@linxiraos/pi-catalog 1.1.6 → 1.1.8

package/src/models.ts

@@ -1,3 +1,4 @@
+import { classifyModel } from "./compat/taxonomy";
 import MODELS from "./models.json" with { type: "json" };
 import type { Api, KnownProvider, Model, ModelCost, TokenCost, Usage } from "./types";
 
@@ -22,7 +23,11 @@ function getProviderModels(provider: string): Map<string, Model<Api>> | undefine
 	for (const id in rawModels) {
 		// models.json rows are complete Models emitted by generate-models.ts;
 		// consuming them verbatim keeps startup allocation-free.
-		providerModels.set(id, rawModels[id as keyof typeof rawModels] as unknown as Model<Api>);
+		const row = rawModels[id as keyof typeof rawModels] as unknown as Model<Api>;
+		// Rows baked before the compat engine (and stale cache snapshots) lack
+		// `identity`; classify lazily so consumers can rely on the field.
+		row.identity ??= classifyModel(provider, id, { lenient: true });
+		providerModels.set(id, row);
 	}
 	modelRegistry.set(provider, providerModels);
 	return providerModels;

package/src/provider-models/bundled-references.ts

@@ -45,7 +45,9 @@ function getGlobalReferences(): Map<string, Model<Api>> {
 	for (const provider of getBundledProviders()) {
 		for (const model of getBundledModels(provider as Parameters<typeof getBundledModels>[0])) {
 			const candidate = model as Model<Api>;
-			if (isZeroCostXaiOAuthReference(candidate)) {
+			// ClinePass limits, pricing, and reasoning controls are gateway-specific;
+			// matching them by bare id would contaminate unrelated proxy models.
+			if (candidate.provider === "cline-pass" || isZeroCostXaiOAuthReference(candidate)) {
 				continue;
 			}
 			const existing = references.get(candidate.id);

