@oh-my-pi/pi-catalog 17.1.2 → 17.1.4

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

@@ -7,6 +7,7 @@ import {
 import { Effort, THINKING_EFFORTS } from "../effort";
 import { FIREWORKS_FAST_SUFFIX, toFireworksPublicModelId } from "../fireworks-model-id";
 import {
+	anthropicModelSupportsThinking,
 	isGlmVisionModelId,
 	isGrokReasoningEffortCapable,
 	isKimiK3ModelId,
@@ -17,7 +18,7 @@ import type { ModelManagerOptions } from "../model-manager";
 import { getBundledModels } from "../models";
 import type { Api, FetchImpl, Model, ModelSpec, OpenAICompat, Provider, ThinkingConfig } from "../types";
 import { discoveryFetch, isAnthropicOAuthToken, isRecord, toBoolean, toNumber, toPositiveNumber } from "../utils";
-import { parseAlibabaTokenPlanCredential } from "../wire/alibaba-token-plan";
+import { ALIBABA_TOKEN_PLAN_BASE_URL, parseAlibabaTokenPlanCredential } from "../wire/alibaba-token-plan";
 import { coreWeaveProjectHeaders } from "../wire/coreweave";
 import {
 	COPILOT_API_HEADERS,
@@ -2434,7 +2435,7 @@ export function alibabaCodingPlanModelManagerOptions(
 // Alibaba Token Plan
 // ---------------------------------------------------------------------------
 
-export const ALIBABA_TOKEN_PLAN_BASE_URL = "https://token-plan.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1";
+export { ALIBABA_TOKEN_PLAN_BASE_URL };
 
 const ALIBABA_TOKEN_PLAN_COST = { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 } as const;
 const ALIBABA_TOKEN_PLAN_COMPAT: OpenAICompat = {
@@ -2556,7 +2557,10 @@ export function alibabaTokenPlanModelManagerOptions(
 ): ModelManagerOptions<"openai-completions"> {
 	const credential = config?.apiKey ? parseAlibabaTokenPlanCredential(config.apiKey) : undefined;
 	const apiKey = credential?.token;
-	const baseUrl = config?.baseUrl ?? ALIBABA_TOKEN_PLAN_BASE_URL;
+	// A region-locked credential (China/custom) dictates the discovery endpoint:
+	// its key only authenticates against its own region, so fetching /models from
+	// any other base URL would 401 (#6682).
+	const baseUrl = credential?.baseUrl ?? config?.baseUrl ?? ALIBABA_TOKEN_PLAN_BASE_URL;
 	return {
 		providerId: "alibaba-token-plan",
 		dynamicModelsAuthoritative: true,
@@ -4279,6 +4283,13 @@ export function githubCopilotModelManagerOptions(config?: GithubCopilotModelMana
 	return {
 		providerId: "github-copilot",
 		dropCachedModelIdsOnStaticMismatch: COPILOT_CACHE_INVALIDATED_MODEL_IDS,
+		// COPILOT_API_HEADERS are compile-time constants (User-Agent + API
+		// version), not credentials. The cache omits all request headers for
+		// safety and can only restore them from a bundled static entry — so a
+		// Copilot model with no bundled reference (e.g. a freshly served
+		// claude-opus-5 and its synthesized -1m sibling) is dropped on offline
+		// reads. Declaring the constant lets the cache restore it by value.
+		restorableHeaderFallback: { ...COPILOT_API_HEADERS },
 		...(apiKey && {
 			fetchDynamicModels: async () => {
 				const longContextVariants: ModelSpec<Api>[] = [];
@@ -4377,6 +4388,17 @@ export function githubCopilotModelManagerOptions(config?: GithubCopilotModelMana
 									contextWindow: defaultTierWindow,
 									maxTokens,
 									headers: { ...COPILOT_API_HEADERS },
+									// Copilot's `/models` advertises no reasoning bit, so a
+									// thinking-capable Claude with no bundled reference would
+									// fall back to `reasoning: false` and lose its effort dial.
+									// Gate on the id classifier (not the transport alone) so a
+									// lagging enterprise catalog serving a pre-thinking Claude
+									// (<= 3.5) over the Messages proxy is not handed a fabricated
+									// dial it would reject; a modern reference-less model (e.g.
+									// claude-opus-5) is marked so `buildModel` derives the ladder.
+									...(api === "anthropic-messages" && anthropicModelSupportsThinking(defaults.id)
+										? { reasoning: true }
+										: {}),
 									...(api === "openai-completions"
 										? {
 												compat: {

