@oh-my-pi/pi-catalog 18.0.4 → 18.0.5

package/CHANGELOG.md

@@ -2,6 +2,17 @@
 
 ## [Unreleased]
 
+## [18.0.5] - 2026-08-25
+
+### Added
+
+- Added built-in DeepInfra provider support (`deepinfra`, `DEEPINFRA_API_KEY`) with live model discovery, including chat models, context windows, pricing, cache-read pricing, vision input, and reasoning-effort capabilities.
+- Added built-in Yolo-Auto provider support with the flat-rate `deepseek-flash-v4` model and live model discovery.
+
+### Fixed
+
+- Fixed the Synthetic provider’s default model to use `hf:zai-org/GLM-5.2` instead of the retired `hf:zai-org/GLM-5.1`.
+
 ## [18.0.4] - 2026-08-24
 
 ### Fixed

package/dist/types/provider-models/descriptors.d.ts

@@ -97,6 +97,16 @@ export declare const CATALOG_PROVIDERS: readonly [{
         readonly envVars: readonly ["CURSOR_API_KEY"];
         readonly oauthProvider: "cursor";
     };
+}, {
+    readonly id: "deepinfra";
+    readonly defaultModel: "deepseek-ai/DeepSeek-V4-Flash-0731";
+    readonly envVars: readonly ["DEEPINFRA_API_KEY"];
+    readonly createModelManagerOptions: (config: ModelManagerConfig) => import("../index.js").ModelManagerOptions<"openai-completions", unknown>;
+    readonly dynamicModelsAuthoritative: true;
+    readonly catalogDiscovery: {
+        readonly label: "DeepInfra";
+        readonly allowUnauthenticated: true;
+    };
 }, {
     readonly id: "deepseek";
     readonly defaultModel: "deepseek-v4-pro";
@@ -361,7 +371,7 @@ export declare const CATALOG_PROVIDERS: readonly [{
     readonly dynamicModelsAuthoritative: true;
 }, {
     readonly id: "synthetic";
-    readonly defaultModel: "hf:zai-org/GLM-5.1";
+    readonly defaultModel: "hf:zai-org/GLM-5.2";
     readonly envVars: readonly ["SYNTHETIC_API_KEY"];
     readonly createModelManagerOptions: (config: ModelManagerConfig) => import("../index.js").ModelManagerOptions<"openai-completions", unknown>;
     readonly dynamicModelsAuthoritative: true;
@@ -469,6 +479,15 @@ export declare const CATALOG_PROVIDERS: readonly [{
     readonly defaultModel: "mimo-v2.5";
     readonly envVars: readonly ["XIAOMI_TOKEN_PLAN_SGP_API_KEY"];
     readonly createModelManagerOptions: (config: ModelManagerConfig) => import("../index.js").ModelManagerOptions<"openai-completions", unknown>;
+}, {
+    readonly id: "yolo-auto";
+    readonly defaultModel: "deepseek-flash-v4";
+    readonly envVars: readonly ["YOLO_AUTO_API_KEY"];
+    readonly createModelManagerOptions: (config: ModelManagerConfig) => import("../index.js").ModelManagerOptions<"openai-completions", unknown>;
+    readonly dynamicModelsAuthoritative: true;
+    readonly catalogDiscovery: {
+        readonly label: "Yolo-Auto";
+    };
 }, {
     readonly id: "zai";
     readonly defaultModel: "glm-5.3";

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

@@ -161,6 +161,19 @@ export interface NovitaModelManagerConfig {
 }
 /** Builds Novita's public model-discovery manager. */
 export declare function novitaModelManagerOptions(config?: NovitaModelManagerConfig): ModelManagerOptions<"openai-completions">;
+export declare const DEEPINFRA_BASE_URL = "https://api.deepinfra.com/v1/openai";
+/** DeepInfra OpenAI-compatible discovery configuration. */
+export interface DeepinfraModelManagerConfig {
+    apiKey?: string;
+    baseUrl?: string;
+    fetch?: FetchImpl;
+}
+/**
+ * Builds DeepInfra's model-discovery manager. The catalog endpoint is public,
+ * so discovery (and keyless `gen:models` generation) works without an API key;
+ * model availability at runtime is still gated on credentials by the registry.
+ */
+export declare function deepinfraModelManagerOptions(config?: DeepinfraModelManagerConfig): ModelManagerOptions<"openai-completions">;
 export interface XaiModelManagerConfig {
     apiKey?: string;
     baseUrl?: string;
@@ -528,6 +541,29 @@ export interface AiandModelManagerConfig {
  * pricing metadata, so discovery is authoritative over the bundled seed.
  */
 export declare function aiandModelManagerOptions(config?: AiandModelManagerConfig): ModelManagerOptions<"openai-completions">;
+/**
+ * Documented Yolo-Auto catalog (yolo-auto.com/docs, 2026-08) bundled so the
+ * provider is usable when generation and first boot have no live key. The
+ * flat-rate `/v1/models` response is authoritative once discovery runs.
+ * The compat block mirrors the provider's documented wire surface: the API
+ * speaks the generic chat template with `reasoning_effort` support and rejects
+ * the `developer` role and `store` param.
+ */
+export declare const YOLO_AUTO_STATIC_MODELS: readonly ModelSpec<"openai-completions">[];
+export interface YoloAutoModelManagerConfig {
+    apiKey?: string;
+    baseUrl?: string;
+    fetch?: FetchImpl;
+}
+/**
+ * Yolo-Auto model manager: OpenAI-compatible chat completions at the
+ * flat-rate `deepseek-flash-v4` endpoint. Live `/v1/models` discovery replaces the
+ * bundled seed once a key is stored or present in `YOLO_AUTO_API_KEY`. Models
+ * the provider adds later inherit metadata (reasoning, thinking, context)
+ * from the global bundled reference index, so new ids work without a per-id
+ * code change.
+ */
+export declare function yoloAutoModelManagerOptions(config?: YoloAutoModelManagerConfig): ModelManagerOptions<"openai-completions">;
 export interface QwenPortalModelManagerConfig {
     apiKey?: string;
     baseUrl?: string;

