@linxiraos/pi-catalog 1.1.26 → 1.2.0

package/src/provider-models/account-access.d.ts

@@ -0,0 +1,11 @@
+import type { Api, ModelSpec } from "../types";
+/**
+ * Merge per-account discovery catalogs into one list, deduped by model id.
+ *
+ * The first catalog to list an id supplies its spec; {@link ModelSpec.accountAccess}
+ * is merged from every catalog listing that id, so credential selection can
+ * route account-gated models to the accounts that serve them. Used by
+ * multi-account discovery for Codex and Antigravity, whose rosters differ per
+ * account.
+ */
+export declare function unionAccountCatalogs<TApi extends Api>(catalogs: readonly (readonly ModelSpec<TApi>[])[]): ModelSpec<TApi>[];

package/src/provider-models/account-access.ts

@@ -0,0 +1,30 @@
+import type { Api, ModelSpec } from "../types";
+
+/**
+ * Merge per-account discovery catalogs into one list, deduped by model id.
+ *
+ * The first catalog to list an id supplies its spec; {@link ModelSpec.accountAccess}
+ * is merged from every catalog listing that id, so credential selection can
+ * route account-gated models to the accounts that serve them. Used by
+ * multi-account discovery for Codex and Antigravity, whose rosters differ per
+ * account.
+ */
+export function unionAccountCatalogs<TApi extends Api>(
+	catalogs: readonly (readonly ModelSpec<TApi>[])[],
+): ModelSpec<TApi>[] {
+	const byId = new Map<string, ModelSpec<TApi>>();
+	for (const catalog of catalogs) {
+		for (const model of catalog) {
+			const existing = byId.get(model.id);
+			if (!existing) {
+				byId.set(model.id, model);
+			} else if (model.accountAccess) {
+				byId.set(model.id, {
+					...existing,
+					accountAccess: { ...existing.accountAccess, ...model.accountAccess },
+				});
+			}
+		}
+	}
+	return [...byId.values()];
+}

package/src/provider-models/cache-provider-id.ts

@@ -1,5 +1,5 @@
 import { CHARM_HYPER_API_BASE_URL, normalizeCharmHyperBaseUrl } from "../wire/charm-hyper";
-import { CODEX_CLIENT_VERSION } from "../wire/codex";
+import { CODEX_BASE_URL, CODEX_CLIENT_VERSION } from "../wire/codex";
 import { CURSOR_DEFAULT_BASE_URL } from "../wire/cursor";
 import { type AccountScope, factoryDroidModelCacheProviderId } from "../wire/factory-droid";
 import { PERSONAL_GITHUB_COPILOT_BASE_URL } from "../wire/github-copilot";
@@ -91,9 +91,15 @@ function cursorCredentialSubject(apiKey: string): string | undefined {
 /** Resolve the cache namespace used by a provider's model-manager options without constructing those options. */
 export function resolveModelCacheProviderId(providerId: string, options: ModelCacheProviderIdOptions = {}): string {
 	switch (providerId) {
-		case "openai-codex":
-			// The backend filters the roster by client version.
-			return `${providerId}:${CODEX_CLIENT_VERSION}`;
+		case "openai-codex": {
+			// The backend filters the roster by client version. A Codex-compatible
+			// gateway publishes its own roster; discovery is authoritative, so the
+			// namespace must follow the endpoint. The official backend keeps the
+			// unhashed namespace so existing caches stay valid.
+			const baseUrl = options.baseUrl?.trim().replace(/\/+$/, "");
+			if (!baseUrl || baseUrl === CODEX_BASE_URL) return `${providerId}:${CODEX_CLIENT_VERSION}`;
+			return `${providerId}:${CODEX_CLIENT_VERSION}:${Bun.hash(baseUrl).toString(36)}`;
+		}
 		case "ollama":
 			return resolveOllamaModelCacheProviderId(providerId, options.baseUrl);
 		case "cursor": {
@@ -102,11 +108,12 @@ export function resolveModelCacheProviderId(providerId: string, options: ModelCa
 			// authoritative model rows.
 			// v3 invalidates zero-price rows written before rich lanes were priced
 			// from the KDL rate card; v4 keys the scope on the token's stable
-			// account subject instead of the rotating JWT.
+			// account subject instead of the rotating JWT; v5 drops rows that still
+			// mark the account default `isProviderDefault`, which startup now honors.
 			const baseUrl = (options.baseUrl ?? CURSOR_DEFAULT_BASE_URL).replace(/\/+$/, "");
 			const apiKey = options.apiKey ?? "";
 			const scope = `${cursorCredentialSubject(apiKey) ?? apiKey}\u0000${baseUrl}`;
-			return `cursor:rich-models-v4:${Bun.hash(scope).toString(36)}`;
+			return `cursor:rich-models-v5:${Bun.hash(scope).toString(36)}`;
 		}
 		case "charm-hyper": {
 			// Discovery is authoritative for this gateway, so a warm cache is served

package/src/provider-models/cline-pass.ts

@@ -34,6 +34,35 @@ const LOW_TO_MAX: ThinkingConfig = {
 	requiresEffort: false,
 };
 
