@gajae-code/ai 0.18.10 → 0.18.12

package/src/bedrock-claude-cache-policy.d.ts

@@ -10,7 +10,7 @@ export type BedrockClaudeGeneration = {
  *  - family-first (3.x era): anthropic.claude-3-5-haiku-20241022-v1:0
  *  - kind-first (4+ era):    anthropic.claude-opus-4-20250514-v1:0,
  *                            anthropic.claude-haiku-4-5-20251001-v1:0
- * Cross-region profiles (us./eu./au./jp./apac./global. prefixes) and inference-profile
+ * Cross-region profiles (us./eu./au./jp./apac./global./in. prefixes) and inference-profile
  * ARNs keep the canonical model id as their final path segment.
  */
 export declare function parseBedrockClaudeGeneration(id: string): BedrockClaudeGeneration | undefined;

package/src/bedrock-claude-cache-policy.ts

@@ -20,13 +20,13 @@ function extractBedrockModelId(id: string): string | undefined {
  *  - family-first (3.x era): anthropic.claude-3-5-haiku-20241022-v1:0
  *  - kind-first (4+ era):    anthropic.claude-opus-4-20250514-v1:0,
  *                            anthropic.claude-haiku-4-5-20251001-v1:0
- * Cross-region profiles (us./eu./au./jp./apac./global. prefixes) and inference-profile
+ * Cross-region profiles (us./eu./au./jp./apac./global./in. prefixes) and inference-profile
  * ARNs keep the canonical model id as their final path segment.
  */
 export function parseBedrockClaudeGeneration(id: string): BedrockClaudeGeneration | undefined {
 	const modelId = extractBedrockModelId(id);
 	if (modelId === undefined) return undefined;
-	const prefix = "(?:(?:us|eu|au|jp|apac|global)\\.)?anthropic\\.claude-";
+	const prefix = "(?:(?:us|eu|au|jp|apac|global|in)\\.)?anthropic\\.claude-";
 	const component = "(?:0|[1-9]\\d?)";
 	const suffix = "(?:-(?:[a-z][a-z0-9]*(?::[a-z0-9]+)?|\\d{8}))*";
 	const familyFirst = new RegExp(`^${prefix}([1-9]\\d?)(?:-(${component}))?-([a-z][a-z0-9]*)${suffix}$`).exec(modelId);

package/src/model-cache.ts

@@ -6,7 +6,7 @@ import { Database } from "bun:sqlite";
 import { getModelDbPath } from "@gajae-code/utils/dirs";
 import type { Api, Model } from "./types";
 
-const CACHE_SCHEMA_VERSION = 5;
+const CACHE_SCHEMA_VERSION = 6;
 
 interface CacheRow {
 	provider_id: string;
@@ -16,6 +16,7 @@ interface CacheRow {
 	static_fingerprint: string;
 	dynamic_model_ids: string | null;
 	dynamic_model_provenance: string | null;
+	discovery_user_agent: string | null;
 	models: string;
 }
 
@@ -38,6 +39,8 @@ interface CacheEntry<TApi extends Api = Api> {
 	/** IDs returned by the authoritative dynamic provider catalog, when retained. */
 	dynamicModelIds: string[] | undefined;
 	dynamicModelProvenance: string | undefined;
+	/** User-Agent used during discovery; invalidates cache if it changes. */
+	discoveryUserAgent: string | undefined;
 }
 
 let sharedDb: Database | null = null;
@@ -63,6 +66,7 @@ function getDb(dbPath?: string): Database {
 			static_fingerprint TEXT NOT NULL DEFAULT '',
 			dynamic_model_ids TEXT,
 			dynamic_model_provenance TEXT,
+			discovery_user_agent TEXT,
 			models TEXT NOT NULL
 		)
 	`);
@@ -84,7 +88,13 @@ export function closeModelCache(dbPath?: string): boolean {
 }
 