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

@@ -0,0 +1,226 @@
+import { Effort } from "../effort";
+import type { ModelSpec, ThinkingConfig } from "../types";
+
+type ClinePassCost = ModelSpec<"openai-completions">["cost"];
+type ClinePassInput = ModelSpec<"openai-completions">["input"];
+
+export interface ClinePassModelMetadata {
+	name: string;
+	contextWindow: number;
+	maxTokens: number;
+	input: ClinePassInput;
+	cost: ClinePassCost;
+	reasoning: boolean;
+	thinking?: ThinkingConfig;
+	tier: "subscription" | "free";
+}
+
+const LOW_HIGH_MAX: ThinkingConfig = {
+	mode: "effort",
+	efforts: [Effort.Low, Effort.High, Effort.Max],
+	defaultLevel: Effort.High,
+	requiresEffort: false,
+};
+
+const HIGH_XHIGH: ThinkingConfig = {
+	mode: "effort",
+	efforts: [Effort.High, Effort.XHigh],
+	requiresEffort: false,
+};
+
+const LOW_TO_MAX: ThinkingConfig = {
+	mode: "effort",
+	efforts: [Effort.Low, Effort.Medium, Effort.High, Effort.XHigh, Effort.Max],
+	requiresEffort: false,
+};
+
+const QWEN_38_EFFORTS: ThinkingConfig = {
+	mode: "effort",
+	efforts: [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh],
+	requiresEffort: true,
+};
+
+const QWEN_37_BUDGETS: ThinkingConfig = {
+	mode: "budget",
+	efforts: [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh, Effort.Max],
+	defaultLevel: Effort.High,
+	requiresEffort: false,
+	// Cline derives these from its 10/20/50/80/95/100% effort ratios and
+	// the model's 131,072-token output ceiling, capped by the advertised
+	// 262,144-token reasoning maximum.
+	effortBudgets: {
+		[Effort.Minimal]: 13_107,
+		[Effort.Low]: 26_214,
+		[Effort.Medium]: 65_536,
+		[Effort.High]: 104_857,
+		[Effort.XHigh]: 124_518,
+		[Effort.Max]: 131_072,
+	},
+};
+
+/**
+ * ClinePass metadata published by Cline CLI 3.0.58 / @cline/llms 0.0.79.
+ * The public recommended-models endpoint remains authoritative for membership;
+ * this snapshot supplies the exact limits, plan pricing, and reasoning controls
+ * that endpoint omits, including a deterministic offline catalog floor.
+ */
+export const CLINE_PASS_MODEL_METADATA: Readonly<Record<string, ClinePassModelMetadata>> = {
+	"kimi-k3": {
+		name: "Kimi K3",
+		contextWindow: 1_048_576,
+		maxTokens: 1_048_576,
+		input: ["text", "image"],
+		cost: { input: 3, output: 15, cacheRead: 0.3, cacheWrite: 0 },
+		reasoning: true,
+		thinking: LOW_HIGH_MAX,
+		tier: "subscription",
+	},
+	"qwen3.8-max": {
+		name: "Qwen3.8 Max",
+		contextWindow: 1_000_000,
+		maxTokens: 131_072,
+		input: ["text", "image"],
+		cost: { input: 2, output: 6, cacheRead: 0.25, cacheWrite: 2.5 },
+		reasoning: true,
+		thinking: QWEN_38_EFFORTS,
+		tier: "subscription",
+	},
+	"deepseek-v4-flash": {
+		name: "DeepSeek V4 Flash",
+		contextWindow: 1_048_576,
+		maxTokens: 384_000,
+		input: ["text"],
+		cost: { input: 0.056, output: 0.112, cacheRead: 0.0112, cacheWrite: 0 },
+		reasoning: true,
+		thinking: HIGH_XHIGH,
+		tier: "subscription",
+	},
+	"glm-5.2": {
+		name: "GLM-5.2",
+		contextWindow: 1_048_576,
+		maxTokens: 131_072,
+		input: ["text"],
+		cost: { input: 0.966, output: 3.036, cacheRead: 0.1932, cacheWrite: 0 },
+		reasoning: true,
+		thinking: LOW_HIGH_MAX,
+		tier: "subscription",
