@sayknow-cli/ai 0.3.8 → 0.3.9

package/dist/types/provider-models/openai-compat.d.ts

@@ -58,6 +58,11 @@ export interface DeepSeekModelManagerConfig {
     baseUrl?: string;
 }
 export declare function deepseekModelManagerOptions(config?: DeepSeekModelManagerConfig): ModelManagerOptions<"openai-completions">;
+export interface DeepInfraModelManagerConfig {
+    apiKey?: string;
+    baseUrl?: string;
+}
+export declare function deepinfraModelManagerOptions(config?: DeepInfraModelManagerConfig): ModelManagerOptions<"openai-completions">;
 export interface FireworksModelManagerConfig {
     apiKey?: string;
     baseUrl?: string;

package/dist/types/providers/cursor/client-version.d.ts

@@ -0,0 +1,10 @@
+/**
+ * The Cursor client version reported to api2.cursor.sh.
+ *
+ * Every call against the Cursor backend must send the same
+ * x-cursor-client-version: the backend gates features and minimum versions on
+ * it, so a drift between the agent Run path and model discovery makes one of
+ * them fail while the other keeps working. Keep this as the single source of
+ * truth for the header value.
+ */
+export declare const CURSOR_CLIENT_VERSION = "cli-2026.02.13-41ac335";

package/dist/types/providers/cursor.d.ts

@@ -1,7 +1,8 @@
 import { type JsonValue } from "@bufbuild/protobuf";
 import type { CursorExecHandlerResult, CursorExecHandlers, CursorToolResultHandler, Message, StreamFunction, StreamOptions, ToolResultMessage } from "../types";
+import { CURSOR_CLIENT_VERSION } from "./cursor/client-version";
 export declare const CURSOR_API_URL = "https://api2.cursor.sh";
-export declare const CURSOR_CLIENT_VERSION = "cli-2026.01.09-231024f";
+export { CURSOR_CLIENT_VERSION };
 /** Drop all cached state + blob bytes for a conversation (F15 bound + session-teardown hook). */
 export declare function disposeCursorConversation(conversationId: string): void;
 export interface CursorOptions extends StreamOptions {

package/dist/types/types.d.ts

@@ -48,7 +48,7 @@ export interface ThinkingConfig {
     /** Provider-specific transport used to encode the selected effort. */
     mode: ThinkingControlMode;
 }
-export type KnownProvider = "alibaba-coding-plan" | "amazon-bedrock" | "azure-openai" | "anthropic" | "google" | "google-gemini-cli" | "google-antigravity" | "google-vertex" | "openai" | "openai-codex" | "kimi-code" | "minimax-code" | "minimax-code-cn" | "github-copilot" | "fireworks" | "firepass" | "fugu" | "gitlab-duo" | "cursor" | "deepseek" | "xai" | "groq" | "cerebras" | "openrouter" | "kilo" | "vercel-ai-gateway" | "zai" | "glm-zcode" | "mistral" | "minimax" | "opencode-go" | "opencode-zen" | "synthetic" | "cloudflare-ai-gateway" | "huggingface" | "litellm" | "moonshot" | "nvidia" | "nanogpt" | "ollama" | "ollama-cloud" | "qianfan" | "qwen-portal" | "together" | "venice" | "vllm" | "xiaomi" | "xiaomi-token-plan-sgp" | "xiaomi-token-plan-ams" | "xiaomi-token-plan-cn" | "zenmux" | "lm-studio";
+export type KnownProvider = "alibaba-coding-plan" | "amazon-bedrock" | "azure-openai" | "anthropic" | "google" | "google-gemini-cli" | "google-antigravity" | "google-vertex" | "openai" | "openai-codex" | "kimi-code" | "minimax-code" | "minimax-code-cn" | "github-copilot" | "fireworks" | "firepass" | "fugu" | "gitlab-duo" | "cursor" | "deepseek" | "deepinfra" | "xai" | "groq" | "cerebras" | "openrouter" | "kilo" | "vercel-ai-gateway" | "zai" | "glm-zcode" | "mistral" | "minimax" | "opencode-go" | "opencode-zen" | "synthetic" | "cloudflare-ai-gateway" | "huggingface" | "litellm" | "moonshot" | "nvidia" | "nanogpt" | "ollama" | "ollama-cloud" | "qianfan" | "qwen-portal" | "together" | "venice" | "vllm" | "xiaomi" | "xiaomi-token-plan-sgp" | "xiaomi-token-plan-ams" | "xiaomi-token-plan-cn" | "zenmux" | "lm-studio";
 export type Provider = KnownProvider | string;
 import type { Effort } from "./model-thinking";
 /** Token budgets for each thinking level (token-based providers only) */
@@ -106,9 +106,9 @@ export type ResolvedServiceTier = Exclude<ServiceTier, "openai-only" | "claude-o
 export declare function resolveServiceTier(serviceTier: ServiceTier | null | undefined, provider: Provider | undefined): ResolvedServiceTier | undefined;
 /**
  * True when the (possibly scoped) tier should be sent as OpenAI's
- * `service_tier` request field for the given provider. Non-OpenAI
- * providers, unsupported tiers (`"auto"`, `"default"`), and scope
- * mismatches all return false.
+ * `service_tier` request field for the given provider. OpenAI accepts
+ * `flex`, `scale`, and `priority`; DeepInfra accepts `priority`.
+ * Unsupported tiers (`"auto"`, `"default"`) and scope mismatches return false.
  */
 export declare function shouldSendServiceTier(serviceTier: ServiceTier | null | undefined, provider: Provider | undefined): boolean;
 /**

