@f5-sales-demo/pi-ai 21.15.1 → 21.17.0

package/CHANGELOG.md

@@ -6,6 +6,9 @@
 
 ### Changed
 
+- Replaced the bundled Vertex Gemini 3.7 Flash model and provider default with GA Gemini 3.8 Flash,
+  retaining HIGH thinking by default and the strict Vertex tool-call contract
+  ([#3751](https://github.com/f5-sales-demo/xcsh/issues/3751)).
 - Replaced Vertex Gemini 3.6 Flash with GA Gemini 3.7 Flash and made full Vertex Gemini Flash and Pro models default to HIGH thinking while preserving supported explicit effort overrides ([#3460](https://github.com/f5-sales-demo/xcsh/issues/3460)).
 
 ### Fixed

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@f5-sales-demo/pi-ai",
-	"version": "21.15.1",
+	"version": "21.17.0",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"homepage": "https://github.com/f5-sales-demo/xcsh",
 	"author": "Can Boluk",
@@ -46,7 +46,7 @@
 		"@anthropic-ai/sdk": "^0.115",
 		"@aws-sdk/client-bedrock-runtime": "^3",
 		"@bufbuild/protobuf": "^2.11",
-		"@f5-sales-demo/pi-utils": "21.15.1",
+		"@f5-sales-demo/pi-utils": "21.17.0",
 		"@google/genai": "^2.15",
 		"@sinclair/typebox": "^0.34",
 		"@smithy/node-http-handler": "^4.4",

package/src/models.json

@@ -10316,9 +10316,9 @@
 		}
 	},
 	"google-vertex": {
-		"gemini-3.7-flash": {
-			"id": "gemini-3.7-flash",
-			"name": "Gemini 3.7 Flash",
+		"gemini-3.8-flash": {
+			"id": "gemini-3.8-flash",
+			"name": "Gemini 3.8 Flash",
 			"api": "google-vertex",
 			"provider": "google-vertex",
 			"baseUrl": "https://{location}-aiplatform.googleapis.com",
@@ -10338,10 +10338,6 @@
 			"thinking": {
 				"mode": "google-level",
 				"supportedLevels": [
-					{
-						"effort": "minimal",
-						"description": "Very brief reasoning"
-					},
 					{
 						"effort": "low",
 						"description": "Light reasoning"

package/src/provider-models/descriptors.ts

@@ -276,7 +276,7 @@ export const DEFAULT_MODEL_PER_PROVIDER: Record<KnownProvider, string> = {
 	"amazon-bedrock": "us.anthropic.claude-opus-4-6-v1",
 	"google-antigravity": "gemini-3-pro-high",
 	"google-gemini-cli": "gemini-2.5-pro",
-	"google-vertex": "gemini-3-pro-preview",
+	"google-vertex": "gemini-3.8-flash",
 	minimax: "MiniMax-M2.5",
 	"minimax-code": "MiniMax-M2.5",
 	"minimax-code-cn": "MiniMax-M2.5",

package/src/providers/google-shared.ts

@@ -65,7 +65,7 @@ function resolveThoughtSignature(isSameProviderAndModel: boolean, signature: str
  * Models with strict Google tool round trips require explicit matching call/response IDs.
  */
 export function requiresToolCallId(modelId: string): boolean {
-	return modelId.startsWith("claude-") || modelId === "gemini-3.7-flash";
+	return modelId.startsWith("claude-") || modelId === "gemini-3.8-flash";
 }
 
 function getGeminiMajorVersion(modelId: string): number | undefined {
@@ -177,7 +177,7 @@ export function convertMessages<T extends GoogleApiType>(model: Model<T>, contex
 							...(requiresToolCallId(model.id) ? { id: block.id } : {}),
 						},
 					};
-					if (model.provider === "google-vertex" && model.id !== "gemini-3.7-flash" && part?.functionCall?.id) {
+					if (model.provider === "google-vertex" && model.id !== "gemini-3.8-flash" && part?.functionCall?.id) {
 						delete part.functionCall.id; // Vertex AI does not support 'id' in functionCall
 					}
 					if (effectiveSignature) {
@@ -230,7 +230,7 @@ export function convertMessages<T extends GoogleApiType>(model: Model<T>, contex
 