package/dist/types/types.d.ts

@@ -121,8 +121,8 @@ export interface Usage {
         total: number;
     };
 }
-export type OpenAIReasoningFormat = "openai" | "openrouter" | "zai" | "kimi" | "qwen" | "qwen-chat-template";
-export type OpenAIReasoningDisableMode = "omit" | "lowest-effort" | "none-effort" | "openrouter-enabled-false" | "venice-disable-thinking" | "zai-thinking-disabled" | "qwen-enable-thinking-false" | "qwen-template-false";
+export type OpenAIReasoningFormat = "openai" | "openrouter" | "zai" | "kimi" | "qwen" | "qwen-chat-template" | "chat-template";
+export type OpenAIReasoningDisableMode = "omit" | "lowest-effort" | "none-effort" | "openrouter-enabled-false" | "venice-disable-thinking" | "zai-thinking-disabled" | "qwen-enable-thinking-false" | "qwen-template-false" | "chat-template-thinking-false";
 export type OpenAIStreamMarkupHealingPattern = "kimi" | "dsml" | "qwen" | "thinking";
 /**
  * Compatibility settings for openai-completions API.

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "18.0.4",
+	"version": "18.0.5",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"homepage": "https://omp.sh",
 	"author": "Stencil Labs, Inc.",
@@ -34,11 +34,11 @@
 		"gen:proto": "bun scripts/generate-protocols.ts"
 	},
 	"dependencies": {
-		"@oh-my-pi/omptype": "18.0.4",
-		"@oh-my-pi/pi-utils": "18.0.4"
+		"@oh-my-pi/omptype": "18.0.5",
+		"@oh-my-pi/pi-utils": "18.0.5"
 	},
 	"devDependencies": {
-		"@oh-my-pi/pi-ai": "18.0.4",
+		"@oh-my-pi/pi-ai": "18.0.5",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/compat/openai.ts

@@ -94,6 +94,8 @@ function resolveReasoningDisableMode(
 			return "qwen-enable-thinking-false";
 		case "qwen-chat-template":
 			return "qwen-template-false";
+		case "chat-template":
+			return "chat-template-thinking-false";
 		default:
 			return "lowest-effort";
 	}

package/src/identity/priority.ts

@@ -22,6 +22,7 @@ const DEFAULT_MODEL_PROVIDER_ORDER = [
 	"fireworks",
 	"cerebras",
 	"baseten",
+	"deepinfra",
 	"openrouter",
 	"aimlapi",
 	"together",

package/src/model-manager.ts

@@ -492,9 +492,17 @@ function mergeDynamicModel<TApi extends Api>(existingModel: Model<TApi>, dynamic
 	// dynamic discovery also pre-applies the correct image fallback for omitted
 	// `supports.vision`, so its explicit `false` must not be OR-upgraded by the
 	// canonical bundled model.
+	// DeepInfra's discovery is authoritative (`dynamicModelsAuthoritative`) and
+	// its `vision`/`vlm` tags are the catalog's whole truth for modality. Every
+	// row shares the single DeepInfra endpoint, so `endpointChanged` never fires
+	// there: without this carve-out a model that dropped those tags would keep
+	// the bundled reference's image support and the agent would go on sending
+	// images to a now text-only route.
 	const endpointChanged = existingModel.baseUrl !== dynamicModel.baseUrl;
 	const dynamicInputAuthoritative =
-		endpointChanged || (existingModel.provider === "github-copilot" && dynamicModel.provider === "github-copilot");
+		endpointChanged ||
+		(existingModel.provider === "github-copilot" && dynamicModel.provider === "github-copilot") ||
+		(existingModel.provider === "deepinfra" && dynamicModel.provider === "deepinfra");
 	const supportsImage = dynamicInputAuthoritative
 		? dynamicModel.input.includes("image")
 		: existingModel.input.includes("image") || dynamicModel.input.includes("image");

package/src/model-thinking.ts

@@ -329,6 +329,14 @@ function getModelDefinedEfforts<TApi extends Api>(
 	spec: ModelSpec<TApi>,
 	compat: CompatOf<TApi>,
 ): readonly Effort[] | undefined {
+	if (
+		isOpenAICompatReasoningApi(spec.api) &&
+		isChatTemplateThinkingFormat(compat) &&
+		spec.thinking?.mode === "effort" &&
+		spec.thinking.efforts.length > 0
+	) {
+		return spec.thinking.efforts;
+	}
 	if (isGlm53ReasoningEffortModelId(spec.id)) {
 		// GLM-5.3+ exposes a uniform wire-exact low/high/max ladder on every
 		// host — unlike GLM-5.2, whose reasoning_effort dialect is
@@ -519,6 +527,9 @@ function isQwenTemplateReasoningEffortCompat(compat: CompatOf<Api>): boolean {
 		compat !== undefined && "qwenTemplateReasoningEffort" in compat && compat.qwenTemplateReasoningEffort === true
 	);
 }
+function isChatTemplateThinkingFormat(compat: CompatOf<Api>): boolean {
+	return compat !== undefined && "thinkingFormat" in compat && compat.thinkingFormat === "chat-template";
+}
 
 function inferDetectedEffortMap<TApi extends Api>(
 	spec: ModelSpec<TApi>,