@f5-sales-demo/pi-ai 21.29.4 → 21.30.1

package/CHANGELOG.md

@@ -4,6 +4,12 @@
 
 ## Unreleased
 
+### Added
+
+- Added native Anthropic Claude Fable 5 and 5.1 catalog and transport support, including always-on
+  adaptive thinking, accurate limits and pricing, strict-tool compatibility, and permanent
+  entitlement/client-version error handling.
+
 ### Changed
 
 - Replaced the bundled Vertex Gemini 3.7 Flash model and provider default with GA Gemini 3.8 Flash,

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@f5-sales-demo/pi-ai",
-	"version": "21.29.4",
+	"version": "21.30.1",
 	"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": "0.123.0",
 		"@aws-sdk/client-bedrock-runtime": "3.1126.0",
 		"@bufbuild/protobuf": "2.14.1",
-		"@f5-sales-demo/pi-utils": "21.29.4",
+		"@f5-sales-demo/pi-utils": "21.30.1",
 		"@google/genai": "2.21.0",
 		"@sinclair/typebox": "0.34.52",
 		"@smithy/node-http-handler": "4.12.1",

package/src/model-thinking.ts

@@ -120,7 +120,12 @@ const ANTHROPIC_ADAPTIVE_EFFORTS: readonly Effort[] = [
  * Adding a version here without a probe is the defect this set exists to
  * prevent: an unprobed model must fail closed to DEFAULT_REASONING_EFFORTS.
  */
-const ANTHROPIC_EXTENDED_EFFORT_VERSIONS: ReadonlySet<string> = new Set(["5.0"]);
+const ANTHROPIC_EXTENDED_EFFORT_MODELS: ReadonlySet<string> = new Set([
+	"fable-5.0",
+	"fable-5.1",
+	"opus-5.0",
+	"sonnet-5.0",
+]);
 
 const GEMINI_3_PRO_EFFORTS: readonly Effort[] = [Effort.Low, Effort.Medium, Effort.High];
 const GEMINI_3_FLASH_EFFORTS: readonly Effort[] = [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High];
@@ -135,7 +140,7 @@ type SemVer = {
 };
 
 type GeminiKind = "pro" | "flash";
-type AnthropicKind = "opus" | "sonnet";
+type AnthropicKind = "fable" | "opus" | "sonnet";
 type OpenAIVariant = "base" | "codex" | "codex-max" | "codex-mini" | "codex-spark" | "mini" | "max" | "nano";
 
 const CODEX_GPT_5_4_PRIORITY_BY_VARIANT: Partial<Record<OpenAIVariant, number>> = {
@@ -453,6 +458,9 @@ function inferModelThinking<TApi extends Api>(model: ApiModel<TApi>): ThinkingCo
 		supportedLevels: efforts.map(effort => ({ effort, description: EFFORT_DESCRIPTIONS[effort] })),
 		defaultLevel: efforts.includes(ReasoningEffort.Medium) ? ReasoningEffort.Medium : efforts[0]!,
 	};
+	if (parsedModel.family === "anthropic" && parsedModel.kind === "fable") {
+		thinking.defaultLevel = Effort.High;
+	}
 	if (
 		model.provider === "google-vertex" &&
 		parsedModel.family === "gemini" &&
@@ -572,7 +580,9 @@ function inferAnthropicSupportedEfforts<TApi extends Api>(
 		// every future Claude the moment it lands in the catalog — the same
 		// open-ended grant that caused this bug. Unprobed versions fail closed to
 		// the conservative range; add one here only with a probe to back it.
-		return ANTHROPIC_EXTENDED_EFFORT_VERSIONS.has(`${parsedModel.version.major}.${parsedModel.version.minor}`)
+		return ANTHROPIC_EXTENDED_EFFORT_MODELS.has(
+			`${parsedModel.kind}-${parsedModel.version.major}.${parsedModel.version.minor}`,
+		)
 			? ANTHROPIC_ADAPTIVE_EFFORTS
 			: DEFAULT_REASONING_EFFORTS;
 	}
@@ -667,7 +677,7 @@ function parseGeminiModel(modelId: string): GeminiModel | null {
 }
 
 function parseAnthropicModel(modelId: string): AnthropicModel | null {
-	const match = /claude-(opus|sonnet)-(\d{1,2}(?:[.-]\d{1,2}){0,2})\b/.exec(modelId);
+	const match = /claude-(fable|opus|sonnet)-(\d{1,2}(?:[.-]\d{1,2}){0,2})\b/.exec(modelId);
 	if (!match) {
 		return null;
 	}
@@ -678,6 +688,12 @@ function parseAnthropicModel(modelId: string): AnthropicModel | null {
 	return { family: "anthropic", kind: match[1] as AnthropicKind, version };
 }
 