+	},
+	"deepseek-v4-pro": {
+		name: "DeepSeek V4 Pro",
+		contextWindow: 1_048_576,
+		maxTokens: 384_000,
+		input: ["text"],
+		cost: { input: 0.522, output: 1.044, cacheRead: 0.0435, cacheWrite: 0 },
+		reasoning: true,
+		thinking: HIGH_XHIGH,
+		tier: "subscription",
+	},
+	"qwen3.7-max": {
+		name: "Qwen3.7 Max",
+		contextWindow: 1_000_000,
+		maxTokens: 131_072,
+		input: ["text"],
+		cost: { input: 1.475, output: 4.425, cacheRead: 0.295, cacheWrite: 1.84375 },
+		reasoning: true,
+		thinking: HIGH_XHIGH,
+		tier: "subscription",
+	},
+	"qwen3.7-plus": {
+		name: "Qwen3.7 Plus",
+		contextWindow: 1_000_000,
+		maxTokens: 131_072,
+		input: ["text", "image"],
+		cost: { input: 0.32, output: 1.28, cacheRead: 0.064, cacheWrite: 0.4 },
+		reasoning: true,
+		thinking: QWEN_37_BUDGETS,
+		tier: "subscription",
+	},
+	"minimax-m3": {
+		name: "MiniMax-M3",
+		contextWindow: 1_048_576,
+		maxTokens: 512_000,
+		input: ["text", "image"],
+		cost: { input: 0.3, output: 1.2, cacheRead: 0.06, cacheWrite: 0 },
+		reasoning: true,
+		tier: "subscription",
+	},
+	"kimi-k2.7-code": {
+		name: "Kimi K2.7 Code",
+		contextWindow: 262_144,
+		maxTokens: 262_144,
+		input: ["text", "image"],
+		cost: { input: 0.67, output: 3.4, cacheRead: 0.19, cacheWrite: 0 },
+		reasoning: true,
+		tier: "subscription",
+	},
+	"kimi-k2.6": {
+		name: "Kimi K2.6",
+		contextWindow: 262_144,
+		maxTokens: 262_144,
+		input: ["text", "image"],
+		cost: { input: 0.95, output: 4, cacheRead: 0.16, cacheWrite: 0 },
+		reasoning: true,
+		tier: "subscription",
+	},
+	"glm-5.3": {
+		name: "GLM-5.3",
+		contextWindow: 1_048_576,
+		maxTokens: 131_072,
+		input: ["text"],
+		cost: { input: 1.4, output: 4.4, cacheRead: 0.26, cacheWrite: 0 },
+		reasoning: true,
+		thinking: LOW_HIGH_MAX,
+		tier: "subscription",
+	},
+	"mimo-v2.5-pro": {
+		name: "MiMo-V2.5-Pro",
+		contextWindow: 1_050_000,
+		maxTokens: 131_072,
+		input: ["text"],
+		cost: { input: 0.435, output: 0.87, cacheRead: 0.0036, cacheWrite: 0 },
+		reasoning: true,
+		tier: "subscription",
+	},
+	"mimo-v2.5": {
+		name: "MiMo-V2.5",
+		contextWindow: 1_050_000,
+		maxTokens: 131_072,
+		input: ["text", "image"],
+		cost: { input: 0.14, output: 0.28, cacheRead: 0.0028, cacheWrite: 0 },
+		reasoning: true,
+		tier: "subscription",
+	},
+	"stealth/ox-alpha": {
+		name: "Ox Alpha (free)",
+		contextWindow: 1_048_576,
+		maxTokens: 131_072,
+		input: ["text", "image"],
+		cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+		reasoning: true,
+		thinking: LOW_TO_MAX,
+		tier: "free",
+	},
+	"deepseek/deepseek-v4-flash": {
+		name: "DeepSeek V4 Flash (free)",
+		contextWindow: 1_048_576,
+		maxTokens: 384_000,
+		input: ["text"],
+		cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+		reasoning: true,
+		thinking: HIGH_XHIGH,
+		tier: "free",
+	},
+	"poolside/laguna-s-2.1:free": {
+		name: "Laguna S 2.1 (free)",
+		contextWindow: 262_144,
+		maxTokens: 32_768,
+		input: ["text"],
+		cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+		reasoning: true,
+		tier: "free",
+	},
+};
+
+export function getClinePassModelMetadata(id: string): ClinePassModelMetadata | undefined {
+	return CLINE_PASS_MODEL_METADATA[id];
+}