 			if (
 				model.provider === "google-vertex" &&
-				model.id !== "gemini-3.7-flash" &&
+				model.id !== "gemini-3.8-flash" &&
 				functionResponsePart.functionResponse?.id
 			) {
 				delete functionResponsePart.functionResponse.id; // Vertex AI does not support 'id' in functionResponse

package/src/providers/google-vertex.ts

@@ -450,7 +450,7 @@ export async function resolveGoogleVertexProject(
 }
 
 export function resolveGoogleVertexLocation(options?: GoogleVertexOptions): string {
-	// Gemini 3.7 Flash Corporate is served through the global endpoint only.
+	// Gemini 3.8 Flash Corporate is served through the global endpoint only.
 	// Ignore inherited provider/environment locations so a request cannot drift.
 	void options;
 	return "global";
@@ -484,28 +484,31 @@ export function buildGoogleVertexParams(
 	options: GoogleVertexOptions = {},
 ): GenerateContentParameters {
 	const contents = convertMessages(model, context);
-	const isGemini37Flash = model.id === "gemini-3.7-flash";
+	const isGemini38Flash = model.id === "gemini-3.8-flash";
+	if (isGemini38Flash && options.thinking?.level === "MINIMAL") {
+		throw new Error("MINIMAL thinking is not supported by Gemini 3.8 Flash");
+	}
 
 	const generationConfig: GoogleVertexSamplingConfig = {};
-	if (!isGemini37Flash && options.temperature !== undefined) {
+	if (!isGemini38Flash && options.temperature !== undefined) {
 		generationConfig.temperature = options.temperature;
 	}
 	if (options.maxTokens !== undefined) {
 		generationConfig.maxOutputTokens = options.maxTokens;
 	}
-	if (!isGemini37Flash && options.topP !== undefined) {
+	if (!isGemini38Flash && options.topP !== undefined) {
 		generationConfig.topP = options.topP;
 	}
-	if (!isGemini37Flash && options.topK !== undefined) {
+	if (!isGemini38Flash && options.topK !== undefined) {
 		generationConfig.topK = options.topK;
 	}
-	if (!isGemini37Flash && options.minP !== undefined) {
+	if (!isGemini38Flash && options.minP !== undefined) {
 		generationConfig.minP = options.minP;
 	}
-	if (!isGemini37Flash && options.presencePenalty !== undefined) {
+	if (!isGemini38Flash && options.presencePenalty !== undefined) {
 		generationConfig.presencePenalty = options.presencePenalty;
 	}
-	if (!isGemini37Flash && options.repetitionPenalty !== undefined) {
+	if (!isGemini38Flash && options.repetitionPenalty !== undefined) {
 		generationConfig.repetitionPenalty = options.repetitionPenalty;
 	}
 
@@ -527,11 +530,11 @@ export function buildGoogleVertexParams(
 		config.toolConfig = undefined;
 	}
 
-	if (model.reasoning && (options.thinking?.enabled || isGemini37Flash)) {
+	if (model.reasoning && (options.thinking?.enabled || isGemini38Flash)) {
 		const cfg: ThinkingConfig = { includeThoughts: true };
 		if (options.thinking?.level !== undefined) {
 			cfg.thinkingLevel = THINKING_LEVEL_MAP[options.thinking.level];
-		} else if (isGemini37Flash) {
+		} else if (isGemini38Flash) {
 			cfg.thinkingLevel = ThinkingLevel.HIGH;
 		} else if (options.thinking?.budgetTokens !== undefined) {
 			cfg.thinkingBudget = options.thinking.budgetTokens;