+/** Whether the model requires adaptive thinking on every request. */
+export function isAnthropicAlwaysThinkingModel<TApi extends Api>(model: ApiModel<TApi>): boolean {
+	const parsed = parseKnownModel(model.id);
+	return model.api === "anthropic-messages" && parsed.family === "anthropic" && parsed.kind === "fable";
+}
+
 function parseOpenAIModel(modelId: string): OpenAIModel | null {
 	const match = /gpt-(\d+(?:\.\d+){0,2})(?:-(codex-spark|codex-mini|codex-max|codex|mini|max|nano))?\b/.exec(modelId);
 	if (!match) {

package/src/models.json

@@ -3158,6 +3158,106 @@
 		}
 	},
 	"anthropic": {
+		"claude-fable-5": {
+			"id": "claude-fable-5",
+			"name": "Claude Fable 5",
+			"api": "anthropic-messages",
+			"provider": "anthropic",
+			"baseUrl": "https://api.anthropic.com",
+			"reasoning": true,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 10,
+				"output": 50,
+				"cacheRead": 1,
+				"cacheWrite": 12.5
+			},
+			"contextWindow": 1000000,
+			"maxTokens": 128000,
+			"thinking": {
+				"mode": "anthropic-adaptive",
+				"supportedLevels": [
+					{
+						"effort": "minimal",
+						"description": "Very brief reasoning"
+					},
+					{
+						"effort": "low",
+						"description": "Light reasoning"
+					},
+					{
+						"effort": "medium",
+						"description": "Balanced reasoning"
+					},
+					{
+						"effort": "high",
+						"description": "Deep reasoning"
+					},
+					{
+						"effort": "xhigh",
+						"description": "Very deep reasoning"
+					},
+					{
+						"effort": "max",
+						"description": "Maximum reasoning"
+					}
+				],
+				"defaultLevel": "high"
+			}
+		},
+		"claude-fable-5-1": {
+			"id": "claude-fable-5-1",
+			"name": "Claude Fable 5.1",
+			"api": "anthropic-messages",
+			"provider": "anthropic",
+			"baseUrl": "https://api.anthropic.com",
+			"reasoning": true,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 10,
+				"output": 50,
+				"cacheRead": 0.25,
+				"cacheWrite": 12.5
+			},
+			"contextWindow": 1000000,
+			"maxTokens": 128000,
+			"thinking": {
+				"mode": "anthropic-adaptive",
+				"supportedLevels": [
+					{
+						"effort": "minimal",
+						"description": "Very brief reasoning"
+					},
+					{
+						"effort": "low",
+						"description": "Light reasoning"
+					},
+					{
+						"effort": "medium",
+						"description": "Balanced reasoning"
+					},
+					{
+						"effort": "high",
+						"description": "Deep reasoning"
+					},
+					{
+						"effort": "xhigh",
+						"description": "Very deep reasoning"
+					},
+					{
+						"effort": "max",
+						"description": "Maximum reasoning"
+					}
+				],
+				"defaultLevel": "high"
+			}
+		},
 		"claude-3-5-sonnet-20240620": {
 			"id": "claude-3-5-sonnet-20240620",
 			"name": "Claude Sonnet 3.5",

package/src/providers/anthropic.ts

@@ -9,7 +9,12 @@ import type {
 	MessageParam,
 } from "@anthropic-ai/sdk/resources/messages";
 import { $env, abortableSleep, isEnoent } from "@f5-sales-demo/pi-utils";
-import { type AnthropicAdaptiveEffort, mapEffortToAnthropicAdaptiveEffort } from "../model-thinking";
+import {
+	type AnthropicAdaptiveEffort,
+	Effort,
+	isAnthropicAlwaysThinkingModel,
+	mapEffortToAnthropicAdaptiveEffort,
+} from "../model-thinking";
 import { calculateCost } from "../models";
 import { getEnvApiKey, OUTPUT_FALLBACK_BUFFER } from "../stream";
 import type {
@@ -186,7 +191,7 @@ function getCacheControl(
 }
 
 // Stealth mode: Mimic Claude Code headers and tool prefixing.
-export const claudeCodeVersion = "2.1.63";
+export const claudeCodeVersion = "2.1.274";
 export const claudeToolPrefix: string = "proxy_";
 export const claudeCodeSystemInstruction = "You are a Claude agent, built on Anthropic's Claude Agent SDK.";
 
