@sayknow-cli/ai 0.2.7 → 0.3.0

package/src/auth-storage.ts

@@ -151,6 +151,29 @@ export interface AuthCredentialSnapshotEntry {
 	identityKey: string | null;
 }
 
+export type AuthCredentialIfAbsentReason =
+	| "inserted"
+	| "skipped-existing"
+	| "skipped-existing-runtime"
+	| "skipped-existing-config"
+	| "skipped-existing-env"
+	| "skipped-existing-fallback"
+	| "skipped-invalid";
+
+export interface AuthCredentialIfAbsentResult {
+	inserted: boolean;
+	reason: AuthCredentialIfAbsentReason;
+	provider: string;
+	entries: StoredAuthCredential[];
+}
+
+export interface AuthCredentialIfAbsentSnapshotResult {
+	inserted: boolean;
+	reason: AuthCredentialIfAbsentReason;
+	provider: string;
+	entries: AuthCredentialSnapshotEntry[];
+}
+
 /**
  * Wire-shaped snapshot exported by {@link AuthStorage.exportSnapshot} and
  * served by the auth-broker server on `GET /v1/snapshot`.
@@ -182,6 +205,7 @@ export interface AuthCredentialStore {
 	tryDisableAuthCredentialIfMatches(id: number, expectedData: string, disabledCause: string): boolean;
 	replaceAuthCredentialsForProvider(provider: string, credentials: AuthCredential[]): StoredAuthCredential[];
 	upsertAuthCredentialForProvider(provider: string, credential: AuthCredential): StoredAuthCredential[];
+	upsertAuthCredentialForProviderIfAbsent(provider: string, credential: AuthCredential): AuthCredentialIfAbsentResult;
 	deleteAuthCredentialsForProvider(provider: string, disabledCause: string): void;
 	getCache(key: string, options?: { includeExpired?: boolean }): string | null;
 	setCache(key: string, value: string, expiresAtSec: number): void;
@@ -254,6 +278,10 @@ export interface AuthCredentialStore {
 	 * post-write read path is consistent.
 	 */
 	upsertAuthCredentialRemote?(provider: string, credential: AuthCredential): Promise<StoredAuthCredential[]>;
+	upsertAuthCredentialRemoteIfAbsent?(
+		provider: string,
+		credential: AuthCredential,
+	): Promise<AuthCredentialIfAbsentResult>;
 	/**
 	 * Optional async write hook for replace-all semantics (e.g. API-key login
 	 * overwriting any previous keys for the same provider). When present,
@@ -1234,6 +1262,56 @@ export class AuthStorage {
 		this.#resetProviderAssignments(provider);
 	}
 