package/dist/types/utils/oauth/types.d.ts

@@ -7,7 +7,7 @@ export type OAuthCredentials = {
     email?: string;
     accountId?: string;
 };
-export type OAuthProvider = "alibaba-coding-plan" | "anthropic" | "cerebras" | "cloudflare-ai-gateway" | "cursor" | "deepseek" | "fireworks" | "firepass" | "fugu" | "github-copilot" | "google-gemini-cli" | "google-antigravity" | "gitlab-duo" | "huggingface" | "kimi-code" | "kilo" | "kagi" | "litellm" | "lm-studio" | "minimax-code" | "minimax-code-cn" | "moonshot" | "nvidia" | "nanogpt" | "ollama" | "ollama-cloud" | "openai-codex" | "openai-codex-device" | "opencode-go" | "opencode-zen" | "parallel" | "perplexity" | "qianfan" | "qwen-portal" | "synthetic" | "tavily" | "together" | "venice" | "vercel-ai-gateway" | "vllm" | "xai" | "glm-zcode" | "xiaomi" | "xiaomi-token-plan-sgp" | "xiaomi-token-plan-ams" | "xiaomi-token-plan-cn" | "zenmux" | "zai";
+export type OAuthProvider = "alibaba-coding-plan" | "anthropic" | "cerebras" | "cloudflare-ai-gateway" | "cursor" | "deepseek" | "deepinfra" | "fireworks" | "firepass" | "fugu" | "github-copilot" | "google-gemini-cli" | "google-antigravity" | "gitlab-duo" | "huggingface" | "kimi-code" | "kilo" | "kagi" | "litellm" | "lm-studio" | "minimax-code" | "minimax-code-cn" | "moonshot" | "nvidia" | "nanogpt" | "ollama" | "ollama-cloud" | "openai-codex" | "openai-codex-device" | "opencode-go" | "opencode-zen" | "parallel" | "perplexity" | "qianfan" | "qwen-portal" | "synthetic" | "tavily" | "together" | "venice" | "vercel-ai-gateway" | "vllm" | "xai" | "glm-zcode" | "xiaomi" | "xiaomi-token-plan-sgp" | "xiaomi-token-plan-ams" | "xiaomi-token-plan-cn" | "zenmux" | "zai";
 export type OAuthProviderId = OAuthProvider | (string & {});
 export type OAuthPrompt = {
     message: string;

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@sayknow-cli/ai",
-	"version": "0.3.8",
+	"version": "0.3.9",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"homepage": "https://sayknow-cli.com",
 	"author": "jaybeyond",
@@ -43,7 +43,7 @@
 	"dependencies": {
 		"@anthropic-ai/sdk": "^0.94.0",
 		"@bufbuild/protobuf": "^2.12.0",
-		"@sayknow-cli/utils": "0.3.8",
+		"@sayknow-cli/utils": "0.3.9",
 		"openai": "^6.36.0",
 		"partial-json": "^0.1.7",
 		"zod": "4.4.3"