package/src/provider-models/descriptors.ts

@@ -9,6 +9,7 @@ import type { ModelManagerConfig, ProviderCatalogEntry, ProviderDescriptor } fro
 import { googleModelManagerOptions, googleVertexModelManagerOptions } from "./google";
 import { ollamaCloudModelManagerOptions } from "./ollama";
 import {
+	abliterationModelManagerOptions,
 	aiandModelManagerOptions,
 	aimlApiModelManagerOptions,
 	alibabaCodingPlanModelManagerOptions,
@@ -17,6 +18,7 @@ import {
 	basetenModelManagerOptions,
 	bedrockMantleModelManagerOptions,
 	cerebrasModelManagerOptions,
+	clinePassModelManagerOptions,
 	cloudflareAiGatewayModelManagerOptions,
 	coreWeaveModelManagerOptions,
 	deepinfraModelManagerOptions,
@@ -69,6 +71,14 @@ import {
 } from "./special";
 
 export const CATALOG_PROVIDERS = [
+	{
+		id: "abliteration",
+		defaultModel: "abliterated-model",
+		envVars: ["ABLITERATION_API_KEY", "ABLIT_KEY"],
+		createModelManagerOptions: (config: ModelManagerConfig) => abliterationModelManagerOptions(config),
+		dynamicModelsAuthoritative: true,
+		catalogDiscovery: { label: "Abliteration" },
+	},
 	{
 		id: "aiand",
 		defaultModel: "moonshotai/kimi-k2.7-code",
@@ -175,6 +185,14 @@ export const CATALOG_PROVIDERS = [
 		dynamicModelsAuthoritative: true,
 		catalogDiscovery: { label: "Devin", envVars: ["DEVIN_API_KEY"], oauthProvider: "devin" },
 	},
+	{
+		id: "cline-pass",
+		defaultModel: "kimi-k3",
+		envVars: ["CLINE_API_KEY"],
+		createModelManagerOptions: (config: ModelManagerConfig) => clinePassModelManagerOptions(config),
+		dynamicModelsAuthoritative: true,
+		catalogDiscovery: { label: "ClinePass", allowUnauthenticated: true },
+	},
 	{
 		id: "firepass",
 		defaultModel: "kimi-k2.6-turbo",

package/src/provider-models/google.ts

@@ -1,10 +1,10 @@
+import { reviewedCollapseTable } from "../compat/collapse";
+import { classifyModel } from "../compat/taxonomy";
 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";
-import { GEMINI_CLI_VARIANT_COLLAPSE_TABLE } from "../variant-collapse";
 
 export interface GoogleModelManagerConfig {
 	apiKey?: string;
@@ -34,6 +34,7 @@ export interface GoogleGeminiCliModelManagerConfig {
 }
 
 const CLOUD_CODE_ASSIST_ENDPOINT = "https://cloudcode-pa.googleapis.com";
+const GEMINI_37_FLASH_CACHE_MIGRATION_MODEL_IDS = ["gemini-3.7-flash"] as const;
 
 function toDiscoveryFetch(fetchImpl: FetchImpl | undefined): typeof fetch | undefined {
 	if (!fetchImpl) {
@@ -51,6 +52,7 @@ export function googleModelManagerOptions(
 	const apiKey = config?.apiKey;
 	return {
 		providerId: "google",
+		dropCachedModelIdsOnStaticMismatch: GEMINI_37_FLASH_CACHE_MIGRATION_MODEL_IDS,
 		...(apiKey
 			? { fetchDynamicModels: () => fetchGeminiModels({ apiKey, fetch: toDiscoveryFetch(config?.fetch) }) }
 			: undefined),
@@ -58,7 +60,10 @@ export function googleModelManagerOptions(
 }
 
 export function googleVertexModelManagerOptions(_config?: GoogleVertexModelManagerConfig): ModelManagerOptions {
-	return { providerId: "google-vertex" };
+	return {
+		providerId: "google-vertex",
+		dropCachedModelIdsOnStaticMismatch: GEMINI_37_FLASH_CACHE_MIGRATION_MODEL_IDS,
+	};
 }
 
 export function googleAntigravityModelManagerOptions(
@@ -91,10 +96,14 @@ export function googleGeminiCliModelManagerOptions(
 			? {
 					fetchDynamicModels: async () => {
 						const fetcher = toDiscoveryFetch(config?.fetch);
+						const collapseTable = reviewedCollapseTable("google-gemini-cli");
+						if (collapseTable === undefined) {
+							throw new Error("missing reviewed collapse table for google-gemini-cli");
+						}
 						const models = await fetchAntigravityDiscoveryModels({
 							token,
 							fetcher,
-							collapseTable: GEMINI_CLI_VARIANT_COLLAPSE_TABLE,
+							collapseTable: collapseTable,
 						});
 						// Antigravity's fetchAvailableModels is unreachable for
 						// credentials without Antigravity entitlement (Code Assist
@@ -104,7 +113,7 @@ export function googleGeminiCliModelManagerOptions(
 							return fetchGeminiCliQuotaModels({ token, projectId: config?.projectId, endpoint, fetcher });
 						}
 						return models
-							.filter(m => isGeminiModelId(m.id))
+							.filter(m => classifyModel("google-gemini-cli", m.id, { lenient: true }).class === "gemini")
 							.map(m => ({
 								...m,
 								provider: "google-gemini-cli" as const,

package/src/provider-models/models-dev-policies.ts

@@ -1,38 +1,13 @@
+import { isExcludedModel } from "../compat/behavior";
 import type { Api, ModelSpec } from "../types";
 
