@sayknow-cli/ai 0.3.11 → 0.3.13

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@sayknow-cli/ai",
-	"version": "0.3.11",
+	"version": "0.3.13",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"homepage": "https://sayknow-cli.com",
 	"author": "jaybeyond",
@@ -27,7 +27,7 @@
 		"api"
 	],
 	"main": "./src/index.ts",
-	"types": "./dist/types/index.d.ts",
+	"types": "./src/index.ts",
 	"bin": {
 		"pi-ai": "./src/cli.ts"
 	},
@@ -43,7 +43,7 @@
 	"dependencies": {
 		"@anthropic-ai/sdk": "^0.94.0",
 		"@bufbuild/protobuf": "^2.12.0",
-		"@sayknow-cli/utils": "0.3.11",
+		"@sayknow-cli/utils": "0.3.13",
 		"openai": "^6.36.0",
 		"partial-json": "^0.1.7",
 		"zod": "4.4.3"
@@ -57,88 +57,87 @@
 	"files": [
 		"src",
 		"README.md",
-		"CHANGELOG.md",
-		"dist/types"
+		"CHANGELOG.md"
 	],
 	"exports": {
 		".": {
-			"types": "./dist/types/index.d.ts",
+			"types": "./src/index.ts",
 			"import": "./src/index.ts"
 		},
 		"./*": {
-			"types": "./dist/types/*.d.ts",
+			"types": "./src/*.ts",
 			"import": "./src/*.ts"
 		},
 		"./auth-broker": {
-			"types": "./dist/types/auth-broker/index.d.ts",
+			"types": "./src/auth-broker/index.ts",
 			"import": "./src/auth-broker/index.ts"
 		},
 		"./auth-broker/*": {
-			"types": "./dist/types/auth-broker/*.d.ts",
+			"types": "./src/auth-broker/*.ts",
 			"import": "./src/auth-broker/*.ts"
 		},
 		"./auth-gateway": {
-			"types": "./dist/types/auth-gateway/index.d.ts",
+			"types": "./src/auth-gateway/index.ts",
 			"import": "./src/auth-gateway/index.ts"
 		},
 		"./auth-gateway/*": {
-			"types": "./dist/types/auth-gateway/*.d.ts",
+			"types": "./src/auth-gateway/*.ts",
 			"import": "./src/auth-gateway/*.ts"
 		},
 		"./models.json": {
-			"types": "./dist/types/models.json.d.d.ts",
+			"types": "./src/models.json.d.ts",
 			"import": "./src/models.json"
 		},
 		"./provider-models": {
-			"types": "./dist/types/provider-models/index.d.ts",
+			"types": "./src/provider-models/index.ts",
 			"import": "./src/provider-models/index.ts"
 		},
 		"./provider-models/*": {
-			"types": "./dist/types/provider-models/*.d.ts",
+			"types": "./src/provider-models/*.ts",
 			"import": "./src/provider-models/*.ts"
 		},
 		"./providers/*": {
-			"types": "./dist/types/providers/*.d.ts",
+			"types": "./src/providers/*.ts",
 			"import": "./src/providers/*.ts"
 		},
 		"./providers/cursor/gen/*": {
-			"types": "./dist/types/providers/cursor/gen/*.d.ts",
+			"types": "./src/providers/cursor/gen/*.ts",
 			"import": "./src/providers/cursor/gen/*.ts"
 		},
 		"./providers/openai-codex/*": {
-			"types": "./dist/types/providers/openai-codex/*.d.ts",
+			"types": "./src/providers/openai-codex/*.ts",
 			"import": "./src/providers/openai-codex/*.ts"
 		},
 		"./usage/*": {
-			"types": "./dist/types/usage/*.d.ts",
+			"types": "./src/usage/*.ts",
 			"import": "./src/usage/*.ts"
 		},
 		"./utils/*": {
-			"types": "./dist/types/utils/*.d.ts",
+			"types": "./src/utils/*.ts",
 			"import": "./src/utils/*.ts"
 		},
 		"./utils/discovery": {
-			"types": "./dist/types/utils/discovery/index.d.ts",
+			"types": "./src/utils/discovery/index.ts",
 			"import": "./src/utils/discovery/index.ts"
 		},
 		"./utils/discovery/*": {
-			"types": "./dist/types/utils/discovery/*.d.ts",
+			"types": "./src/utils/discovery/*.ts",
 			"import": "./src/utils/discovery/*.ts"
 		},
 		"./utils/oauth": {
-			"types": "./dist/types/utils/oauth/index.d.ts",
+			"types": "./src/utils/oauth/index.ts",
 			"import": "./src/utils/oauth/index.ts"
 		},
 		"./utils/oauth/*": {
-			"types": "./dist/types/utils/oauth/*.d.ts",
+			"types": "./src/utils/oauth/*.ts",
 			"import": "./src/utils/oauth/*.ts"
 		},
 		"./utils/schema": {
-			"types": "./dist/types/utils/schema/index.d.ts",
+			"types": "./src/utils/schema/index.ts",
 			"import": "./src/utils/schema/index.ts"
 		},
 		"./utils/schema/*": {
-			"types": "./dist/types/utils/schema/*.d.ts",
+			"types": "./src/utils/schema/*.ts",
 			"import": "./src/utils/schema/*.ts"
 		},
 		"./*.js": "./src/*.ts"

