@oh-my-pi/pi-catalog 17.4.2 → 18.0.1

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

@@ -63,13 +63,13 @@ export function resolveModelCacheProviderId(providerId: string, options: ModelCa
 		}
 		case "opencode-go":
 		case "opencode-zen": {
-			// v2: muse-spark-1.2 rows cached before the reasoning/thinking
-			// recovery carry `reasoning: false` and must be refetched.
+			// v3: gateway-first rows cached before stencil enrichment carry null
+			// limits and `reasoning: false`; use a fresh namespace so they refetch.
 			const configuredBaseUrl = options.baseUrl ?? getDefaultModelDiscoveryBaseUrl(providerId)!;
 			const trimmedBaseUrl = configuredBaseUrl.endsWith("/") ? configuredBaseUrl.slice(0, -1) : configuredBaseUrl;
 			const discoveryBaseUrl = trimmedBaseUrl.endsWith("/v1") ? trimmedBaseUrl : `${trimmedBaseUrl}/v1`;
 			const scope = `${options.apiKey ?? ""}\u0000${discoveryBaseUrl}`;
-			return `${providerId}:models-v2:${Bun.hash(scope).toString(36)}`;
+			return `${providerId}:models-v3:${Bun.hash(scope).toString(36)}`;
 		}
 		case "github-copilot": {
 			// Copilot model specs bake in the plan-specific endpoint (personal vs

package/src/provider-models/google.ts

@@ -1,5 +1,6 @@
 import { fetchAntigravityDiscoveryModels } from "../discovery/antigravity";
 import { fetchGeminiModels } from "../discovery/gemini";
+import { fetchGeminiCliQuotaModels } from "../discovery/gemini-cli";
 import { isGeminiModelId } from "../identity/family";
 import type { ModelManagerOptions } from "../model-manager";
 import type { FetchImpl } from "../types";
@@ -26,6 +27,8 @@ export interface GoogleAntigravityModelManagerConfig {
 
 export interface GoogleGeminiCliModelManagerConfig {
 	oauthToken?: string;
+	/** GCP project id required by Workspace/Standard credentials for quota discovery. */
+	projectId?: string;
 	endpoint?: string;
 	fetch?: FetchImpl;
 }
@@ -87,13 +90,18 @@ export function googleGeminiCliModelManagerOptions(
 		...(token
 			? {
 					fetchDynamicModels: async () => {
+						const fetcher = toDiscoveryFetch(config?.fetch);
 						const models = await fetchAntigravityDiscoveryModels({
 							token,
-							fetcher: toDiscoveryFetch(config?.fetch),
+							fetcher,
 							collapseTable: GEMINI_CLI_VARIANT_COLLAPSE_TABLE,
 						});
+						// 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) {
-							return null;
+							return fetchGeminiCliQuotaModels({ token, projectId: config?.projectId, endpoint, fetcher });
 						}
 						return models
 							.filter(m => isGeminiModelId(m.id))

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

@@ -2594,6 +2594,16 @@ function openCodeModelManagerOptions(
 		];
 		return hints.includes("openai-responses") ? "openai-responses" : undefined;
 	};
+	const resolveApi = (id: string, defaultApi: Api): Api => {
+		const base = openCodeBaseModelId(id);
+		return (
+			apiOverrides[id] ??
+			(base ? apiOverrides[base] : undefined) ??
+			references.get(id)?.api ??
+			fallbackApi(id, base) ??
+			defaultApi
+		);
+	};
 	return {
 		providerId,
 		cacheProviderId: resolveModelCacheProviderId(providerId, { apiKey, baseUrl: discoveryBaseUrl }),
@@ -2604,6 +2614,18 @@ function openCodeModelManagerOptions(
 		// completions after the pin shipped). Sibling-catalog drift is bounded
 		// by the 2h cache TTL instead.
 		dropCachedModelIdsOnStaticMismatch: Object.keys(apiOverrides),
+		modelsDev: {
+			fetch: () => fetchWellKnownModels(config?.fetch),
+			map: payload => {
+				if (!isRecord(payload)) return [];
+				return mapModelsDevToModels(payload, OPENCODE_MODELS_DEV_DESCRIPTORS)
+					.filter(model => model.provider === providerId)
+					.map(model => {
+						const api = resolveApi(model.id, "openai-completions");
+						return { ...model, api, baseUrl: openCodeBaseUrlForApi(api, basePath) };
+					});
+			},
+		},
 		...(apiKey && {
 			fetchDynamicModels: () =>
 				fetchOpenAICompatibleModels<Api>({
@@ -2614,16 +2636,9 @@ function openCodeModelManagerOptions(
 					mapModel: (entry, defaults) => {
 						const reference = references.get(defaults.id);
 						const name = toModelName(entry.name, reference?.name ?? defaults.name);
-						const base = openCodeBaseModelId(defaults.id);
-						// Pins win over bundled references (stale bundled routes
-						// must not stick), and a base-id pin covers its billing
-						// variants; the responses fallback covers gateway-first ids.
-						const api =
-							apiOverrides[defaults.id] ??
-							(base ? apiOverrides[base] : undefined) ??
-							reference?.api ??
-							fallbackApi(defaults.id, base) ??
-							defaults.api;
+						// Pins and bundled routing hints win over the metadata-only
+						// stencil fallback; the fallback never selects a transport.
+						const api = resolveApi(defaults.id, defaults.api);
 						const baseUrl = openCodeBaseUrlForApi(api, basePath);
 						if (isMuseSparkModelId(defaults.id)) {
 							// Gateway lists these as bare ids with no capability
@@ -6255,6 +6270,29 @@ const MODELS_DEV_PROVIDER_DESCRIPTORS_GOOGLE_VERTEX: readonly ModelsDevProviderD
 	}),
 ];
 