package/src/providers/openai-codex-responses.ts

@@ -63,7 +63,12 @@ import {
 	transformRequestBody,
 } from "./openai-codex/request-transformer";
 import { parseCodexError } from "./openai-codex/response-handler";
-import { encodeTextSignatureV1, mapOpenAIResponsesStopReason, parseTextSignature } from "./openai-responses-shared";
+import {
+	encodeTextSignatureV1,
+	mapOpenAIResponsesStopReason,
+	parseTextSignature,
+	resolveAssistantMessagePhase,
+} from "./openai-responses-shared";
 import { transformMessages } from "./transform-messages";
 
 export interface OpenAICodexResponsesOptions extends StreamOptions {
@@ -747,11 +752,21 @@ function handleCodexStreamEvent(args: {
 		runtime.currentBlock = createOutputBlockForItem(item);
 		if (!runtime.currentBlock) return firstTokenTime;
 		output.content.push(runtime.currentBlock);
-		stream.push({
-			type: getOutputBlockStartEventType(runtime.currentBlock),
-			contentIndex: blockIndex(),
-			partial: output,
-		});
+		if (runtime.currentBlock.type === "text") {
+			stream.push({
+				type: "text_start",
+				contentIndex: blockIndex(),
+				phase: runtime.currentBlock.phase ?? "final_answer",
+				partial: output,
+			});
+		} else {
+			const type = getOutputBlockStartEventType(runtime.currentBlock);
+			if (type === "thinking_start") {
+				stream.push({ type, contentIndex: blockIndex(), partial: output });
+			} else {
+				stream.push({ type: "toolcall_start", contentIndex: blockIndex(), partial: output });
+			}
+		}
 		return firstTokenTime;
 	}
 
@@ -838,7 +853,7 @@ function createOutputBlockForItem(item: CodexEventItem): CodexOutputBlock | null
 		return { type: "thinking", thinking: "" };
 	}
 	if (item.type === "message") {
-		return { type: "text", text: "" };
+		return { type: "text", text: "", phase: resolveAssistantMessagePhase(item.phase) };
 	}
 	if (item.type === "function_call") {
 		return {
@@ -986,12 +1001,14 @@ function handleOutputItemDone(
 		runtime.currentBlock.text = item.content
 			.map(content => (content.type === "output_text" ? content.text : content.refusal))
 			.join("");
-		const phase = item.phase === "commentary" || item.phase === "final_answer" ? item.phase : undefined;
+		const phase = resolveAssistantMessagePhase(item.phase);
+		runtime.currentBlock.phase = phase;
 		runtime.currentBlock.textSignature = encodeTextSignatureV1(item.id, phase);
 		stream.push({
 			type: "text_end",
 			contentIndex: blockIndex(),
 			content: runtime.currentBlock.text,
+			phase,
 			partial: output,
 		});
 		runtime.currentBlock = null;
@@ -2175,7 +2192,7 @@ function convertMessages(model: Model<"openai-codex-responses">, context: Contex
 						content: [{ type: "output_text", text: textBlock.text.toWellFormed(), annotations: [] }],
 						status: "completed",
 						id: msgId,
-						phase: parsedSignature?.phase,
+						phase: parsedSignature?.phase ?? textBlock.phase,
 					} satisfies ResponseOutputMessage);
 					continue;
 				}

package/src/providers/openai-responses-shared.ts

@@ -14,6 +14,7 @@ import { calculateCost } from "../models";
 import type {
 	Api,
 	AssistantMessage,
+	AssistantMessagePhase,
 	ImageContent,
 	Model,
 	StopReason,
@@ -33,6 +34,10 @@ export function encodeTextSignatureV1(id: string, phase?: TextSignatureV1["phase
 	return JSON.stringify(payload);
 }
 