+	#toSnapshotEntries(provider: string, stored: StoredAuthCredential[]): AuthCredentialSnapshotEntry[] {
+		return stored.map(entry => {
+			const persisted = entry.credential;
+			const redacted: SnapshotCredential =
+				persisted.type === "api_key" ? persisted : { ...persisted, refresh: REMOTE_REFRESH_SENTINEL };
+			return {
+				id: entry.id,
+				provider: entry.provider,
+				credential: redacted,
+				identityKey: resolveCredentialIdentityKey(provider, persisted),
+			};
+		});
+	}
+
+	#snapshotSkipResult(provider: string, reason: AuthCredentialIfAbsentReason): AuthCredentialIfAbsentSnapshotResult {
+		return {
+			inserted: false,
+			reason,
+			provider,
+			entries: this.exportSnapshot().credentials.filter(entry => entry.provider === provider),
+		};
+	}
+
+	async importCredentialIfAbsent(
+		provider: string,
+		credential: AuthCredential,
+	): Promise<AuthCredentialIfAbsentSnapshotResult> {
+		if (this.#runtimeOverrides.has(provider)) return this.#snapshotSkipResult(provider, "skipped-existing-runtime");
+		if (this.#configOverrides.has(provider)) return this.#snapshotSkipResult(provider, "skipped-existing-config");
+		if (this.#getCredentialsForProvider(provider).length > 0)
+			return this.#snapshotSkipResult(provider, "skipped-existing");
+		if (getEnvApiKey(provider)) return this.#snapshotSkipResult(provider, "skipped-existing-env");
+		if (this.#fallbackResolver?.(provider)) return this.#snapshotSkipResult(provider, "skipped-existing-fallback");
+
+		const result = this.#store.upsertAuthCredentialRemoteIfAbsent
+			? await this.#store.upsertAuthCredentialRemoteIfAbsent(provider, credential)
+			: this.#store.upsertAuthCredentialForProviderIfAbsent(provider, credential);
+		this.#setStoredCredentials(
+			provider,
+			result.entries.map(entry => ({ id: entry.id, credential: entry.credential })),
+		);
+		this.#resetProviderAssignments(provider);
+		return {
+			inserted: result.inserted,
+			reason: result.reason,
+			provider: result.provider,
+			entries: this.#toSnapshotEntries(provider, result.entries),
+		};
+	}
+
 	async #upsertOAuthCredential(provider: string, credential: OAuthCredential): Promise<void> {
 		const stored = this.#store.upsertAuthCredentialRemote
 			? await this.#store.upsertAuthCredentialRemote(provider, credential)
@@ -1515,6 +1593,14 @@ export class AuthStorage {
 				});
 				break;
 			}
+			case "glm-zcode": {
+				const { loginGlmZcode } = await import("./utils/oauth/glm-zcode");
+				credentials = await loginGlmZcode({
+					...ctrl,
+					onManualCodeInput: ctrl.onManualCodeInput ?? manualCodeInput,
+				});
+				break;
+			}
 			case "fireworks": {
 				const { loginFireworks } = await import("./utils/oauth/fireworks");
 				const apiKey = await loginFireworks(ctrl);
@@ -3343,17 +3429,7 @@ export class AuthStorage {
 			stored.map(entry => ({ id: entry.id, credential: entry.credential })),
 		);
 		this.#resetProviderAssignments(provider);
-		return stored.map(entry => {
-			const persisted = entry.credential;
-			const redacted: SnapshotCredential =
-				persisted.type === "api_key" ? persisted : { ...persisted, refresh: REMOTE_REFRESH_SENTINEL };
-			return {
-				id: entry.id,
-				provider: entry.provider,
-				credential: redacted,
-				identityKey: resolveCredentialIdentityKey(provider, persisted),
-			};
-		});
+		return this.#toSnapshotEntries(provider, stored);
 	}
 
 	/**
@@ -3989,6 +4065,48 @@ export class SqliteAuthCredentialStore implements AuthCredentialStore {
 		return result;
 	}
 
+	upsertAuthCredentialForProviderIfAbsent(provider: string, credential: AuthCredential): AuthCredentialIfAbsentResult {
+		let serialized: SerializedCredentialRecord | null;
+		try {
+			serialized = serializeCredential(provider, credential);
+		} catch {
+			serialized = null;
+		}
+		if (!serialized) {
+			return {
+				inserted: false,
+				reason: "skipped-invalid",
+				provider,
+				entries: this.listAuthCredentials(provider),
+			};
+		}
+
+		const writeIfAbsent = this.#db.transaction(
+			(providerName: string, record: SerializedCredentialRecord): AuthCredentialIfAbsentResult => {
+				const existingRows = this.#listActiveByProviderStmt.all(providerName) as AuthRow[];
+				const existing: StoredAuthCredential[] = [];
+				for (const row of existingRows) {
+					const activeCredential = deserializeCredential(row);
+					if (!activeCredential) continue;
+					existing.push(toStoredAuthCredential(row, activeCredential));
+				}
+				if (existing.length > 0) {
+					return { inserted: false, reason: "skipped-existing", provider: providerName, entries: existing };
+				}
+
+				this.#insertStmt.get(providerName, record.credentialType, record.data, record.identityKey);
+				return {
+					inserted: true,
+					reason: "inserted",
+					provider: providerName,
+					entries: this.listAuthCredentials(providerName),
+				};
+			},
+		);
+
+		return writeIfAbsent.immediate(provider, serialized);
+	}
+
 	/**
 	 * Hard-deletes disabled rows for a provider when an active row with the same identity exists.
 	 * This prevents unbounded accumulation of soft-deleted credentials while preserving