package/src/provider-models/descriptors.ts

@@ -13,6 +13,7 @@ import {
 	anthropicModelManagerOptions,
 	cerebrasModelManagerOptions,
 	cloudflareAiGatewayModelManagerOptions,
+	deepinfraModelManagerOptions,
 	deepseekModelManagerOptions,
 	firepassModelManagerOptions,
 	fireworksModelManagerOptions,
@@ -168,6 +169,12 @@ export const PROVIDER_DESCRIPTORS: readonly ProviderDescriptor[] = [
 		config => deepseekModelManagerOptions(config),
 		catalog("DeepSeek", ["DEEPSEEK_API_KEY"]),
 	),
+	catalogDescriptor(
+		"deepinfra",
+		"deepseek-ai/DeepSeek-V3.2",
+		config => deepinfraModelManagerOptions(config),
+		catalog("DeepInfra", ["DEEPINFRA_API_KEY"]),
+	),
 	descriptor("mistral", "devstral-medium-latest", config => mistralModelManagerOptions(config)),
 	catalogDescriptor(
 		"nvidia",

package/src/provider-models/openai-compat.ts

@@ -665,6 +665,21 @@ export function deepseekModelManagerOptions(
 ): ModelManagerOptions<"openai-completions"> {
 	return createSimpleOpenAICompletionsOptions("deepseek", "https://api.deepseek.com", config);
 }
+export interface DeepInfraModelManagerConfig {
+	apiKey?: string;
+	baseUrl?: string;
+}
+
+export function deepinfraModelManagerOptions(
+	config?: DeepInfraModelManagerConfig,
+): ModelManagerOptions<"openai-completions"> {
+	return createSimpleOpenAICompletionsOptions(
+		"deepinfra" as Parameters<typeof getBundledModels>[0],
+		"https://api.deepinfra.com/v1/openai",
+		config,
+	);
+}
+
 // ---------------------------------------------------------------------------
 // 7.5 Fireworks
 // ---------------------------------------------------------------------------
@@ -2254,6 +2269,8 @@ const MODELS_DEV_PROVIDER_DESCRIPTORS_CORE: readonly ModelsDevProviderDescriptor
 	}),
 	// --- xAI ---
 	openAiCompletionsDescriptor("xai", "xai", "https://api.x.ai/v1"),
+	// --- DeepInfra ---
+	openAiCompletionsDescriptor("deepinfra", "deepinfra", "https://api.deepinfra.com/v1/openai"),
 	// --- DeepSeek ---
 	openAiCompletionsDescriptor("deepseek", "deepseek", "https://api.deepseek.com", {
 		// Only ship the v4 family as built-ins; older deepseek-chat / deepseek-reasoner

package/src/providers/cursor/client-version.ts

@@ -0,0 +1,10 @@
+/**
+ * The Cursor client version reported to api2.cursor.sh.
+ *
+ * Every call against the Cursor backend must send the same
+ * x-cursor-client-version: the backend gates features and minimum versions on
+ * it, so a drift between the agent Run path and model discovery makes one of
+ * them fail while the other keeps working. Keep this as the single source of
+ * truth for the header value.
+ */
+export const CURSOR_CLIENT_VERSION = "cli-2026.02.13-41ac335";

package/src/providers/cursor.ts

@@ -31,6 +31,7 @@ import { parseStreamingJson } from "../utils/json-parse";
 import { formatErrorMessageWithRetryAfter } from "../utils/retry-after";
 import { flattenToolRootCombinators, toolWireSchema } from "../utils/schema";
 import { COMPOSER_EDIT_DISCIPLINE_PROMPT, isComposerHarnessModel } from "./composer-discipline";
+import { CURSOR_CLIENT_VERSION } from "./cursor/client-version";
 import type { McpToolDefinition } from "./cursor/gen/agent_pb";
 import {
 	AgentClientMessageSchema,
@@ -131,7 +132,7 @@ import {
 } from "./cursor/gen/agent_pb";
 
 export const CURSOR_API_URL = "https://api2.cursor.sh";
-export const CURSOR_CLIENT_VERSION = "cli-2026.01.09-231024f";
+export { CURSOR_CLIENT_VERSION };
 
 const conversationStateCache = new Map<string, ConversationStateStructure>();
 const conversationBlobStores = new Map<string, Map<string, Uint8Array>>();

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

@@ -4,7 +4,7 @@
  * 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 || "0.46.0";
+	const version = process.env.SKC_AI_GEMINI_CLI_VERSION || process.env.PI_AI_GEMINI_CLI_VERSION || "0.49.0";
 	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/providers/openai-completions.ts