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 "@sayknow-cli/utils";
+import { getModelDbPath } from "@sayknow-cli/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/model-thinking.ts

@@ -47,6 +47,7 @@ const DEFAULT_REASONING_EFFORTS_WITH_XHIGH_AND_MAX: readonly Effort[] = [
 const GEMINI_3_PRO_EFFORTS: readonly Effort[] = [Effort.Low, Effort.High];
 const GEMINI_3_FLASH_EFFORTS: readonly Effort[] = [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High];
 const GPT_5_2_PLUS_EFFORTS: readonly Effort[] = [Effort.Low, Effort.Medium, Effort.High, Effort.XHigh];
+const GPT_5_6_PLUS_EFFORTS: readonly Effort[] = [Effort.Low, Effort.Medium, Effort.High, Effort.XHigh, Effort.Max];
 const GPT_5_5_DEFAULT_EFFORT = Effort.XHigh;
 
 const GPT_5_1_CODEX_MINI_EFFORTS: readonly Effort[] = [Effort.Medium, Effort.High];
@@ -60,7 +61,18 @@ type SemVer = {
 
 type GeminiKind = "pro" | "flash";
 type AnthropicKind = "opus" | "sonnet";
-type OpenAIVariant = "base" | "codex" | "codex-max" | "codex-mini" | "codex-spark" | "mini" | "max" | "nano";
+type OpenAIVariant =
+	| "base"
+	| "codex"
+	| "codex-max"
+	| "codex-mini"
+	| "codex-spark"
+	| "luna"
+	| "mini"
+	| "max"
+	| "nano"
+	| "sol"
+	| "terra";
 
 const CODEX_GPT_5_4_PRIORITY_BY_VARIANT: Partial<Record<OpenAIVariant, number>> = {
 	base: 0,
@@ -465,11 +477,25 @@ function applyGpt55ContextWindow(model: ApiModel<Api>, parsedModel: OpenAIModel)
 	}
 	return false;
 }
+const GPT_5_6_TIER_VARIANTS: ReadonlySet<OpenAIVariant> = new Set(["base", "sol", "terra", "luna"]);
+
+function applyGpt56ContextWindow(model: ApiModel<Api>, parsedModel: OpenAIModel): boolean {
+	if (!semverGte(parsedModel.version, "5.6") || !GPT_5_6_TIER_VARIANTS.has(parsedModel.variant)) {
+		return false;
+	}
+	// GPT-5.6 tiers enforce a ~373K usable prompt budget on both transports
+	// (matches the openai-codex live catalog), despite the 1M+ marketing window.
+	model.contextWindow = 373_000;
+	return true;
+}
 