package/src/cli.ts

@@ -109,6 +109,7 @@ Providers:
   kagi              Kagi
   tavily            Tavily
   zai               Z.AI (GLM Coding Plan)
+  glm-zcode         GLM ZCode OAuth (unofficial, opt-in; at your own risk)
   deepseek          DeepSeek
   xai               xAI
   nanogpt           NanoGPT

package/src/model-thinking.ts

@@ -404,13 +404,11 @@ function applyGeneratedModelPolicy(model: ApiModel<Api>): void {
 	if (model.provider === "zai" && model.id === "glm-5.2") {
 		model.contextWindow = 1_000_000;
 	}
-	// MiniMax-M3: official MiniMax docs (platform.minimax.io/docs/guides/models-intro)
-	// document a 1M context window, but models.dev and the bundled catalog both report
-	// 512K. The stale 512K survives generate-models (provider-scoped models bypass the
-	// models.dev refresh in applyGlobalModelsDevFallback), tripping auto-compaction /
-	// context-cap thresholds 2x early on MiniMax sessions. Pin to the true 1M.
+	// MiniMax-M3: MiniMax exposes a 1M context tier, but usage beyond 512K is
+	// billed separately. Keep bundled/default metadata at the billing-safe 512K
+	// unless an explicit paid-tier contract is added.
 	if (model.provider !== "opencode-go" && model.id === "minimax-m3") {
-		model.contextWindow = 1_000_000;
+		model.contextWindow = 512_000;
 	}
 }
 
@@ -455,10 +453,13 @@ function inferGeneratedApplyPatchToolType(
 }
 
 function applyGpt55ContextWindow(model: ApiModel<Api>, parsedModel: OpenAIModel): boolean {
-	// gpt-5.5 is a 400K-context model. OpenAI code backend discovery can omit the
-	// context window, falling back to the 272K default, which incorrectly trips
-	// context-cap / auto-promote thresholds (a ~272K session would look over-cap
-	// and demote to gpt-5.4). Pin gpt-5.5 to its true 400K window.
+	// OpenAI Codex reports GPT-5.5 with a 272K prompt budget. Keep the generated
+	// bundle aligned with the backend limit so compaction fires before the prompt
+	// crosses the usable Codex window instead of trusting stale 400K snapshots.
+	if (model.provider === "openai-codex" && parsedModel.variant === "base" && semverEqual(parsedModel.version, "5.5")) {
+		model.contextWindow = 272000;
+		return true;
+	}
 	if (parsedModel.variant === "base" && semverEqual(parsedModel.version, "5.5")) {
 		model.contextWindow = 400000;
 		return true;