@@ -624,6 +629,7 @@ function isProviderRetryableStreamEnvelopeError(error: unknown): boolean {
 export function isProviderRetryableError(error: unknown): boolean {
 	if (!(error instanceof Error)) return false;
 	const msg = error.message.toLowerCase();
+	if (isAnthropicPermanentErrorMessage(msg)) return false;
 	return (
 		/rate.?limit|too many requests|overloaded|service.?unavailable|internal_error|stream error.*received from peer|1302|timed?\s*out while waiting for the first event|timeout waiting for first/i.test(
 			msg,
@@ -633,6 +639,26 @@ export function isProviderRetryableError(error: unknown): boolean {
 	);
 }
 
+export function isAnthropicPermanentErrorMessage(message: string): boolean {
+	return /credits_required|obsolete[_ -]?client|client[_ -]?(?:is[_ -]?)?too[_ -]?old|version of (?:claude code|the client) is no longer supported/i.test(
+		message,
+	);
+}
+
+export function rewriteAnthropicPermanentError(errorMessage: string): string {
+	if (/credits_required/i.test(errorMessage)) {
+		return "Anthropic access requires credits for this model. Check the account's Claude subscription or API billing entitlement, then retry.";
+	}
+	if (
+		/obsolete[_ -]?client|client[_ -]?(?:is[_ -]?)?too[_ -]?old|version of (?:claude code|the client) is no longer supported/i.test(
+			errorMessage,
+		)
+	) {
+		return "Anthropic rejected this xcsh client version. Upgrade xcsh to the latest release, then retry.";
+	}
+	return errorMessage;
+}
+
 /**
  * Map an Anthropic `citations_delta` citation to a `WebCitation`.
  *
@@ -1114,6 +1140,9 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 			output.stopReason = activeAbortTracker.wasCallerAbort() ? "aborted" : "error";
 			output.errorMessage = firstEventTimeoutError?.message ?? (await finalizeErrorMessage(error, rawRequestDump));
 			output.errorMessage = rewriteCopilotAuthError(output.errorMessage, error, model.provider);
+			if (model.provider === "anthropic") {
+				output.errorMessage = rewriteAnthropicPermanentError(output.errorMessage);
+			}
 			output.duration = Date.now() - startTime;
 			if (firstTokenTime) output.ttft = firstTokenTime - startTime;
 			stream.push({ type: "error", reason: output.stopReason, error: output });
@@ -1250,7 +1279,7 @@ export function buildAnthropicClientOptions(args: AnthropicClientOptionsArgs): A
 		apiKey: oauthToken ? null : apiKey,
 		authToken: oauthToken ? apiKey : undefined,
 		baseURL: baseUrl,
-		maxRetries: 5,
+		maxRetries: 0,
 		dangerouslyAllowBrowser: true,
 		defaultHeaders,
 		logLevel: ANTHROPIC_SDK_LOG_LEVEL,
@@ -1529,6 +1558,12 @@ function buildParams(
 	options?: AnthropicOptions,
 ): MessageCreateParamsStreaming {
 	const { cacheControl } = getCacheControl(baseUrl, options?.cacheRetention);
+	const alwaysThinking = isAnthropicAlwaysThinkingModel(model);
+	const toolChoice =
+		alwaysThinking &&
+		(options?.toolChoice === "any" || (typeof options?.toolChoice === "object" && options.toolChoice.type === "tool"))
+			? "auto"
+			: options?.toolChoice;
 	const params: AnthropicSamplingParams = {
 		model: model.id,
 		messages: convertAnthropicMessages(context.messages, model, isOAuthToken),
@@ -1536,7 +1571,7 @@ function buildParams(
 		stream: true,
 	};
 
-	if (options?.temperature !== undefined) {
+	if (!alwaysThinking && options?.temperature !== undefined) {
 		params.temperature = options.temperature;
 	}
 	if (options?.topP !== undefined) {
@@ -1547,22 +1582,26 @@ function buildParams(
 	}
 
 	if (context.tools) {
-		const toolChoice = options?.toolChoice;
 		const forcedToolName =
 			model.provider === "anthropic" &&
 			supportsAnthropicStrictToolUse(model.id) &&
-			typeof toolChoice === "object" &&
-			toolChoice.type === "tool"
-				? toolChoice.name
+			typeof options?.toolChoice === "object" &&
+			options.toolChoice.type === "tool"
+				? options.toolChoice.name
 				: undefined;
 		params.tools = convertTools(context.tools, isOAuthToken, forcedToolName);
 	}
 