package/src/types.ts

@@ -341,6 +341,12 @@ export interface OpenAICompat {
 	 * to opt a host out.
 	 */
 	toolSchemaFlavor?: "moonshot-mfjs" | "grammar" | "none";
+	/**
+	 * Stream-watchdog first-event timeout in ms.
+	 * Set to `0` to allow unbounded prompt processing. Default: auto-detected
+	 * (disabled for local OpenAI-compatible backends).
+	 */
+	streamFirstEventTimeoutMs?: number;
 	/**
 	 * Stream-watchdog idle-timeout floor in ms for slow reasoning hosts.
 	 * Default: auto-detected (GLM coding-plan hosts, direct DeepSeek reasoning).
@@ -565,6 +571,8 @@ export interface ResolvedOpenAISharedCompat {
 	requiresAssistantContentForToolCalls: boolean;
 	stripDeepseekSpecialTokens: boolean;
 	streamMarkupHealingPattern?: OpenAIStreamMarkupHealingPattern;
+	/** See {@link OpenAICompat.streamFirstEventTimeoutMs}. */
+	streamFirstEventTimeoutMs?: number;
 	reasoningDeltasMayBeCumulative: boolean;
 	emptyLengthFinishIsContextError: boolean;
 	usesOpenAIToolCallIdLimit: boolean;
@@ -645,6 +653,7 @@ export type ResolvedOpenAICompat = ResolvedOpenAISharedCompat &
 			| "extraBody"
 			| "toolStrictMode"
 			| "toolSchemaFlavor"
+			| "streamFirstEventTimeoutMs"
 			| "streamIdleTimeoutMs"
 			| "cacheControlFormat"
 			| "thinkingKeep"
@@ -816,6 +825,8 @@ export interface Model<TApi extends Api = Api> {
 	supportsTools?: boolean;
 	/** Whether this model accepts the GA OpenAI Responses `{ type: "computer" }` native tool. */
 	supportsComputerUse?: boolean;
+	/** Verbatim explicit computer-use support from the spec; undefined when `buildModel` inferred the runtime value. */
+	supportsComputerUseConfig?: boolean;
 	/** GitLab Duo Workflow root namespace selected during catalog discovery. */
 	gitlabDuoWorkflowRootNamespaceId?: string;
 	/** Cursor `max_mode` request flag returned by `GetUsableModels` for premium models that require max mode. */
@@ -901,7 +912,8 @@ export interface Model<TApi extends Api = Api> {
  * vocabulary of `buildModel`. Identical to `Model` except `compat` carries the
  * sparse override shape and nothing is resolved yet.
  */
-export interface ModelSpec<TApi extends Api = Api> extends Omit<Model<TApi>, "compat" | "compatConfig"> {
+export interface ModelSpec<TApi extends Api = Api>
+	extends Omit<Model<TApi>, "compat" | "compatConfig" | "supportsComputerUseConfig"> {
 	/** Sparse compatibility overrides; resolved into `Model.compat` by `buildModel`. */
 	compat?: CompatConfigOf<TApi>;
 }

package/src/wire/alibaba-token-plan.ts

@@ -1,6 +1,23 @@
+/**
+ * International (Singapore) Token Plan endpoint. Default region; keys issued by
+ * the international product authenticate only here.
+ */
+export const ALIBABA_TOKEN_PLAN_BASE_URL = "https://token-plan.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1";
+/**
+ * China (Beijing) Token Plan endpoint (百炼 Token Plan). Keys are region-locked:
+ * a Beijing-issued key is rejected by the international endpoint with
+ * `invalid_api_key`, and vice versa (#6682).
+ */
+export const ALIBABA_TOKEN_PLAN_CN_BASE_URL = "https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1";
+
 export interface AlibabaTokenPlanCredential {
 	token: string;
 	cookie?: string;
+	/**
+	 * Region base URL the key authenticates against. Absent means the default
+	 * international endpoint ({@link ALIBABA_TOKEN_PLAN_BASE_URL}).
+	 */
+	baseUrl?: string;
 }
 
 const TOKEN_PATTERN = /^sk-[A-Za-z0-9._~+/-]+={0,2}$/;
@@ -10,18 +27,28 @@ export function parseAlibabaTokenPlanCredential(value: string): AlibabaTokenPlan
 	if (!trimmed) return null;
 	if (!trimmed.startsWith("{")) return TOKEN_PATTERN.test(trimmed) ? { token: trimmed } : null;
 	try {
-		const parsed = JSON.parse(trimmed) as { token?: unknown; cookie?: unknown };
+		const parsed = JSON.parse(trimmed) as { token?: unknown; cookie?: unknown; baseUrl?: unknown };
 		if (typeof parsed.token !== "string" || !TOKEN_PATTERN.test(parsed.token.trim())) return null;
 		if (parsed.cookie !== undefined && typeof parsed.cookie !== "string") return null;
+		if (parsed.baseUrl !== undefined && typeof parsed.baseUrl !== "string") return null;
 		const token = parsed.token.trim();
 		const cookie = parsed.cookie?.trim();
-		return cookie ? { token, cookie } : { token };
+		const baseUrl = parsed.baseUrl?.trim();
+		const credential: AlibabaTokenPlanCredential = { token };
+		if (cookie) credential.cookie = cookie;
+		if (baseUrl) credential.baseUrl = baseUrl;
+		return credential;
 	} catch {
 		return null;
 	}
 }
 
-export function serializeAlibabaTokenPlanCredential(token: string, cookie: string): string {
+export function serializeAlibabaTokenPlanCredential(token: string, cookie: string, baseUrl?: string): string {
 	const trimmedCookie = cookie.trim();
-	return trimmedCookie ? JSON.stringify({ token, cookie: trimmedCookie }) : token;
+	const trimmedBaseUrl = baseUrl?.trim();
+	if (!trimmedCookie && !trimmedBaseUrl) return token;
+	const payload: { token: string; cookie?: string; baseUrl?: string } = { token };
+	if (trimmedCookie) payload.cookie = trimmedCookie;
+	if (trimmedBaseUrl) payload.baseUrl = trimmedBaseUrl;
+	return JSON.stringify(payload);
 }