package/src/models.json

@@ -10486,6 +10486,31 @@
 				"maxLevel": "high"
 			}
 		},
+		"gemini-3.5-flash": {
+			"id": "gemini-3.5-flash",
+			"name": "Gemini 3.5 Flash",
+			"api": "google-gemini-cli",
+			"provider": "google-gemini-cli",
+			"baseUrl": "https://cloudcode-pa.googleapis.com",
+			"reasoning": true,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 1.5,
+				"output": 9,
+				"cacheRead": 0.15,
+				"cacheWrite": 0
+			},
+			"contextWindow": 1048576,
+			"maxTokens": 65536,
+			"thinking": {
+				"mode": "google-level",
+				"minLevel": "minimal",
+				"maxLevel": "high"
+			}
+		},
 		"gemini-3-flash-preview": {
 			"id": "gemini-3-flash-preview",
 			"name": "Gemini 3 Flash Preview",
@@ -37453,7 +37478,7 @@
 				"cacheRead": 0.12,
 				"cacheWrite": 0
 			},
-			"contextWindow": 1000000,
+			"contextWindow": 512000,
 			"maxTokens": 128000,
 			"thinking": {
 				"mode": "budget",
@@ -37673,7 +37698,7 @@
 				"cacheRead": 0.12,
 				"cacheWrite": 0
 			},
-			"contextWindow": 1000000,
+			"contextWindow": 512000,
 			"maxTokens": 128000,
 			"thinking": {
 				"mode": "budget",
@@ -37965,7 +37990,7 @@
 				"cacheRead": 0,
 				"cacheWrite": 0
 			},
-			"contextWindow": 1000000,
+			"contextWindow": 512000,
 			"maxTokens": 128000,
 			"compat": {
 				"supportsStore": false,
@@ -38300,7 +38325,7 @@
 				"cacheRead": 0,
 				"cacheWrite": 0
 			},
-			"contextWindow": 1000000,
+			"contextWindow": 512000,
 			"maxTokens": 128000,
 			"compat": {
 				"supportsStore": false,
@@ -56285,7 +56310,7 @@
 				"cacheRead": 0.5,
 				"cacheWrite": 0
 			},
-			"contextWindow": 400000,
+			"contextWindow": 272000,
 			"maxTokens": 128000,
 			"preferWebsockets": true,
 			"priority": 9,
@@ -68507,7 +68532,7 @@
 				"cacheRead": 0,
 				"cacheWrite": 0
 			},
-			"contextWindow": 1000000,
+			"contextWindow": 512000,
 			"maxTokens": 128000,
 			"compat": {
 				"supportsUsageInStreaming": false
@@ -75673,6 +75698,32 @@
 			}
 		}
 	},
+	"glm-zcode": {
+		"glm-5.2": {
+			"id": "glm-5.2",
+			"name": "GLM-5.2 (ZCode)",
+			"api": "anthropic-messages",
+			"provider": "glm-zcode",
+			"baseUrl": "https://api.z.ai/api/anthropic",
+			"reasoning": true,
+			"input": [
+				"text"
+			],
+			"cost": {
+				"input": 0,
+				"output": 0,
+				"cacheRead": 0,
+				"cacheWrite": 0
+			},
+			"contextWindow": 1000000,
+			"maxTokens": 131072,
+			"thinking": {
+				"mode": "budget",
+				"minLevel": "minimal",
+				"maxLevel": "xhigh"
+			}
+		}
+	},
 	"zenmux": {
 		"anthropic/claude-3.5-haiku": {
 			"id": "anthropic/claude-3.5-haiku",
@@ -79707,4 +79758,4 @@
 			}
 		}
 	}
-}
+}

package/src/provider-models/descriptors.ts

@@ -43,7 +43,7 @@ import {
 	xiaomiModelManagerOptions,
 	zenmuxModelManagerOptions,
 } from "./openai-compat";
-import { cursorModelManagerOptions, zaiModelManagerOptions } from "./special";
+import { cursorModelManagerOptions, glmZcodeModelManagerOptions, zaiModelManagerOptions } from "./special";
 
 /** Catalog discovery configuration for providers that support endpoint-based model listing. */
 export interface CatalogDiscoveryConfig {
@@ -299,6 +299,12 @@ export const PROVIDER_DESCRIPTORS: readonly ProviderDescriptor[] = [
 		catalog("ZenMux", ["ZENMUX_API_KEY"]),
 	),
 	catalogDescriptor("zai", "glm-5.2", config => zaiModelManagerOptions(config), catalog("zAI", ["ZAI_API_KEY"])),
+	catalogDescriptor(
+		"glm-zcode",
+		"glm-5.2",
+		config => glmZcodeModelManagerOptions(config),
+		catalog("GLM ZCode (unofficial)", ["GLM_ZCODE_API_KEY"], { oauthProvider: "glm-zcode" }),
+	),
 	descriptor("github-copilot", "gpt-4o", config => githubCopilotModelManagerOptions(config)),
 	descriptor("google", "gemini-2.5-pro", config => googleModelManagerOptions(config)),
 	catalogDescriptor(

package/src/provider-models/openai-compat.ts

@@ -2278,6 +2278,12 @@ const MODELS_DEV_PROVIDER_DESCRIPTORS_CORE: readonly ModelsDevProviderDescriptor
 const MODELS_DEV_PROVIDER_DESCRIPTORS_CODING_PLANS: readonly ModelsDevProviderDescriptor[] = [
 	// --- zAI ---
 	anthropicMessagesDescriptor("zai-coding-plan", "zai", "https://api.z.ai/api/anthropic"),
+	// --- GLM ZCode (unofficial Z.AI OAuth) ---
+	anthropicMessagesDescriptor(
+		"glm-zcode-coding-plan",
+		"glm-zcode",
+		process.env.ZCODE_PLAN_ANTHROPIC_BASE_URL ?? "https://api.z.ai/api/anthropic",
+	),
 	// --- Xiaomi ---
 	openAiCompletionsDescriptor("xiaomi", "xiaomi", "https://api.xiaomimimo.com/v1", {
 		defaultContextWindow: 262144,