-const BEDROCK_MANTLE_OPENAI_MODEL_IDS: Readonly<Record<string, true>> = {
-	"openai.gpt-5.4": true,
-	"openai.gpt-5.5": true,
-	"openai.gpt-5.6-luna": true,
-	"openai.gpt-5.6-sol": true,
-	"openai.gpt-5.6-terra": true,
-};
-
 /**
- * Remove models.dev rows that OMP cannot route successfully.
+ * Remove models.dev rows that OMP cannot route successfully; the roster
+ * policy lives in the `exclude-models` behavior rules.
  *
  * Generation and runtime refresh share this policy so a live catalog cannot
  * reintroduce selectors deliberately excluded from the bundled catalog.
  */
 export function filterModelsDevCatalogRows<TApi extends Api>(models: readonly ModelSpec<TApi>[]): ModelSpec<TApi>[] {
-	return models.filter(model => {
-		if (model.provider === "amazon-bedrock") {
-			// AWS does not document the jp. Opus 5 profile, and the openai.gpt-5.x
-			// rows are Mantle-only ids that Bedrock rejects or misroutes.
-			return model.id !== "jp.anthropic.claude-opus-5" && !BEDROCK_MANTLE_OPENAI_MODEL_IDS[model.id];
-		}
-		if (model.provider === "zai") {
-			// [1m] is a Claude Code selector convention, not an inference id.
-			return !model.id.endsWith("[1m]");
-		}
-		if (model.provider === "fireworks" || model.provider === "firepass") {
-			// Control-plane resource ids are not accepted by the inference API.
-			return !model.id.startsWith("accounts/fireworks/");
-		}
-		if (model.provider === "xiaomi" || model.provider.startsWith("xiaomi-token-plan-")) {
-			// Text-chat transports cannot serve Xiaomi's audio-only models.
-			return !model.id.includes("-tts") && !model.id.includes("-asr");
-		}
-		return true;
-	});
+	return models.filter(model => !isExcludedModel(model.provider, model.id));
 }

package/src/provider-models/ollama.ts

@@ -1,6 +1,7 @@
 import { fetchWithRetry } from "@linxiraos/pi-utils";
+import { compareRevision, parseRevision } from "../compat/revision";
+import { classifyModel } from "../compat/taxonomy";
 import { Effort } from "../effort";
-import { isGlm52ReasoningEffortModelId } from "../identity/family";
 import type { ModelManagerOptions } from "../model-manager";
 import type { FetchImpl, ModelSpec, ThinkingConfig } from "../types";
 import { discoveryFetch } from "../utils";
@@ -96,7 +97,19 @@ function getThinkingConfig(modelId: string, capabilities: string[] | undefined):
 	if (!capabilities?.includes("thinking")) {
 		return undefined;
 	}
-	if (isGlm52ReasoningEffortModelId(modelId)) {
+	const identity = classifyModel("ollama-cloud", modelId, { lenient: true });
+	const revision = identity.revision === undefined ? undefined : parseRevision(identity.revision);
+	const floor = parseRevision(identity.family === "flash" ? "5.3" : "5.2");
+	const isGlmEffortModel =
+		identity.class === "glm" &&
+		(identity.family === undefined ||
+			identity.family === "air" ||
+			identity.family === "turbo" ||
+			identity.family === "flash") &&
+		revision !== undefined &&
+		floor !== undefined &&
+		compareRevision(revision, floor) >= 0;
+	if (isGlmEffortModel) {
 		return OLLAMA_CLOUD_GLM_52_THINKING;
 	}
 	return { mode: "effort", efforts: [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High] };