@gajae-code/ai 0.9.2 → 0.9.3

package/CHANGELOG.md

@@ -9,6 +9,9 @@
 
 ### Fixed
 
+- Refreshed the default Gemini CLI impersonation version to 0.50.0 so the spoofed User-Agent freshness gate passes for the 0.9.2 release.
+- Hid the non-callable `google-antigravity/gemini-3.1-pro-high` selector from bundled, dynamic, and cached Antigravity catalogs after live Cloud Code Assist calls returned HTTP 400; `google-antigravity/gemini-3.1-pro-low:high` remains the working high-thinking path.
+- Preserved Anthropic tool-use arguments supplied on `content_block_start` when no `input_json_delta` chunks follow, preventing finished tool calls from collapsing back to `{}`.
 - Refreshed the default Gemini CLI impersonation version to 0.50.0 so the spoofed User-Agent freshness gate passes for the 0.9.2 release.
 
 ## [0.9.1] - 2026-07-08

package/dist/types/model-retirements.d.ts

@@ -0,0 +1,6 @@
+export declare const RETIRED_MODEL_KEYS: readonly ["google-antigravity/gemini-3.1-pro-high"];
+export declare function isRetiredModelKey(provider: string, modelId: string): boolean;
+export declare function isRetiredModel(model: {
+    provider: string;
+    id: string;
+}): boolean;

package/dist/types/models.d.ts

@@ -1,6 +1,14 @@
-import MODELS from "./models.json";
 import type { Api, KnownProvider, Model, Usage } from "./types";
-export type GeneratedProvider = keyof typeof MODELS;
+/**
+ * Static bundled model registry loaded lazily from `models.json`.
+ *
+ * This module intentionally exposes compile-time defaults only.
+ * It does not include runtime discovery, models.dev overlays, or on-disk cache state.
+ *
+ * For runtime-aware resolution, use `createModelManager()` / `resolveProviderModels()`.
+ */
+type BundledCatalog = typeof import("./models.json");
+export type GeneratedProvider = keyof BundledCatalog;
 export declare function getBundledModel<TApi extends Api = Api>(provider: GeneratedProvider, modelId: string): Model<TApi>;
 export declare function getBundledProviders(): KnownProvider[];
 export declare function getBundledModels(provider: GeneratedProvider): Model<Api>[];
@@ -10,3 +18,4 @@ export declare function calculateCost<TApi extends Api>(model: Model<TApi>, usag
  * Returns false if either model is null or undefined.
  */
 export declare function modelsAreEqual<TApi extends Api>(a: Model<TApi> | null | undefined, b: Model<TApi> | null | undefined): boolean;
+export {};

package/dist/types/utils/discovery/antigravity.d.ts

@@ -51,6 +51,12 @@ export interface FetchAntigravityDiscoveryModelsOptions {
     signal?: AbortSignal;
     /** Optional fetch implementation override for tests. */
     fetcher?: typeof fetch;
+    /**
+     * Provider id the caller assigns to returned models. Scopes retired-selector
+     * filtering (e.g. `google-gemini-cli` reuses this helper and remaps rows).
+     * Default: `google-antigravity`.
+     */
+    targetProvider?: "google-antigravity" | "google-gemini-cli";
 }
 /**
  * Fetches discoverable Antigravity models and normalizes them into canonical model entries.

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@gajae-code/ai",
-	"version": "0.9.2",
+	"version": "0.9.3",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"homepage": "https://gajae-code.com",
 	"author": "Yeachan-Heo and Gajae Code Contributors",
@@ -40,7 +40,7 @@
 	"dependencies": {
 		"@anthropic-ai/sdk": "^0.94.0",
 		"@bufbuild/protobuf": "^2.12.0",
-		"@gajae-code/utils": "0.9.2",
+		"@gajae-code/utils": "0.9.3",
 		"openai": "^6.36.0",
 		"partial-json": "^0.1.7",
 		"zod": "4.4.3"

package/src/model-cache.ts

@@ -3,7 +3,7 @@
  * Replaces per-provider JSON files with a single cache.db.
  */
 import { Database } from "bun:sqlite";
-import { getModelDbPath } from "@gajae-code/utils";
+import { getModelDbPath } from "@gajae-code/utils/dirs";
 import type { Api, Model } from "./types";
 
 const CACHE_SCHEMA_VERSION = 3;