+const MINIMAL_TO_XHIGH: ThinkingConfig = {
+	mode: "effort",
+	efforts: [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh],
+	defaultLevel: Effort.Medium,
+	requiresEffort: false,
+};
+
+const DEEPSEEK_V41_FLASH_FREE: ClinePassModelMetadata = {
+	name: "DeepSeek V4.1 Flash (free)",
+	contextWindow: 1_048_576,
+	maxTokens: 384_000,
+	input: ["text", "image"],
+	cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+	reasoning: true,
+	thinking: LOW_HIGH_MAX,
+	tier: "free",
+};
+
+const MUSE_SPARK_13_CONTRIBUTOR_FREE: ClinePassModelMetadata = {
+	name: "Muse Spark 1.3 (C) (free)",
+	contextWindow: 1_048_576,
+	maxTokens: 131_072,
+	input: ["text", "image"],
+	cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+	reasoning: true,
+	thinking: MINIMAL_TO_XHIGH,
+	tier: "free",
+};
+
 const QWEN_38_EFFORTS: ThinkingConfig = {
 	mode: "effort",
 	efforts: [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh],
@@ -219,6 +248,10 @@ export const CLINE_PASS_MODEL_METADATA: Readonly<Record<string, ClinePassModelMe
 		reasoning: true,
 		tier: "free",
 	},
+	"cline-free/deepseek-v4.1-flash": DEEPSEEK_V41_FLASH_FREE,
+	"deepseek/deepseek-v4.1-flash": DEEPSEEK_V41_FLASH_FREE,
+	"cline-free/muse-spark-1.3-contributor": MUSE_SPARK_13_CONTRIBUTOR_FREE,
+	"meta/muse-spark-1.3-contributor": MUSE_SPARK_13_CONTRIBUTOR_FREE,
 };
 
 export function getClinePassModelMetadata(id: string): ClinePassModelMetadata | undefined {

package/src/provider-models/descriptors.ts

@@ -29,6 +29,7 @@ import {
 	clinePassModelManagerOptions,
 	cloudflareAiGatewayModelManagerOptions,
 	commandCodeModelManagerOptions,
+	coralbricksModelManagerOptions,
 	coreWeaveModelManagerOptions,
 	deepinfraModelManagerOptions,
 	deepseekModelManagerOptions,
@@ -78,6 +79,7 @@ import {
 	zhipuCodingPlanModelManagerOptions,
 } from "./openai-compat";
 import {
+	cloudflareWorkersAiModelManagerOptions,
 	cursorModelManagerOptions,
 	devinModelManagerOptions,
 	factoryDroidModelManagerOptions,
@@ -104,7 +106,9 @@ const MODEL_MANAGER_FACTORIES: Readonly<Partial<Record<KnownProvider, ModelManag
 	cerebras: config => cerebrasModelManagerOptions(config),
 	"charm-hyper": config => charmHyperModelManagerOptions(config),
 	"cloudflare-ai-gateway": config => cloudflareAiGatewayModelManagerOptions(config),
+	"cloudflare-workers-ai": () => cloudflareWorkersAiModelManagerOptions(),
 	commandcode: config => commandCodeModelManagerOptions(config),
+	coralbricks: config => coralbricksModelManagerOptions(config),
 	cursor: config => cursorModelManagerOptions(config),
 	deepinfra: config => deepinfraModelManagerOptions(config),
 	deepseek: config => deepseekModelManagerOptions(config),

