
 	"abortSignal",
@@ -1000,12 +1032,17 @@ var GoogleModel = class {
 		}
 		const toolCalls = this.extractToolCalls(response);
 		const finishReason = toolCalls ? "tool_calls" : mapFinishReason(response.candidates?.[0]?.finishReason);
-		const usage = (0, _warlock_js_ai.attachCost)(this.extractUsage(response), this.pricing);
+		const usage = this.extractUsage(response);
+		const grounding = /* @__PURE__ */ new Map();
+		applyGrounding(usage, grounding, response);
+		const providerMetadata = groundingMetadata(grounding);
+		(0, _warlock_js_ai.attachCost)(usage, this.pricing);
 		this.logger.debug(LOG_MODULE, "response", "generateContent call succeeded", {
 			finishReason,
 			usage
 		});
 		return {
+			...providerMetadata ? { providerMetadata } : {},
 			content: response.text ?? "",
 			finishReason,
 			usage,
@@ -1016,10 +1053,9 @@ var GoogleModel = class {
 	async countTokens(messages, options) {
 		const { systemInstruction, contents } = toGoogleContents(messages);
 		const config = this.buildConfig(systemInstruction, options);
-		if (!this.ai.vertexai && (config.systemInstruction !== void 0 || config.tools !== void 0)) throw new _warlock_js_ai.InvalidRequestError("The Google SDK countTokens endpoint in Developer API mode cannot count system instructions or tools. Use Vertex mode for this preflight, or SDK.count() for an offline estimate.");
-		if (config.cachedContent !== void 0) throw new _warlock_js_ai.InvalidRequestError("The Google SDK countTokens endpoint cannot count a cachedContent reference.");
+		if (config.cachedContent !== void 0) throw new _warlock_js_ai.InvalidRequestError("The Google SDK countTokens endpoint cannot count a cachedContent reference.", { reason: "count-unsupported" });
 		try {
-			const response = await this.ai.models.countTokens({
+			const response = this.ai.vertexai ? await this.ai.models.countTokens({
 				model: this.name,
 				contents,
 				config: {
@@ -1028,7 +1064,7 @@ var GoogleModel = class {
 					...config.abortSignal ? { abortSignal: config.abortSignal } : {},
 					...config.httpOptions ? { httpOptions: config.httpOptions } : {}
 				}
-			});
+			}) : await this.countDeveloperTokens(contents, config);
 			if (response.totalTokens === void 0) throw new _warlock_js_ai.InvalidRequestError("Google countTokens returned no totalTokens.");
 			return response.totalTokens;
 		} catch (thrown) {
@@ -1036,6 +1072,42 @@ var GoogleModel = class {
 		}
 	}
 	/**
+	* REST counts the full generation request, unlike the typed Developer API
+	* converter in @google/genai 2.19.0. See https://ai.google.dev/api/tokens.
+	* extraBody keeps the SDK's auth, fetch, base URL, retries and cancellation.
+	*/
+	countDeveloperTokens(contents, config) {
+		config.abortSignal?.throwIfAborted();
+		const { systemInstruction, serviceTier, tools, toolConfig, safetySettings, cachedContent, abortSignal, httpOptions, ...generationConfig } = config;
+		delete generationConfig.automaticFunctionCalling;
+		const model = this.name.startsWith("models/") ? this.name : `models/${this.name}`;
+		return this.ai.models.countTokens({
+			model: this.name,
+			contents,
+			config: {
+				abortSignal,
+				httpOptions: {
+					...httpOptions,
+					extraBody: {
+						...httpOptions?.extraBody,
+						contents: void 0,
+						generateContentRequest: {
+							model,
+							contents,
+							...serviceTier !== void 0 ? { serviceTier } : {},
+							...systemInstruction !== void 0 ? { systemInstruction: { parts: [{ text: systemInstruction }] } } : {},
+							...tools !== void 0 ? { tools } : {},
+							...toolConfig !== void 0 ? { toolConfig } : {},
+							...safetySettings !== void 0 ? { safetySettings } : {},
+							...cachedContent !== void 0 ? { cachedContent } : {},
+							...Object.keys(generationConfig).length > 0 ? { generationConfig } : {}
+						}
+					}
+				}
+			}
+		});
+	}
+	/**
 	* Incremental streaming completion via `generateContentStream`.
 	* Yields neutral `ModelStreamChunk`s — `delta` for text, `tool-call`
 	* per function call (Gemini emits a fully-formed call, not partial
@@ -1072,8 +1144,10 @@ var GoogleModel = class {
 			total: 0
 		};
 		const want = (0, _warlock_js_ai.resolveStreamChunks)(options);
+		const grounding = /* @__PURE__ */ new Map();
 		try {
 			for await (const chunk of iterable) {
+				applyGrounding(usage, grounding, chunk);
 				if (want.reasoning) {
 					for (const part of chunk.candidates?.[0]?.content?.parts ?? []) if (part.thought === true && typeof part.text === "string" && part.text !== "") yield {
 						type: "reasoning",
@@ -1116,10 +1190,12 @@ var GoogleModel = class {
 			finishReason,
 			usage
 		});
+		const providerMetadata = groundingMetadata(grounding);
 		yield {
 			type: "done",
 			finishReason,
-			usage
+			usage,
+			...providerMetadata ? { providerMetadata } : {}
 		};
 	}
 	/**
@@ -1393,6 +1469,7 @@ var GoogleSDK = class {
 
 //#endregion
 exports.GOOGLE_MODEL_PRICING = GOOGLE_MODEL_PRICING;
+exports.GOOGLE_SERVER_TOOL_PRICING = GOOGLE_SERVER_TOOL_PRICING;
 exports.GeminiImageModel = GeminiImageModel;
 exports.GoogleImageModel = GoogleImageModel;
 exports.GoogleSDK = GoogleSDK;