package/src/model-manager.ts

@@ -1,8 +1,8 @@
 import { readModelCache, writeModelCache } from "./model-cache";
+import { isRetiredModel, isRetiredModelKey } from "./model-retirements";
 import { applyGeneratedModelPolicies, enrichModelThinking } from "./model-thinking";
 import { type GeneratedProvider, getBundledModels } from "./models";
 import type { Api, Model, Provider } from "./types";
-import { isRecord } from "./utils";
 
 const DEFAULT_CACHE_TTL_MS = 2 * 60 * 60 * 1000;
 const NON_AUTHORITATIVE_RETRY_MS = 5 * 60 * 1000;
@@ -85,7 +85,14 @@ function passModelList<TApi extends Api>(value: unknown): Model<TApi>[] {
 	}
 	const out: Model<TApi>[] = [];
 	for (const item of value) {
-		if (item === null || typeof item !== "object" || typeof (item as { id: unknown }).id !== "string") {
+		if (item === null || typeof item !== "object") {
+			continue;
+		}
+		const candidate = item as { id?: unknown; provider?: unknown };
+		if (typeof candidate.id !== "string") {
+			continue;
+		}
+		if (typeof candidate.provider === "string" && isRetiredModelKey(candidate.provider, candidate.id)) {
 			continue;
 		}
 		out.push(enrichModelThinking(item as Model<TApi>));
@@ -382,7 +389,7 @@ function normalizeModelList<TApi extends Api>(value: unknown): Model<TApi>[] {
 	}
 	const models: Model<TApi>[] = [];
 	for (const item of value) {
-		if (isModelLike(item)) {
+		if (isModelLike(item) && !isRetiredModel(item)) {
 			models.push(enrichModelThinking(item as Model<TApi>));
 		}
 	}
@@ -441,6 +448,10 @@ function isModelLike(value: unknown): value is Model<Api> {
 	return true;
 }
 
+function isRecord(value: unknown): value is Record<string, unknown> {
+	return typeof value === "object" && value !== null;
+}
+
 function isModelInputArray(value: unknown): value is ("text" | "image")[] {
 	if (!Array.isArray(value) || value.length === 0) {
 		return false;

package/src/model-retirements.ts

@@ -0,0 +1,13 @@
+// Retired from advertised catalogs because Cloud Code Assist rejects live calls
+// with HTTP 400. The callable high-thinking path is gemini-3.1-pro-low:high.
+export const RETIRED_MODEL_KEYS = ["google-antigravity/gemini-3.1-pro-high"] as const;
+
+const RETIRED_MODEL_KEY_SET = new Set<string>(RETIRED_MODEL_KEYS);
+
+export function isRetiredModelKey(provider: string, modelId: string): boolean {
+	return RETIRED_MODEL_KEY_SET.has(`${provider}/${modelId}`);
+}
+
+export function isRetiredModel(model: { provider: string; id: string }): boolean {
+	return isRetiredModelKey(model.provider, model.id);
+}

package/src/models.json

@@ -10978,35 +10978,6 @@
 				]
 			}
 		},
-		"gemini-3.1-pro-high": {
-			"id": "gemini-3.1-pro-high",
-			"name": "Gemini 3.1 Pro (High) (Antigravity)",
-			"api": "google-gemini-cli",
-			"provider": "google-antigravity",
-			"baseUrl": "https://daily-cloudcode-pa.sandbox.googleapis.com",
-			"reasoning": true,
-			"input": [
-				"text",
-				"image"
-			],
-			"cost": {
-				"input": 0,
-				"output": 0,
-				"cacheRead": 0,
-				"cacheWrite": 0
-			},
-			"contextWindow": 1048576,
-			"maxTokens": 65535,
-			"thinking": {
-				"mode": "google-level",
-				"minLevel": "low",
-				"maxLevel": "high",
-				"levels": [
-					"low",
-					"high"
-				]
-			}
-		},
 		"gemini-3.1-pro-low": {
 			"id": "gemini-3.1-pro-low",
 			"name": "Gemini 3.1 Pro (Low) (Antigravity)",

package/src/models.ts

@@ -1,26 +1,42 @@
+import { createRequire } from "node:module";
+import { isRetiredModelKey } from "./model-retirements";
 import { applyGeneratedModelPolicies, enrichModelThinking } from "./model-thinking";
-import MODELS from "./models.json" with { type: "json" };
 import type { Api, KnownProvider, Model, Usage } from "./types";
 import { isClaudeForcedToolChoiceIncapableModelId } from "./utils/tool-choice-capability";
 