 function migrateCacheSchema(db: Database): void {
-	const columns = db.prepare("PRAGMA table_info(model_cache)").all() as TableInfoRow[];
+	const tableInfo = db.query("PRAGMA table_info(model_cache)");
+	let columns: TableInfoRow[];
+	try {
+		columns = tableInfo.all() as TableInfoRow[];
+	} finally {
+		tableInfo.finalize();
+	}
 	if (!columns.some(column => column.name === "static_fingerprint")) {
 		db.run("ALTER TABLE model_cache ADD COLUMN static_fingerprint TEXT NOT NULL DEFAULT ''");
 	}
@@ -94,7 +104,10 @@ function migrateCacheSchema(db: Database): void {
 	if (!columns.some(column => column.name === "dynamic_model_provenance")) {
 		db.run("ALTER TABLE model_cache ADD COLUMN dynamic_model_provenance TEXT");
 	}
-	db.run("UPDATE model_cache SET version = ? WHERE version IN (2, 3, 4)", [CACHE_SCHEMA_VERSION]);
+	if (!columns.some(column => column.name === "discovery_user_agent")) {
+		db.run("ALTER TABLE model_cache ADD COLUMN discovery_user_agent TEXT");
+	}
+	db.run("UPDATE model_cache SET version = ? WHERE version IN (2, 3, 4, 5)", [CACHE_SCHEMA_VERSION]);
 }
 