+const OPENCODE_MODELS_DEV_DESCRIPTORS: readonly ModelsDevProviderDescriptor[] = [
+	openAiCompletionsDescriptor("opencode", "opencode-zen", "https://opencode.ai/zen/v1", {
+		filterModel: filterActiveToolCallModels,
+		resolveApi: (modelId, raw) =>
+			resolveApiByRules(
+				modelId,
+				raw,
+				OPENCODE_ZEN_API_RESOLUTION.rules,
+				OPENCODE_ZEN_API_RESOLUTION.defaultResolution,
+			),
+	}),
+	openAiCompletionsDescriptor("opencode-go", "opencode-go", "https://opencode.ai/zen/go/v1", {
+		filterModel: filterActiveToolCallModels,
+		resolveApi: (modelId, raw) =>
+			resolveApiByRules(
+				modelId,
+				raw,
+				OPENCODE_GO_API_RESOLUTION.rules,
+				OPENCODE_GO_API_RESOLUTION.defaultResolution,
+			),
+	}),
+];
+
 const MODELS_DEV_PROVIDER_DESCRIPTORS_SPECIALIZED: readonly ModelsDevProviderDescriptor[] = [
 	// --- Azure OpenAI ---
 	// OpenAI-family models hosted on Azure, served via the Responses API. baseUrl
@@ -6276,28 +6314,8 @@ const MODELS_DEV_PROVIDER_DESCRIPTORS_SPECIALIZED: readonly ModelsDevProviderDes
 	),
 	// --- Mistral ---
 	openAiCompletionsDescriptor("mistral", "mistral", "https://api.mistral.ai/v1"),
-	// --- OpenCode Zen ---
-	openAiCompletionsDescriptor("opencode", "opencode-zen", "https://opencode.ai/zen/v1", {
-		filterModel: filterActiveToolCallModels,
-		resolveApi: (modelId, raw) =>
-			resolveApiByRules(
-				modelId,
-				raw,
-				OPENCODE_ZEN_API_RESOLUTION.rules,
-				OPENCODE_ZEN_API_RESOLUTION.defaultResolution,
-			),
-	}),
-	// --- OpenCode Go ---
-	openAiCompletionsDescriptor("opencode-go", "opencode-go", "https://opencode.ai/zen/go/v1", {
-		filterModel: filterActiveToolCallModels,
-		resolveApi: (modelId, raw) =>
-			resolveApiByRules(
-				modelId,
-				raw,
-				OPENCODE_GO_API_RESOLUTION.rules,
-				OPENCODE_GO_API_RESOLUTION.defaultResolution,
-			),
-	}),
+	// --- OpenCode Zen / Go ---
+	...OPENCODE_MODELS_DEV_DESCRIPTORS,
 	// --- GitHub Copilot ---
 	openAiCompletionsDescriptor("github-copilot", "github-copilot", COPILOT_BASE_URL, {
 		defaultContextWindow: 128000,

package/src/types.ts

@@ -157,6 +157,7 @@ export type OpenAIReasoningDisableMode =
 	| "lowest-effort"
 	| "none-effort"
 	| "openrouter-enabled-false"
+	| "venice-disable-thinking"
 	| "zai-thinking-disabled"
 	| "qwen-enable-thinking-false"
 	| "qwen-template-false";
@@ -880,6 +881,12 @@ export type ModelTokenizer =
 // Model interface for the unified model system
 export interface Model<TApi extends Api = Api> {
 	id: string;
+	/**
+	 * Whether provider-bound private-use glyphs require reversible ASCII tokenization.
+	 * Materialized by `buildModel`; request handlers read this capability instead of
+	 * inferring it from the transport API.
+	 */
+	requiresGlyphTokenization?: boolean;
 	/**
 	 * Model id to send on the wire when it differs from `id`. Used by catalog
 	 * variants that present one upstream model under several local entries —
@@ -999,6 +1006,18 @@ export interface Model<TApi extends Api = Api> {
 	 * `options.isOAuth = true` for the underlying provider call.
 	 */
 	isOAuth?: boolean;
+	/**
+	 * Amazon Bedrock Guardrail id or ARN attached to every Converse request for
+	 * this model. Set from `providers.amazon-bedrock.guardrailIdentifier`; the
+	 * streaming layer forwards it as `options.guardrailIdentifier` so accounts
+	 * that gate `bedrock:InvokeModel*` on the `bedrock:GuardrailIdentifier`
+	 * condition key stop returning an explicit deny.
+	 */
+	guardrailIdentifier?: string;
+	/** Bedrock guardrail version. Defaults to `"DRAFT"` at request time when unset. */
+	guardrailVersion?: string;
+	/** Bedrock guardrail trace verbosity. */
+	guardrailTrace?: "enabled" | "disabled" | "enabled_full";
 }
 
 /**
@@ -1007,7 +1026,7 @@ export interface Model<TApi extends Api = Api> {
  * sparse override shape and nothing is resolved yet.
  */
 export interface ModelSpec<TApi extends Api = Api>
-	extends Omit<Model<TApi>, "compat" | "compatConfig" | "supportsComputerUseConfig"> {
+	extends Omit<Model<TApi>, "compat" | "compatConfig" | "requiresGlyphTokenization" | "supportsComputerUseConfig"> {
 	/** Sparse compatibility overrides; resolved into `Model.compat` by `buildModel`. */
 	compat?: CompatConfigOf<TApi>;
 }