 function applyOpenAICatalogPolicy(model: ApiModel<Api>, parsedModel: OpenAIModel): void {
 	if (applyGpt55ContextWindow(model, parsedModel)) {
 		return;
 	}
+	if (applyGpt56ContextWindow(model, parsedModel)) {
+		return;
+	}
 	// OpenAI code backend models: 400K figure includes output budget; input window is 272K.
 	if (parsedModel.variant.startsWith("codex") && parsedModel.variant !== "codex-spark") {
 		model.contextWindow = 272000;
@@ -582,6 +608,9 @@ function inferOpenAISupportedEfforts(model: OpenAIModel): readonly Effort[] {
 	if (model.variant === "codex-mini" && semverEqual(model.version, "5.1")) {
 		return GPT_5_1_CODEX_MINI_EFFORTS;
 	}
+	if (semverGte(model.version, "5.6")) {
+		return GPT_5_6_PLUS_EFFORTS;
+	}
 	if (semverGte(model.version, "5.2")) {
 		return GPT_5_2_PLUS_EFFORTS;
 	}
@@ -714,7 +743,10 @@ function parseAnthropicModel(modelId: string): AnthropicModel | null {
 }
 
 function parseOpenAIModel(modelId: string): OpenAIModel | null {
-	const match = /gpt-(\d+(?:\.\d+){0,2})(?:-(codex-spark|codex-mini|codex-max|codex|mini|max|nano))?$/.exec(modelId);
+	const match =
+		/gpt-(\d+(?:\.\d+){0,2})(?:-(codex-spark|codex-mini|codex-max|codex|luna|mini|max|nano|sol|terra))?$/.exec(
+			modelId,
+		);
 	if (!match) {
 		return null;
 	}

package/src/models.ts

@@ -1,26 +1,46 @@
+import { readFileSync } from "node:fs";
+import { isRetiredModelKey } from "./model-retirements";
 import { applyGeneratedModelPolicies, enrichModelThinking } from "./model-thinking";
-import MODELS from "./models.json" with { type: "json" };
+// `with { type: "file" }` is embedded by `bun build --compile` and resolves to
+// the bunfs path inside standalone binaries (and to the on-disk path in dev).
+// A plain `createRequire` of a `.json` listed as an extra compile entrypoint is
+// NOT emitted into the bunfs, and its cwd-fallback masks the failure whenever
+// the process runs inside a repo checkout — see PR body for the minimal repro.
+import modelsJsonPath from "./models.json" with { type: "file" };
 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");
+
+let bundledCatalog: BundledCatalog | undefined;
+let providerNames: KnownProvider[] | undefined;
 const providerModelRegistry: Map<string, Map<string, Model<Api>>> = new Map();
 
+function getBundledCatalog(): BundledCatalog {
+	// TS types a .json import as its contents; at runtime `with { type: "file" }`
+	// yields the file path (bunfs path in compiled binaries, disk path in dev).
+	bundledCatalog ??= JSON.parse(readFileSync(modelsJsonPath as unknown as string, "utf8")) 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 +72,7 @@ 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 +82,7 @@ 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/amazon-bedrock.ts

@@ -7,7 +7,7 @@
  * Bun's native `HTTPS_PROXY` support.
  */
 
-import { $env, $flag, extractHttpStatusFromError, fetchWithRetry } from "@sayknow-cli/utils";
+import { $credentialEnv, $env, $flag, extractHttpStatusFromError, fetchWithRetry } from "@sayknow-cli/utils";
 import type { Effort } from "../model-thinking";
 import { mapEffortToAnthropicAdaptiveEffort, requireSupportedEffort } from "../model-thinking";
 import { calculateCost } from "../models";
@@ -39,8 +39,10 @@ import {
 	markToolChoiceIncapability,
 	resolveToolChoice,
 } from "../utils/tool-choice-capability";
+import { isValidBedrockBearerToken } from "./aws-credential-config";
 import { resolveAwsCredentials } from "./aws-credentials";
 import { decodeEventStream } from "./aws-eventstream";
+import type { AwsCredentials } from "./aws-sigv4";
 import { signRequest } from "./aws-sigv4";
 import { transformMessages } from "./transform-messages";
 