-	if (options?.thinkingEnabled && model.reasoning) {
+	if ((options?.thinkingEnabled || alwaysThinking) && model.reasoning) {
 		const mode = model.thinking?.mode;
-		const requestedEffort = options.reasoning;
+		const requestedEffort = options?.reasoning;
 		const effort =
-			options.effort ?? (requestedEffort ? mapEffortToAnthropicAdaptiveEffort(model, requestedEffort) : undefined);
+			options?.effort ??
+			(requestedEffort
+				? mapEffortToAnthropicAdaptiveEffort(model, requestedEffort)
+				: alwaysThinking
+					? mapEffortToAnthropicAdaptiveEffort(model, Effort.High)
+					: undefined);
 
 		if (mode === "anthropic-adaptive") {
 			params.thinking = { type: "adaptive" };
@@ -1572,7 +1611,7 @@ function buildParams(
 		} else {
 			params.thinking = {
 				type: "enabled",
-				budget_tokens: options.thinkingBudgetTokens || 1024,
+				budget_tokens: options?.thinkingBudgetTokens || 1024,
 			};
 			if (mode === "anthropic-budget-effort" && effort) {
 				params.output_config = { effort };
@@ -1580,7 +1619,7 @@ function buildParams(
 		}
 		// Anthropic requires temperature=1 when thinking is enabled; override any
 		// caller-supplied value (e.g. temperature=0 set for deterministic non-interactive mode).
-		params.temperature = 1;
+		if (!alwaysThinking) params.temperature = 1;
 	}
 
 	const metadataUserId = resolveAnthropicMetadataUserId(options?.metadata?.user_id, isOAuthToken);
@@ -1588,16 +1627,16 @@ function buildParams(
 		params.metadata = { user_id: metadataUserId };
 	}
 
-	if (options?.toolChoice) {
-		if (typeof options.toolChoice === "string") {
-			params.tool_choice = { type: options.toolChoice };
-		} else if (isOAuthToken && options.toolChoice.name) {
+	if (toolChoice) {
+		if (typeof toolChoice === "string") {
+			params.tool_choice = { type: toolChoice };
+		} else if (isOAuthToken && toolChoice.name) {
 			params.tool_choice = {
-				...options.toolChoice,
-				name: applyClaudeToolPrefix(options.toolChoice.name),
+				...toolChoice,
+				name: applyClaudeToolPrefix(toolChoice.name),
 			};
 		} else {
-			params.tool_choice = options.toolChoice;
+			params.tool_choice = toolChoice;
 		}
 	}
 

package/src/providers/google-gemini-cli.ts

@@ -79,7 +79,7 @@ export function resolveAntigravityServingModelId(requestedModelId: string): stri
  * GeminiCLI/VERSION/MODEL (PLATFORM; ARCH; SURFACE)
  */
 export function getGeminiCliUserAgent(modelId = "gemini-3.1-pro-preview"): string {
-	const version = process.env.PI_AI_GEMINI_CLI_VERSION || "0.35.3";
+	const version = process.env.PI_AI_GEMINI_CLI_VERSION || "0.60.0";
 	const platform = process.platform === "win32" ? "win32" : process.platform;
 	const arch = process.arch === "x64" ? "x64" : process.arch;
 	return `GeminiCLI/${version}/${modelId} (${platform}; ${arch}; terminal)`;

package/src/stream.ts

@@ -1,11 +1,13 @@
 import { $env, $pickenv } from "@f5-sales-demo/pi-utils";
 import { getCustomApi } from "./api-registry";
-import type { Effort, ReasoningEffort } from "./model-thinking";
 import {
 	clampEffortThroughXHigh,
+	Effort,
 	type EffortThroughXHigh,
+	isAnthropicAlwaysThinkingModel,
 	mapEffortToAnthropicAdaptiveEffort,
 	mapEffortToGoogleThinkingLevel,
+	type ReasoningEffort,
 	requireSupportedEffort,
 	requireSupportedReasoningEffort,
 } from "./model-thinking";
@@ -428,8 +430,9 @@ export function mapOptionsForApi<TApi extends Api>(
 
 	switch (model.api) {
 		case "anthropic-messages": {
-			// Explicitly disable thinking when reasoning is not specified or model doesn't support it
-			const reasoning = options?.reasoning;
+			// Fable requires adaptive thinking on every request. Other Anthropic models
+			// retain the caller-controlled behavior.
+			const reasoning = options?.reasoning ?? (isAnthropicAlwaysThinkingModel(model) ? Effort.High : undefined);
 			if (!reasoning || !model.reasoning) {
 				return castApi<"anthropic-messages">({
 					...base,
@@ -438,7 +441,7 @@ export function mapOptionsForApi<TApi extends Api>(
 				});
 			}
 
-			let thinkingBudget = options.thinkingBudgets?.[reasoning] ?? ANTHROPIC_THINKING[reasoning];
+			let thinkingBudget = options?.thinkingBudgets?.[reasoning] ?? ANTHROPIC_THINKING[reasoning];
 			if (thinkingBudget <= 0) {
 				return castApi<"anthropic-messages">({
 					...base,