package/src/provider-models/google.d.ts

@@ -11,8 +11,30 @@ export interface GoogleVertexModelManagerConfig {
     signal?: AbortSignal;
     fetch?: FetchImpl;
 }
+/** One Antigravity OAuth account whose `fetchAvailableModels` roster discovery reads. */
+export interface GoogleAntigravityAccount {
+    /** OAuth access token used for `Authorization: Bearer ...`. */
+    accessToken: string;
+    /**
+     * Credential identity recorded in {@link ModelSpec.accountAccess} for every
+     * model this account serves (see `oauthAccountKey` in `@linxiraos/pi-ai`).
+     * When any account lacks one, no model records per-account access.
+     */
+    accountKey?: string;
+}
 export interface GoogleAntigravityModelManagerConfig {
-    oauthToken?: string;
+    /**
+     * Resolves every configured Antigravity account at discovery time. Rosters
+     * are plan-scoped (Claude 5.5 is served to some accounts only) and
+     * authoritative, so each account's roster is fetched and the results are
+     * unioned; reading one account's roster would prune models only its
+     * siblings serve (#14924).
+     *
+     * Returns `null` to abort discovery (e.g. an account's credential failed to
+     * refresh), keeping the previous/bundled catalog instead of caching a
+     * partial one.
+     */
+    resolveAccounts?: () => Promise<readonly GoogleAntigravityAccount[] | null>;
     endpoint?: string;
     fetch?: FetchImpl;
 }

package/src/provider-models/google.ts

@@ -1,10 +1,13 @@
+import { logger } from "@linxiraos/pi-utils";
 import { reviewedCollapseTable } from "../compat/collapse";
 import { classifyModel } from "../compat/taxonomy";
+import { providerEntry } from "../compat/providers";
 import { fetchAntigravityDiscoveryModels } from "../discovery/antigravity";
 import { fetchGeminiModels } from "../discovery/gemini";
 import { fetchGeminiCliQuotaModels } from "../discovery/gemini-cli";
 import type { ModelManagerOptions } from "../model-manager";
-import type { FetchImpl } from "../types";
+import type { FetchImpl, ModelSpec } from "../types";
+import { unionAccountCatalogs } from "./account-access";
 
 export interface GoogleModelManagerConfig {
 	apiKey?: string;
@@ -19,8 +22,31 @@ export interface GoogleVertexModelManagerConfig {
 	fetch?: FetchImpl;
 }
 
+/** One Antigravity OAuth account whose `fetchAvailableModels` roster discovery reads. */
+export interface GoogleAntigravityAccount {
+	/** OAuth access token used for `Authorization: Bearer ...`. */
+	accessToken: string;
+	/**
+	 * Credential identity recorded in {@link ModelSpec.accountAccess} for every
+	 * model this account serves (see `oauthAccountKey` in `@linxiraos/pi-ai`).
+	 * When any account lacks one, no model records per-account access.
+	 */
+	accountKey?: string;
+}
+
 export interface GoogleAntigravityModelManagerConfig {
-	oauthToken?: string;
+	/**
+	 * Resolves every configured Antigravity account at discovery time. Rosters
+	 * are plan-scoped (Claude 5.5 is served to some accounts only) and
+	 * authoritative, so each account's roster is fetched and the results are
+	 * unioned; reading one account's roster would prune models only its
+	 * siblings serve (#14924).
+	 *
+	 * Returns `null` to abort discovery (e.g. an account's credential failed to
+	 * refresh), keeping the previous/bundled catalog instead of caching a
+	 * partial one.
+	 */
+	resolveAccounts?: () => Promise<readonly GoogleAntigravityAccount[] | null>;
 	endpoint?: string;
 	fetch?: FetchImpl;
 }
@@ -70,17 +96,49 @@ export function googleVertexModelManagerOptions(_config?: GoogleVertexModelManag
 export function googleAntigravityModelManagerOptions(
 	config?: GoogleAntigravityModelManagerConfig,
 ): ModelManagerOptions<"google-gemini-cli"> {
-	const token = config?.oauthToken;
+	const resolveAccounts = config?.resolveAccounts;
 	return {
 		providerId: "google-antigravity",
-		...(token
+		dynamicModelsAuthoritative: providerEntry("google-antigravity")?.dynamicModelsAuthoritative === true,
+		...(resolveAccounts
 			? {
-					fetchDynamicModels: () =>
-						fetchAntigravityDiscoveryModels({
-							token,
-							endpoint: config?.endpoint,
-							fetcher: toDiscoveryFetch(config?.fetch),
-						}),
+					fetchDynamicModels: async () => {
+						const accounts = await resolveAccounts();
+						if (!accounts || accounts.length === 0) return null;
+						const fetcher = toDiscoveryFetch(config?.fetch);
+						const rosters = await Promise.all(
+							accounts.map(async account => ({
+								accountKey: account.accountKey,
+								result: await fetchAntigravityDiscoveryModels({
+									token: account.accessToken,
+									endpoint: config?.endpoint,
+									fetcher,
+								}),
+							})),
+						);
+						const catalogs: { accountKey: string | undefined; models: ModelSpec<"google-gemini-cli">[] }[] = [];
+						for (const { accountKey, result } of rosters) {
+							// A transient failure would leave the union partial; keep the previous catalog.
+							if (!result) return null;
+							if (result.rejectedStatus !== undefined) {
+								logger.warn("Antigravity model discovery skipped an account whose credential was rejected", {
+									accountKey,
+									status: result.rejectedStatus,
+								});
+								continue;
+							}
+							catalogs.push({ accountKey, models: result.models });
+						}
+						if (catalogs.length === 0) return null;
+						const tagAccess = catalogs.every(catalog => catalog.accountKey !== undefined);
+						return unionAccountCatalogs(
+							catalogs.map(({ accountKey, models }) =>
+								tagAccess && accountKey !== undefined
+									? models.map(model => ({ ...model, accountAccess: { [accountKey]: {} } }))
+									: models,
+							),
+						);
+					},
 				}
 			: undefined),
 	};
@@ -101,19 +159,19 @@ export function googleGeminiCliModelManagerOptions(
 						if (collapseTable === undefined) {
 							throw new Error("missing reviewed collapse table for google-gemini-cli");
 						}
-						const models = await fetchAntigravityDiscoveryModels({
+						const result = await fetchAntigravityDiscoveryModels({
 							token,
 							fetcher,
-							collapseTable: collapseTable,
+							collapseTable,
 						});
 						// Antigravity's fetchAvailableModels is unreachable for
 						// credentials without Antigravity entitlement (Code Assist
 						// Standard returns HTTP 403). Fall back to the account's own
 						// retrieveUserQuota list on Cloud Code Assist.
-						if (models === null) {
+						if (result === null || result.rejectedStatus !== undefined) {
 							return fetchGeminiCliQuotaModels({ token, projectId: config?.projectId, endpoint, fetcher });
 						}
-						return models
+						return result.models
 							.filter(m => classifyModel("google-gemini-cli", m.id, { lenient: true }).class === "gemini")
 							.map(m => ({
 								...m,