 /**
- * Static bundled model registry loaded from `models.json`.
+ * Static bundled model registry loaded lazily from `models.json`.
  *
  * This module intentionally exposes compile-time defaults only.
  * It does not include runtime discovery, models.dev overlays, or on-disk cache state.
  *
  * For runtime-aware resolution, use `createModelManager()` / `resolveProviderModels()`.
  */
-const providerNames = Object.keys(MODELS) as KnownProvider[];
+type BundledCatalog = typeof import("./models.json");
+
+const require = createRequire(import.meta.url);
+const COMPILED_MODELS_PATH = "./packages/ai/src/models.json";
+let bundledCatalog: BundledCatalog | undefined;
+let providerNames: KnownProvider[] | undefined;
 const providerModelRegistry: Map<string, Map<string, Model<Api>>> = new Map();
 
+function getBundledCatalog(): BundledCatalog {
+	bundledCatalog ??= require(
+		process.env.PI_COMPILED === "true" ? COMPILED_MODELS_PATH : "./models.json",
+	) as BundledCatalog;
+	return bundledCatalog;
+}
+
 function getProviderModels(provider: GeneratedProvider): Map<string, Model<Api>> | undefined {
 	const cached = providerModelRegistry.get(provider);
 	if (cached) return cached;
-	const models = MODELS[provider];
+	const models = getBundledCatalog()[provider];
 	if (!models) return undefined;
 	const providerModels = new Map<string, Model<Api>>();
 	for (const [id, model] of Object.entries(models)) {
+		if (isRetiredModelKey(provider, id)) {
+			continue;
+		}
 		providerModels.set(id, applyBundledCompatDefaults(enrichModelThinking(model as Model<Api>)));
 	}
 	providerModelRegistry.set(provider, providerModels);
@@ -52,7 +68,7 @@ function applyBundledCompatDefaults(model: Model<Api>): Model<Api> {
 	return policyModels[0] ?? normalized;
 }
 
-export type GeneratedProvider = keyof typeof MODELS;
+export type GeneratedProvider = keyof BundledCatalog;
 
 export function getBundledModel<TApi extends Api = Api>(provider: GeneratedProvider, modelId: string): Model<TApi> {
 	const providerModels = getProviderModels(provider);
@@ -62,6 +78,7 @@ export function getBundledModel<TApi extends Api = Api>(provider: GeneratedProvi
 export function getBundledProviders(): KnownProvider[] {
 	// Defensive copy: the old eager path returned a fresh Array.from(...), so
 	// callers may freely mutate their result without corrupting enumeration.
+	providerNames ??= Object.keys(getBundledCatalog()) as KnownProvider[];
 	return providerNames.slice();
 }
 

package/src/provider-models/google.ts

@@ -75,6 +75,7 @@ export function googleGeminiCliModelManagerOptions(
 						const models = await fetchAntigravityDiscoveryModels({
 							token,
 							endpoint,
+							targetProvider: "google-gemini-cli",
 						});
 						if (models === null) {
 							return null;

package/src/providers/anthropic.ts

@@ -1513,7 +1513,9 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 										partial: output,
 									});
 								} else if (block.type === "toolCall") {
-									block.arguments = parseStreamingJson(block.partialJson);
+									if (block.partialJson.trim()) {
+										block.arguments = parseStreamingJson(block.partialJson);
+									}
 									delete (block as { partialJson?: string }).partialJson;
 									stream.push({
 										type: "toolcall_end",