@@ -35,7 +35,7 @@ import {
 	type ToolChoice,
 	type ToolResultMessage,
 } from "../types";
-import { normalizeSystemPrompts } from "../utils";
+import { normalizeSystemPrompts, sanitizeJsonStrings } from "../utils";
 import { createAbortSourceTracker } from "../utils/abort";
 import { AssistantMessageEventStream } from "../utils/event-stream";
 import { toFirepassWireModelId, toFireworksWireModelId } from "../utils/fireworks-model-id";
@@ -166,7 +166,7 @@ function normalizeStreamingContentText(content: unknown): string {
 function serializeToolArguments(value: unknown): string {
 	if (value && typeof value === "object" && !Array.isArray(value)) {
 		try {
-			return JSON.stringify(value);
+			return JSON.stringify(sanitizeJsonStrings(value));
 		} catch {
 			return "{}";
 		}
@@ -178,7 +178,7 @@ function serializeToolArguments(value: unknown): string {
 		try {
 			const parsed = JSON.parse(trimmed);
 			if (parsed && typeof parsed === "object" && !Array.isArray(parsed)) {
-				return JSON.stringify(parsed);
+				return JSON.stringify(sanitizeJsonStrings(parsed));
 			}
 		} catch {}
 		return "{}";

package/src/rate-limit-utils.ts

@@ -13,7 +13,7 @@ export type RateLimitReason =
 const QUOTA_EXHAUSTED_BACKOFF_MS = 30 * 60 * 1000; // 30 min
 const RATE_LIMIT_EXCEEDED_BACKOFF_MS = 30 * 1000; // 30s
 const MODEL_CAPACITY_BASE_MS = 45 * 1000; // 45s base
-const MODEL_CAPACITY_JITTER_MS = 30 * 1000; // ±15s
+const MODEL_CAPACITY_JITTER_MS = 30 * 1000; // uniform +0–30s above base → 45–75s total
 const SERVER_ERROR_BACKOFF_MS = 20 * 1000; // 20s
 
 /**
@@ -93,8 +93,10 @@ export function calculateRateLimitBackoffMs(reason: RateLimitReason): number {
 }
 
 /** Detect usage/quota limit errors in error messages (persistent, requires credential switch). */
+// ZAI reports durable token exhaustion as "[1310][Weekly/Monthly Limit Exhausted...]".
+// Keep this explicit so generic "rate limit exhausted, retry..." throttles remain retryable.
 const USAGE_LIMIT_PATTERN =
-	/usage.?limit|usage_limit_reached|usage_not_included|limit_reached|model.?limit|model_limit_reached|message.?limit|message_limit_reached|limit for this model|quota.?exceeded|out_of_credits|request would exceed your account.?s rate limit|resource has been exhausted[^\n]*(?:quota|limit)/i;
+	/usage.?limit|usage_limit_reached|usage_not_included|limit_reached|model.?limit|model_limit_reached|message.?limit|message_limit_reached|limit for this model|weekly\/monthly\s+limit\s+exhausted|quota.?exceeded|out_of_credits|request would exceed your account.?s rate limit|resource has been exhausted[^\n]*(?:quota|limit)/i;
 export function isUsageLimitError(errorMessage: string): boolean {
 	return USAGE_LIMIT_PATTERN.test(errorMessage);
 }

package/src/stream.ts

@@ -98,6 +98,7 @@ const serviceProviderMap: Record<string, KeyResolver> = {
 	"opencode-zen": "OPENCODE_API_KEY",
 	cursor: "CURSOR_ACCESS_TOKEN",
 	deepseek: "DEEPSEEK_API_KEY",
+	deepinfra: "DEEPINFRA_API_KEY",
 	"openai-codex": "OPENAI_CODEX_OAUTH_TOKEN",
 	"azure-openai": "AZURE_OPENAI_API_KEY",
 	"azure-openai-responses": "AZURE_OPENAI_API_KEY",

package/src/types.ts

@@ -116,6 +116,7 @@ export type KnownProvider =
 	| "gitlab-duo"
 	| "cursor"
 	| "deepseek"
+	| "deepinfra"
 	| "xai"
 	| "groq"
 	| "cerebras"
@@ -224,17 +225,20 @@ export function resolveServiceTier(
 