 export function readModelCache<TApi extends Api>(
@@ -102,13 +115,24 @@ export function readModelCache<TApi extends Api>(
 	ttlMs: number,
 	now: () => number,
 	dbPath?: string,
+	expectedDiscoveryUserAgent?: string,
 ): CacheEntry<TApi> | null {
 	try {
 		const db = getDb(dbPath);
-		const row = db.query<CacheRow, [string]>("SELECT * FROM model_cache WHERE provider_id = ?").get(providerId);
+		const select = db.query<CacheRow, [string]>("SELECT * FROM model_cache WHERE provider_id = ?");
+		let row: CacheRow | null;
+		try {
+			row = select.get(providerId);
+		} finally {
+			select.finalize();
+		}
 		if (!row || row.version !== CACHE_SCHEMA_VERSION) {
 			return null;
 		}
+		// Invalidate cache if discovery User-Agent has changed
+		if (expectedDiscoveryUserAgent !== undefined && row.discovery_user_agent !== expectedDiscoveryUserAgent) {
+			return null;
+		}
 		const models = JSON.parse(row.models) as Model<TApi>[];
 		const ageMs = now() - row.updated_at;
 		const fresh = Number.isFinite(ageMs) && ageMs >= 0 && ageMs <= ttlMs;
@@ -120,6 +144,7 @@ export function readModelCache<TApi extends Api>(
 			staticFingerprint: row.static_fingerprint ?? "",
 			dynamicModelIds: row.dynamic_model_ids === null ? undefined : (JSON.parse(row.dynamic_model_ids) as string[]),
 			dynamicModelProvenance: row.dynamic_model_provenance ?? undefined,
+			discoveryUserAgent: row.discovery_user_agent ?? undefined,
 		};
 	} catch {
 		return null;
@@ -135,12 +160,13 @@ export function writeModelCache<TApi extends Api>(
 	dbPath?: string,
 	dynamicModelIds?: readonly string[],
 	dynamicModelProvenance?: string,
+	discoveryUserAgent?: string,
 ): void {
 	try {
 		const db = getDb(dbPath);
 		db.run(
-			`INSERT OR REPLACE INTO model_cache (provider_id, version, updated_at, authoritative, static_fingerprint, dynamic_model_ids, dynamic_model_provenance, models)
-			 VALUES (?, ?, ?, ?, ?, ?, ?, ?)`,
+			`INSERT OR REPLACE INTO model_cache (provider_id, version, updated_at, authoritative, static_fingerprint, dynamic_model_ids, dynamic_model_provenance, discovery_user_agent, models)
+			 VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)`,
 			[
 				providerId,
 				CACHE_SCHEMA_VERSION,
@@ -149,6 +175,7 @@ export function writeModelCache<TApi extends Api>(
 				staticFingerprint,
 				dynamicModelIds === undefined ? null : JSON.stringify(dynamicModelIds),
 				dynamicModelProvenance ?? null,
+				discoveryUserAgent ?? null,
 				JSON.stringify(models),
 			],
 		);
@@ -166,11 +193,12 @@ export function insertModelCacheIfAbsent<TApi extends Api>(
 	dbPath?: string,
 	dynamicModelIds?: readonly string[],
 	dynamicModelProvenance?: string,
+	discoveryUserAgent?: string,
 ): boolean {
 	try {
 		const result = getDb(dbPath).run(
-			`INSERT INTO model_cache (provider_id, version, updated_at, authoritative, static_fingerprint, dynamic_model_ids, dynamic_model_provenance, models)
-			 VALUES (?, ?, ?, ?, ?, ?, ?, ?)
+			`INSERT INTO model_cache (provider_id, version, updated_at, authoritative, static_fingerprint, dynamic_model_ids, dynamic_model_provenance, discovery_user_agent, models)
+			 VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)
 			 ON CONFLICT(provider_id) DO NOTHING`,
 			[
 				providerId,
@@ -180,6 +208,7 @@ export function insertModelCacheIfAbsent<TApi extends Api>(
 				staticFingerprint,
 				dynamicModelIds === undefined ? null : JSON.stringify(dynamicModelIds),
 				dynamicModelProvenance ?? null,
+				discoveryUserAgent ?? null,
 				JSON.stringify(models),
 			],
 		);
@@ -195,6 +224,7 @@ export function updateModelCacheIfUnchanged<TApi extends Api>(
 	expectedDynamicModelIds: readonly string[] | undefined,
 	expectedDynamicModelProvenance: string | undefined,
 	expectedModels: readonly Model<TApi>[],
+	expectedDiscoveryUserAgent: string | undefined,
 	updatedAt: number,
 	models: Model<TApi>[],
 	authoritative: boolean,
@@ -202,30 +232,35 @@ export function updateModelCacheIfUnchanged<TApi extends Api>(
 	dbPath?: string,
 	dynamicModelIds?: readonly string[],
 	dynamicModelProvenance?: string,
+	discoveryUserAgent?: string,
 ): boolean {
 	try {
 		const expectedIds = expectedDynamicModelIds === undefined ? null : JSON.stringify(expectedDynamicModelIds);
 		const provenance = expectedDynamicModelProvenance ?? null;
+		const expectedUserAgent = expectedDiscoveryUserAgent ?? null;
 		const nextIds = dynamicModelIds === undefined ? null : JSON.stringify(dynamicModelIds);
 		const nextProvenance = dynamicModelProvenance ?? null;
 		const expectedModelsJson = JSON.stringify(expectedModels);
 		const result = getDb(dbPath).run(
 			`UPDATE model_cache
 			 SET updated_at = ?, authoritative = ?, static_fingerprint = ?, dynamic_model_ids = ?,
-			     dynamic_model_provenance = ?, models = ?
+			     dynamic_model_provenance = ?, discovery_user_agent = ?, models = ?
 			 WHERE provider_id = ? AND updated_at = ?
-			   AND dynamic_model_ids IS ? AND dynamic_model_provenance IS ? AND models = ?`,
+			   AND dynamic_model_ids IS ? AND dynamic_model_provenance IS ?
+			   AND discovery_user_agent IS ? AND models = ?`,
 			[
 				updatedAt,
 				authoritative ? 1 : 0,
 				staticFingerprint,
 				nextIds,
 				nextProvenance,
+				discoveryUserAgent ?? null,
 				JSON.stringify(models),
 				providerId,
 				expectedUpdatedAt,
 				expectedIds,
 				provenance,
+				expectedUserAgent,
 				expectedModelsJson,
 			],
 		);

package/src/model-manager.ts

@@ -60,6 +60,8 @@ export interface ModelManagerOptions<TApi extends Api = Api, TModelsDevPayload =
 	canPublishCache?: () => boolean;
 	/** Credential-and-endpoint identity required to reuse dynamic catalog IDs. */
 	cacheDynamicModelProvenance?: string;
+	/** Discovery User-Agent (e.g., for Antigravity): cache invalidates on change. */
+	discoveryUserAgent?: string;
 }
 