package/src/stream.ts

@@ -12,22 +12,15 @@ import {
 import type { BedrockOptions } from "./providers/amazon-bedrock";
 import type { AnthropicOptions } from "./providers/anthropic";
 import type { CursorOptions } from "./providers/cursor";
-import { isGitLabDuoModel, streamGitLabDuo } from "./providers/gitlab-duo";
 import type { GoogleOptions } from "./providers/google";
 import type { GoogleGeminiCliOptions } from "./providers/google-gemini-cli";
 import type { GoogleVertexOptions } from "./providers/google-vertex";
-import { isKimiModel, streamKimi } from "./providers/kimi";
 import type { OllamaChatOptions } from "./providers/ollama";
 import type { OpenAICompletionsOptions } from "./providers/openai-completions";
-import { streamPiNative } from "./providers/pi-native-client";
 // Heavy provider stream functions are imported lazily via register-builtins,
-// which wraps each provider module in a dynamic import. This keeps the
-// AWS SDK, google-auth-library, @google/genai, @bufbuild/protobuf, and
-// other provider SDKs out of the CLI startup parse graph. The
-// gitlab-duo / kimi / synthetic providers stay eager because their modules
-// export routing predicates (isGitLabDuoModel, isKimiModel, isSyntheticModel)
-// that must be callable synchronously before streaming begins, and their
-// modules are thin wrappers with no heavy SDK dependencies.
+// which wraps each provider module in a dynamic import. Thin provider routing
+// modules are also loaded lazily below by returning an outer stream and piping
+// the dynamically imported inner stream into it.
 import {
 	streamAnthropic,
 	streamAzureOpenAIResponses,
@@ -41,7 +34,6 @@ import {
 	streamOpenAICompletions,
 	streamOpenAIResponses,
 } from "./providers/register-builtins";
-import { isSyntheticModel, streamSynthetic } from "./providers/synthetic";
 import type {
 	Api,
 	AssistantMessage,
@@ -240,6 +232,45 @@ export function formatProviderCredentialHint(provider: string): string {
 	}
 	return parts.join(" ");
 }
+function pipeAssistantStream(
+	outer: AssistantMessageEventStream,
+	inner: AssistantMessageEventStream,
+	signal?: AbortSignal,
+): void {
+	void (async () => {
+		try {
+			for await (const event of inner) {
+				outer.push(event);
+				// The inner provider stream owns abort semantics (it receives the
+				// same signal), but stop forwarding as soon as the consumer
+				// aborted so a misbehaving inner stream cannot keep the pipe
+				// buffering events indefinitely.
+				if (signal?.aborted && !outer.done) {
+					outer.end(await inner.result());
+					return;
+				}
+			}
+			if (!outer.done) outer.end(await inner.result());
+		} catch (error) {
+			outer.fail(error);
+		}
+	})();
+}
+
+function streamFromLazyImport(
+	createInner: () => Promise<AssistantMessageEventStream>,
+	signal?: AbortSignal,
+): AssistantMessageEventStream {
+	const outer = new AssistantMessageEventStream();
+	void (async () => {
+		try {
+			pipeAssistantStream(outer, await createInner(), signal);
+		} catch (error) {
+			outer.fail(error);
+		}
+	})();
+	return outer;
+}
 
 /**
  * Build an actionable "missing API key" error for a provider, used by the
@@ -262,15 +293,21 @@ export function stream<TApi extends Api>(
 		return customApiProvider.stream(model, context, options as StreamOptions);
 	}
 
-	if (isGitLabDuoModel(model)) {
+	if (model.provider === "gitlab-duo") {
 		const apiKey = (options as StreamOptions | undefined)?.apiKey || getEnvApiKey(model.provider);
 		if (!apiKey) {
 			throw new Error(formatMissingApiKeyError(model.provider));
 		}
-		return streamGitLabDuo(model, context, {
-			...(options as SimpleStreamOptions | undefined),
-			apiKey,
-		});
+		return streamFromLazyImport(
+			async () => {
+				const { streamGitLabDuo } = await import("./providers/gitlab-duo");
+				return streamGitLabDuo(model, context, {
+					...(options as SimpleStreamOptions | undefined),
+					apiKey,
+				});
+			},
+			(options as StreamOptions | undefined)?.signal,
+		);
 	}
 
 	// Vertex AI uses Application Default Credentials, not API keys
@@ -446,7 +483,10 @@ export function streamSimple<TApi extends Api>(
 	// extension-registered APIs can't accidentally override a configured
 	// pi-native transport.
 	if (model.transport === "pi-native") {
-		return streamPiNative(model, context, options);
+		return streamFromLazyImport(async () => {
+			const { streamPiNative } = await import("./providers/pi-native-client");
+			return streamPiNative(model, context, options);
+		}, options?.signal);
 	}
 
 	// Check custom API registry (extension-provided APIs)
@@ -471,31 +511,40 @@ export function streamSimple<TApi extends Api>(
 	}
 
 	// GitLab Duo - wraps Anthropic/OpenAI behind GitLab AI Gateway direct access tokens
-	if (isGitLabDuoModel(model)) {
-		return streamGitLabDuo(model, context, {
-			...options,
-			apiKey,
-		});
+	if (model.provider === "gitlab-duo") {
+		return streamFromLazyImport(async () => {
+			const { streamGitLabDuo } = await import("./providers/gitlab-duo");
+			return streamGitLabDuo(model, context, {
+				...options,
+				apiKey,
+			});
+		}, options?.signal);
 	}
 