 /**
  * True when the (possibly scoped) tier should be sent as OpenAI's
- * `service_tier` request field for the given provider. Non-OpenAI
- * providers, unsupported tiers (`"auto"`, `"default"`), and scope
- * mismatches all return false.
+ * `service_tier` request field for the given provider. OpenAI accepts
+ * `flex`, `scale`, and `priority`; DeepInfra accepts `priority`.
+ * Unsupported tiers (`"auto"`, `"default"`) and scope mismatches return false.
  */
 export function shouldSendServiceTier(
 	serviceTier: ServiceTier | null | undefined,
 	provider: Provider | undefined,
 ): boolean {
-	if (provider !== "openai" && provider !== "openai-codex") return false;
 	const resolved = resolveServiceTier(serviceTier, provider);
-	return resolved === "flex" || resolved === "scale" || resolved === "priority";
+	if (provider === "openai" || provider === "openai-codex") {
+		return resolved === "flex" || resolved === "scale" || resolved === "priority";
+	}
+	if (provider === "deepinfra") return resolved === "priority";
+	return false;
 }
 
 /**
@@ -252,7 +256,9 @@ export function getPriorityPremiumRequests(
 	if (resolveServiceTier(serviceTier, provider) !== "priority") return 0;
 	// Only providers that realize `priority` on the wire bill the user.
 	// Everywhere else, the field is silently dropped and nothing is charged.
-	return provider === "openai" || provider === "openai-codex" || provider === "anthropic" ? 1 : 0;
+	return provider === "openai" || provider === "openai-codex" || provider === "anthropic" || provider === "deepinfra"
+		? 1
+		: 0;
 }
 
 export interface ProviderSessionState {

package/src/utils/discovery/cursor.ts

@@ -2,11 +2,11 @@ import * as http2 from "node:http2";
 import { create, fromBinary, toBinary } from "@bufbuild/protobuf";
 import * as z from "zod/v4";
 import { getBundledModels } from "../../models";
+import { CURSOR_CLIENT_VERSION } from "../../providers/cursor/client-version";
 import { GetUsableModelsRequestSchema, GetUsableModelsResponseSchema } from "../../providers/cursor/gen/agent_pb";
 import type { Model } from "../../types";
 
 const CURSOR_DEFAULT_BASE_URL = "https://api2.cursor.sh";
-const CURSOR_DEFAULT_CLIENT_VERSION = "cli-2026.02.13-41ac335";
 const CURSOR_GET_USABLE_MODELS_PATH = "/agent.v1.AgentService/GetUsableModels";
 
 const DEFAULT_CONTEXT_WINDOW = 200_000;
@@ -91,7 +91,7 @@ function buildRequestHeaders(options: CursorModelDiscoveryOptions): Record<strin
 		te: "trailers",
 		authorization: `Bearer ${options.apiKey}`,
 		"x-ghost-mode": "true",
-		"x-cursor-client-version": options.clientVersion ?? CURSOR_DEFAULT_CLIENT_VERSION,
+		"x-cursor-client-version": options.clientVersion ?? CURSOR_CLIENT_VERSION,
 		"x-cursor-client-type": "cli",
 	};
 }

package/src/utils/oauth/fugu.ts

@@ -3,10 +3,10 @@ import { createApiKeyLogin } from "./api-key-login";
 
 export const loginFugu = createApiKeyLogin({
 	providerLabel: "Sakana Fugu",
-	authUrl: "https://fugu.sakana.ai/",
+	authUrl: "https://console.sakana.ai/api-keys",
 	instructions: "Create or copy your Sakana Fugu API key",
 	promptMessage: "Paste your Sakana Fugu API key",
-	placeholder: "fugu_...",
+	placeholder: "fish_...",
 	validation: {
 		kind: "models-endpoint",
 		provider: "Sakana Fugu",

package/src/utils/oauth/index.ts

@@ -60,6 +60,11 @@ const builtInOAuthProviders: OAuthProviderInfo[] = [
 		name: "DeepSeek",
 		available: true,
 	},
+	{
+		id: "deepinfra",
+		name: "DeepInfra",
+		available: true,
+	},
 	{
 		id: "xai",
 		name: "xAI",
@@ -359,6 +364,7 @@ export async function refreshOAuthToken(
 		case "fireworks":
 		case "firepass":
 		case "fugu":
+		case "deepinfra":
 		case "nvidia":
 		case "nanogpt":
 		case "synthetic":

package/src/utils/oauth/types.ts

@@ -15,6 +15,7 @@ export type OAuthProvider =
 	| "cloudflare-ai-gateway"
 	| "cursor"
 	| "deepseek"
+	| "deepinfra"
 	| "fireworks"
 	| "firepass"
 	| "fugu"