 /**
@@ -146,7 +148,13 @@ export async function resolveProviderModels<TApi extends Api = Api, TModelsDevPa
 	const provenance = options.cacheDynamicModelProvenance;
 	const now = options.now ?? Date.now;
 	const ttlMs = options.cacheTtlMs ?? DEFAULT_CACHE_TTL_MS;
-	const legacyCache = readModelCache<TApi>(options.providerId, ttlMs, now, options.cacheDbPath);
+	const legacyCache = readModelCache<TApi>(
+		options.providerId,
+		ttlMs,
+		now,
+		options.cacheDbPath,
+		options.discoveryUserAgent,
+	);
 	if (
 		strategy !== "online-if-uncached" ||
 		typeof options.fetchDynamicModels !== "function" ||
@@ -185,7 +193,7 @@ async function resolveProviderModelsUncoalesced<TApi extends Api = Api, TModelsD
 	const staticModels = passModelList<TApi>(
 		options.staticModels ?? getBundledModels(options.providerId as GeneratedProvider),
 	);
-	const cache = readModelCache<TApi>(options.providerId, ttlMs, now, dbPath);
+	const cache = readModelCache<TApi>(options.providerId, ttlMs, now, dbPath, options.discoveryUserAgent);
 	const dynamicFetcher = options.fetchDynamicModels;
 	const hasDynamicFetcher = typeof dynamicFetcher === "function";
 	const cacheDynamicModelIdsCurrent =
@@ -262,6 +270,7 @@ async function resolveProviderModelsUncoalesced<TApi extends Api = Api, TModelsD
 				dbPath,
 				cache.dynamicModelIds,
 				cache.dynamicModelProvenance,
+				options.discoveryUserAgent,
 			);
 		}
 		return {
@@ -315,6 +324,7 @@ async function resolveProviderModelsUncoalesced<TApi extends Api = Api, TModelsD
 					dbPath,
 					dynamicModels.map(model => model.id),
 					options.cacheDynamicModelProvenance,
+					options.discoveryUserAgent,
 				);
 			}
 		} else {
@@ -322,10 +332,9 @@ async function resolveProviderModelsUncoalesced<TApi extends Api = Api, TModelsD
 			// stale state remains visible while retry backoff still applies. Re-read
 			// the current row and trust only that row's provenance and dynamic IDs,
 			// not the initial cacheProvenanceMismatch snapshot. A concurrent writer
-			// may have replaced the provider row; if the latest row is unbound or
-			// belongs to another context, use no latest fallback and do not
-			// overwrite or downgrade it.
-			const latestCache = readModelCache<TApi>(options.providerId, ttlMs, now, dbPath);
+			// may have replaced the provider row. This reread filters mismatched
+			// discovery User-Agents, while the checks below enforce dynamic provenance.
+			const latestCache = readModelCache<TApi>(options.providerId, ttlMs, now, dbPath, options.discoveryUserAgent);
 			const latestCacheMatchesCurrentContext =
 				options.cacheDynamicModelProvenance === undefined
 					? latestCache !== null && latestCache.dynamicModelIds === undefined
@@ -353,6 +362,7 @@ async function resolveProviderModelsUncoalesced<TApi extends Api = Api, TModelsD
 						latestCache.dynamicModelIds,
 						latestCache.dynamicModelProvenance,
 						latestCache.models,
+						latestCache.discoveryUserAgent,
 						now(),
 						snapshotModels,
 						false,
@@ -360,6 +370,7 @@ async function resolveProviderModelsUncoalesced<TApi extends Api = Api, TModelsD
 						dbPath,
 						options.cacheDynamicModelProvenance === undefined ? undefined : [],
 						options.cacheDynamicModelProvenance,
+						options.discoveryUserAgent,
 					);
 					if (!updated) cacheModels = [];
 				} else if (latestCache == null) {
@@ -372,6 +383,7 @@ async function resolveProviderModelsUncoalesced<TApi extends Api = Api, TModelsD
 						dbPath,
 						options.cacheDynamicModelProvenance === undefined ? undefined : [],
 						options.cacheDynamicModelProvenance,
+						options.discoveryUserAgent,
 					);
 					if (!inserted) cacheModels = [];
 				}

package/src/model-thinking.ts

@@ -95,7 +95,7 @@ type SemVer = {
 	patch: number;
 };
 