 	// Kimi Code - route to dedicated handler that wraps OpenAI or Anthropic API
-	if (isKimiModel(model)) {
-		// Pass raw SimpleStreamOptions - streamKimi handles mapping internally
-		return streamKimi(model as Model<"openai-completions">, context, {
-			...options,
-			apiKey,
-			format: options?.kimiApiFormat ?? "anthropic",
-		});
+	if (model.provider === "kimi-code") {
+		return streamFromLazyImport(async () => {
+			const { streamKimi } = await import("./providers/kimi");
+			// Pass raw SimpleStreamOptions - streamKimi handles mapping internally
+			return streamKimi(model as Model<"openai-completions">, context, {
+				...options,
+				apiKey,
+				format: options?.kimiApiFormat ?? "anthropic",
+			});
+		}, options?.signal);
 	}
 
 	// Synthetic - route to dedicated handler that wraps OpenAI or Anthropic API
-	if (isSyntheticModel(model)) {
-		// Pass raw SimpleStreamOptions - streamSynthetic handles mapping internally
-		return streamSynthetic(model as Model<"openai-completions">, context, {
-			...options,
-			apiKey,
-			format: options?.syntheticApiFormat ?? "openai", // Default to OpenAI format
-		});
+	if (model.provider === "synthetic") {
+		return streamFromLazyImport(async () => {
+			const { streamSynthetic } = await import("./providers/synthetic");
+			// Pass raw SimpleStreamOptions - streamSynthetic handles mapping internally
+			return streamSynthetic(model as Model<"openai-completions">, context, {
+				...options,
+				apiKey,
+				format: options?.syntheticApiFormat ?? "openai", // Default to OpenAI format
+			});
+		}, options?.signal);
 	}
 
 	const providerOptions = mapOptionsForApi(model, options, apiKey);

package/src/utils/discovery/antigravity.ts

@@ -1,7 +1,7 @@
 import * as z from "zod/v4";
+import { isRetiredModelKey } from "../../model-retirements";
 import { getAntigravityUserAgent } from "../../providers/google-gemini-headers";
 import type { Model } from "../../types";
-import { toPositiveNumber } from "../../utils";
 
 const DEFAULT_ANTIGRAVITY_DISCOVERY_ENDPOINTS = [
 	"https://daily-cloudcode-pa.googleapis.com",
@@ -162,6 +162,12 @@ export interface FetchAntigravityDiscoveryModelsOptions {
 	signal?: AbortSignal;
 	/** Optional fetch implementation override for tests. */
 	fetcher?: typeof fetch;
+	/**
+	 * Provider id the caller assigns to returned models. Scopes retired-selector
+	 * filtering (e.g. `google-gemini-cli` reuses this helper and remaps rows).
+	 * Default: `google-antigravity`.
+	 */
+	targetProvider?: "google-antigravity" | "google-gemini-cli";
 }
 
 /**
@@ -174,6 +180,7 @@ export async function fetchAntigravityDiscoveryModels(
 	options: FetchAntigravityDiscoveryModelsOptions,
 ): Promise<Model<"google-gemini-cli">[] | null> {
 	const fetcher = options.fetcher ?? fetch;
+	const targetProvider = options.targetProvider ?? "google-antigravity";
 	const endpoints = options.endpoint
 		? [trimTrailingSlashes(options.endpoint)]
 		: DEFAULT_ANTIGRAVITY_DISCOVERY_ENDPOINTS.map(trimTrailingSlashes);
@@ -214,7 +221,7 @@ export async function fetchAntigravityDiscoveryModels(
 		const models: Model<"google-gemini-cli">[] = [];
 
 		for (const [modelId, model] of Object.entries(parsed.models ?? {})) {
-			if (ANTIGRAVITY_DISCOVERY_DENYLIST.has(modelId)) {
+			if (ANTIGRAVITY_DISCOVERY_DENYLIST.has(modelId) || isRetiredModelKey(targetProvider, modelId)) {
 				continue;
 			}
 			if (model.isInternal === true) {
@@ -256,6 +263,13 @@ function parseAntigravityDiscoveryResponse(value: unknown): AntigravityDiscovery
 	return parsed.data;
 }
 
+function toPositiveNumber(value: unknown, fallback: number): number {
+	if (typeof value !== "number" || !Number.isFinite(value) || value <= 0) {
+		return fallback;
+	}
+	return value;
+}
+
 function trimTrailingSlashes(value: string): string {
 	return value.replace(/\/+$/, "");
 }