+export function resolveAssistantMessagePhase(phase: unknown): AssistantMessagePhase {
+	return phase === "commentary" ? "commentary" : "final_answer";
+}
+
 export function parseTextSignature(
 	signature: string | undefined,
 ): { id: string; phase?: TextSignatureV1["phase"] } | undefined {
@@ -152,7 +157,7 @@ export function convertResponsesAssistantMessage<TApi extends Api>(
 				content: [{ type: "output_text", text: block.text.toWellFormed(), annotations: [] }],
 				status: "completed",
 				id: msgId,
-				phase: parsedSignature?.phase,
+				phase: parsedSignature?.phase ?? block.phase,
 			} satisfies ResponseOutputMessage);
 			continue;
 		}
@@ -262,9 +267,14 @@ export async function processResponsesStream<TApi extends Api>(
 				stream.push({ type: "thinking_start", contentIndex: blockIndex(), partial: output });
 			} else if (item.type === "message") {
 				currentItem = item;
-				currentBlock = { type: "text", text: "" };
+				currentBlock = { type: "text", text: "", phase: resolveAssistantMessagePhase(item.phase) };
 				output.content.push(currentBlock);
-				stream.push({ type: "text_start", contentIndex: blockIndex(), partial: output });
+				stream.push({
+					type: "text_start",
+					contentIndex: blockIndex(),
+					phase: currentBlock.phase ?? "final_answer",
+					partial: output,
+				});
 			} else if (item.type === "function_call") {
 				currentItem = item;
 				currentBlock = {
@@ -380,11 +390,13 @@ export async function processResponsesStream<TApi extends Api>(
 				currentBlock.text = item.content
 					.map(part => (part.type === "output_text" ? (part.text ?? "") : (part.refusal ?? "")))
 					.join("");
-				currentBlock.textSignature = encodeTextSignatureV1(item.id, item.phase ?? undefined);
+				currentBlock.phase = resolveAssistantMessagePhase(item.phase);
+				currentBlock.textSignature = encodeTextSignatureV1(item.id, currentBlock.phase);
 				stream.push({
 					type: "text_end",
 					contentIndex: blockIndex(),
 					content: currentBlock.text,
+					phase: currentBlock.phase ?? "final_answer",
 					partial: output,
 				});
 				currentBlock = null;

package/src/types.ts

@@ -258,10 +258,12 @@ export type StreamFunction<TApi extends Api> = (
 	options: OptionsForApi<TApi>,
 ) => AssistantMessageEventStream;
 
+export type AssistantMessagePhase = "commentary" | "final_answer";
+
 export interface TextSignatureV1 {
 	v: 1;
 	id: string;
-	phase?: "commentary" | "final_answer";
+	phase?: AssistantMessagePhase;
 }
 
 /**
@@ -288,6 +290,8 @@ export interface WebCitation {
 export interface TextContent {
 	type: "text";
 	text: string;
+	/** Semantic role supplied by Responses. Absence has final-answer semantics. */
+	phase?: AssistantMessagePhase;
 	textSignature?: string; // e.g., for OpenAI responses, message metadata (legacy id string or TextSignatureV1 JSON)
 	/** Structured citations for this span, when the provider ran a server-side search. */
 	citations?: WebCitation[];
@@ -468,9 +472,15 @@ export interface Context {
 
 export type AssistantMessageEvent =
 	| { type: "start"; contentIndex?: undefined; partial: AssistantMessage }
-	| { type: "text_start"; contentIndex: number; partial: AssistantMessage }
+	| { type: "text_start"; contentIndex: number; phase?: AssistantMessagePhase; partial: AssistantMessage }
 	| { type: "text_delta"; contentIndex: number; delta: string; partial: AssistantMessage }
-	| { type: "text_end"; contentIndex: number; content: string; partial: AssistantMessage }
+	| {
+			type: "text_end";
+			contentIndex: number;
+			content: string;
+			phase?: AssistantMessagePhase;
+			partial: AssistantMessage;
+	  }
 	| { type: "thinking_start"; contentIndex: number; partial: AssistantMessage }
 	| { type: "thinking_delta"; contentIndex: number; delta: string; partial: AssistantMessage }
 	| { type: "thinking_end"; contentIndex: number; content: string; partial: AssistantMessage }