-type GeminiKind = "pro" | "flash";
+type GeminiKind = "pro" | "flash" | "argon";
 type AnthropicKind = "opus" | "sonnet" | "fable";
 type OpenAIVariant =
 	| "astra"
@@ -328,6 +328,23 @@ export function modelSupportsReasoningControl<TApi extends Api>(
 export function applyGeneratedModelPolicies(models: ApiModel<Api>[]): void {
 	for (let index = 0; index < models.length; index++) {
 		const source = models[index]!;
+		const inputCost = nonNegativeFinitePrice(source.cost.input);
+		const outputCost = nonNegativeFinitePrice(source.cost.output);
+		const cacheReadCost = nonNegativeFinitePrice(source.cost.cacheRead);
+		const cacheWriteCost = nonNegativeFinitePrice(source.cost.cacheWrite);
+		if (
+			inputCost !== source.cost.input ||
+			outputCost !== source.cost.output ||
+			cacheReadCost !== source.cost.cacheRead ||
+			cacheWriteCost !== source.cost.cacheWrite
+		) {
+			source.cost = {
+				input: inputCost,
+				output: outputCost,
+				cacheRead: cacheReadCost,
+				cacheWrite: cacheWriteCost,
+			};
+		}
 		if (source.provider === "xai" && (source.id === "grok-4.5" || source.id === "grok-4.6")) {
 			source.reasoning = true;
 		}
@@ -359,6 +376,10 @@ export function applyGeneratedModelPolicies(models: ApiModel<Api>[]): void {
 	}
 }
 
+function nonNegativeFinitePrice(value: number): number {
+	return Number.isFinite(value) && value >= 0 ? value : 0;
+}
+
 /**
  * Link OpenAI model variants to their context promotion targets.
  *
@@ -729,6 +750,17 @@ function applyGeneratedModelPolicy(model: ApiModel<Api>): void {
 	) {
 		model.contextWindow = 1_000_000;
 	}
+	// Google Gemini 4 Argon supports up to 65,536 output tokens across provider
+	// aliases. Strip catalog provider prefixes and OpenRouter's batch suffix.
+	// https://ai.google.dev/gemini-api/docs/models/gemini-4-argon
+	const canonicalId = model.id
+		.toLowerCase()
+		.split("/")
+		.at(-1)
+		?.replace(/:batch$/, "");
+	if (canonicalId === "gemini-4-argon") {
+		model.maxTokens = Math.min(model.maxTokens, 65_536);
+	}
 }
 
 function scrubGeneratedModelName(name: string): string {
@@ -979,6 +1011,8 @@ function inferGeminiSupportedEfforts(model: GeminiModel): readonly Effort[] {
 	if (model.kind === "pro") {
 		return GEMINI_3_PRO_EFFORTS;
 	}
+	// Gemini 3.7+ Flash and Gemini 4 Argon accept low/medium/high only.
+	// https://ai.google.dev/gemini-api/docs/models/gemini-4-argon
 	if (semverGte(model.version, "3.7")) {
 		return GEMINI_3_7_FLASH_EFFORTS;
 	}
@@ -1054,11 +1088,7 @@ function inferThinkingControlMode<TApi extends Api>(
 		case "google-generative-ai":
 		case "google-gemini-cli":
 		case "google-vertex":
-			return parsedModel.family === "gemini" &&
-				semverGte(parsedModel.version, "3.0") &&
-				parsedModel.version.major === 3
-				? "google-level"
-				: "budget";
+			return parsedModel.family === "gemini" && semverGte(parsedModel.version, "3.0") ? "google-level" : "budget";
 
 		case "anthropic-messages":
 			if (parsedModel.family === "anthropic") {
@@ -1104,7 +1134,7 @@ function parseGeminiModel(modelId: string): GeminiModel | null {
 	if (modelId.endsWith(GEMINI_SUFFIX)) {
 		modelId = modelId.slice(0, -GEMINI_SUFFIX.length);
 	}
-	const match = /gemini-(\d+(?:\.\d+){0,2})-(pro|flash)\b/.exec(modelId);
+	const match = /gemini-(\d+(?:\.\d+){0,2})-(pro|flash|argon)\b/.exec(modelId);
 	if (!match) {
 		return null;
 	}