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

@@ -69,7 +69,7 @@ export interface OpenAIModelManagerConfig {
     baseUrl?: string;
     fetch?: FetchImpl;
 }
-export declare function openaiModelManagerOptions(config?: OpenAIModelManagerConfig): ModelManagerOptions<"openai-responses">;
+export declare function openaiModelManagerOptions(config?: OpenAIModelManagerConfig): ModelManagerOptions<Api>;
 /**
  * Re-derive the generated pro-reasoning aliases (`gpt-5.6-*-pro`) for the
  * first-party `openai` gpt-5.6 rows. Each alias inherits the base row's
@@ -154,6 +154,19 @@ export interface DeepinfraModelManagerConfig {
  * model availability at runtime is still gated on credentials by the registry.
  */
 export declare function deepinfraModelManagerOptions(config?: DeepinfraModelManagerConfig): ModelManagerOptions<"openai-completions">;
+export declare const CORALBRICKS_BASE_URL = "https://inference.coralbricks.ai/v1";
+/** CoralBricks OpenAI-compatible discovery configuration. */
+export interface CoralbricksModelManagerConfig {
+    apiKey?: string;
+    baseUrl?: string;
+    fetch?: FetchImpl;
+}
+/**
+ * Builds CoralBricks' model-discovery manager. `/v1/models` is key-protected
+ * (401 without a bearer key), so a keyless config serves only the bundled
+ * reviewed seed rows; with a key, live rows are authoritative over the bundle.
+ */
+export declare function coralbricksModelManagerOptions(config?: CoralbricksModelManagerConfig): ModelManagerOptions<"openai-completions">;
 export interface XaiModelManagerConfig {
     apiKey?: string;
     baseUrl?: string;
@@ -339,6 +352,11 @@ export interface OpenRouterModelManagerConfig {
     baseUrl?: string;
     fetch?: FetchImpl;
 }
+/**
+ * OpenRouter's Decisions API lives at `/api/alpha`, a sibling of the `/api/v1`
+ * chat root; derive it so a custom gateway base URL keeps both aligned.
+ */
+export declare function openrouterDecisionsBaseUrl(chatBaseUrl: string): string;
 export declare function openrouterModelManagerOptions(config?: OpenRouterModelManagerConfig): ModelManagerOptions<Api>;
 export interface ZenMuxModelManagerConfig {
     apiKey?: string;