@@ -162,6 +164,8 @@ interface MetadataEvent {
 	};
 }
 
+type BedrockAuthMode = { kind: "bearer"; token: string } | { kind: "sigv4"; credentials: AwsCredentials };
+
 export const streamBedrock: StreamFunction<"bedrock-converse-stream"> = (
 	model: Model<"bedrock-converse-stream">,
 	context: Context,
@@ -233,34 +237,44 @@ export const streamBedrock: StreamFunction<"bedrock-converse-stream"> = (
 				body: commandInput,
 			};
 
-			let credentials: { accessKeyId: string; secretAccessKey: string; sessionToken?: string };
-			if ($flag("AWS_BEDROCK_SKIP_AUTH")) {
-				credentials = { accessKeyId: "dummy-access-key", secretAccessKey: "dummy-secret-key" };
-			} else {
-				credentials = await resolveAwsCredentials({
-					profile: options.profile,
-					region,
-					signal: options.signal,
-				});
+			const bearerToken = $credentialEnv("AWS_BEARER_TOKEN_BEDROCK");
+			if (bearerToken && !isValidBedrockBearerToken(bearerToken)) {
+				throw new Error("AWS_BEARER_TOKEN_BEDROCK contains unsafe control characters.");
 			}
-
+			const authMode: BedrockAuthMode = bearerToken
+				? { kind: "bearer", token: bearerToken }
+				: {
+						kind: "sigv4",
+						credentials: $flag("AWS_BEDROCK_SKIP_AUTH")
+							? { accessKeyId: "dummy-access-key", secretAccessKey: "dummy-secret-key" }
+							: await resolveAwsCredentials({ profile: options.profile, region, signal: options.signal }),
+					};
 			const bodyText = JSON.stringify(commandInput);
 			const body = new TextEncoder().encode(bodyText);
 			const baseHeaders: Record<string, string> = {
 				"content-type": "application/json",
 				accept: "application/vnd.amazon.eventstream",
 			};
-			const signed = await signRequest({
-				method: "POST",
-				host,
-				path: urlPath,
-				body,
-				region,
-				service: "bedrock",
-				credentials,
-				headers: baseHeaders,
-			});
-			const requestHeaders: Record<string, string> = { ...baseHeaders, ...signed };
+			const buildRequestHeaders = async (requestBody: Uint8Array): Promise<Record<string, string>> => {
+				const headers = new Headers(baseHeaders);
+				if (authMode.kind === "bearer") {
+					headers.set("authorization", `Bearer ${authMode.token}`);
+					return Object.fromEntries(headers);
+				}
+				const signed = await signRequest({
+					method: "POST",
+					host,
+					path: urlPath,
+					body: requestBody,
+					region,
+					service: "bedrock",
+					credentials: authMode.credentials,
+					headers: baseHeaders,
+				});
+				for (const [name, value] of Object.entries(signed)) headers.set(name, value);
+				return Object.fromEntries(headers);
+			};
+			const requestHeaders = await buildRequestHeaders(body);
 			const sentForcedToolChoice = Boolean(toolConfig?.toolChoice?.any || toolConfig?.toolChoice?.tool);
 			let fallbackRan = false;
 			const retryWithoutForcedToolChoice = async (reason: string) => {
@@ -279,20 +293,10 @@ export const streamBedrock: StreamFunction<"bedrock-converse-stream"> = (
 				stripBedrockForcedToolChoiceForRetry(commandInput);
 				const retryBodyText = JSON.stringify(commandInput);
 				const retryBody = new TextEncoder().encode(retryBodyText);
-				const retrySigned = await signRequest({
-					method: "POST",
-					host,
-					path: urlPath,
-					body: retryBody,
-					region,
-					service: "bedrock",
-					credentials,
-					headers: baseHeaders,
-				});
 				if (rawRequestDump) rawRequestDump.body = commandInput;
 				return fetchWithRetry(url, {
 					method: "POST",
-					headers: { ...baseHeaders, ...retrySigned },
+					headers: await buildRequestHeaders(retryBody),
 					body: retryBody,
 					signal: options.signal,
 					maxAttempts: 1,