package/src/provider-models/special.ts

@@ -65,3 +65,15 @@ export interface ZaiModelManagerConfig {}
 export function zaiModelManagerOptions(_config: ZaiModelManagerConfig = {}): ModelManagerOptions<"anthropic-messages"> {
 	return { providerId: "zai" };
 }
+
+// ---------------------------------------------------------------------------
+// GLM ZCode (unofficial Z.AI OAuth)
+// ---------------------------------------------------------------------------
+
+export interface GlmZcodeModelManagerConfig {}
+
+export function glmZcodeModelManagerOptions(
+	_config: GlmZcodeModelManagerConfig = {},
+): ModelManagerOptions<"anthropic-messages"> {
+	return { providerId: "glm-zcode" };
+}

package/src/providers/anthropic.ts

@@ -1,5 +1,6 @@
 import * as nodeCrypto from "node:crypto";
 import * as fs from "node:fs";
+import * as os from "node:os";
 import { scheduler } from "node:timers/promises";
 import * as tls from "node:tls";
 import Anthropic, { type ClientOptions as AnthropicSdkClientOptions } from "@anthropic-ai/sdk";
@@ -93,6 +94,13 @@ export type AnthropicHeaderOptions = {
 	stream?: boolean;
 	modelHeaders?: Record<string, string>;
 	isCloudflareAiGateway?: boolean;
+	/**
+	 * Attach ZCode client "source" headers (User-Agent: ZCode/<ver>, X-Title,
+	 * X-ZCode-Agent: glm, X-Platform, etc.) so api.z.ai recognizes the caller as
+	 * the ZCode client, exactly like ZCode's `buildZCodeSourceHeaders` does for
+	 * GLM providers. glm-zcode only.
+	 */
+	zcodeSourceHeaders?: boolean;
 };
 
 export function normalizeAnthropicBaseUrl(baseUrl?: string): string | undefined {
@@ -161,6 +169,67 @@ const sharedHeaders = {
 	"X-App": "cli",
 };
 
+// ZCode bakes its app version and runtime env at build time. Mirror the values
+// from the analyzed ZCode 3.1.2 desktop bundle (`resolveRuntimeZCodeEnv` returns
+// "production" for non-test builds). Both are overridable for forward-compat.
+const ZCODE_APP_VERSION = process.env.ZCODE_APP_VERSION?.trim() || "3.1.2";
+const ZCODE_RELEASE_CHANNEL = process.env.ZCODE_RELEASE_CHANNEL?.trim() || "production";
+
+// Mirrors ZCode's `normalizePrintableHeaderValue`: only printable ASCII passes.
+function normalizePrintableHeaderValue(value: string | undefined): string | undefined {
+	const trimmed = value?.trim();
+	if (trimmed && /^[\x20-\x7e]+$/.test(trimmed)) return trimmed;
+	return undefined;
+}
+
+// Mirrors ZCode's `normalizeOsCategory`.
+function normalizeOsCategory(platform: NodeJS.Platform): string {
+	switch (platform) {
+		case "darwin":
+			return "macos";
+		case "win32":
+			return "windows";
+		default:
+			return "linux";
+	}
+}
+
+/**
+ * Replicates ZCode's `buildZCodeSourceHeaders()` + GLM `X-ZCode-Agent` tag
+ * (host bundle `Bl` / `buildConnectivitySourceHeaders` for GLM providers), so
+ * api.z.ai sees skc's glm-zcode requests as the ZCode client. Dynamic values
+ * (platform/arch, locale, timezone, OS version) are resolved at runtime exactly
+ * as ZCode does; printable-ASCII-only and conditionally omitted when empty.
+ */
+export function buildZCodeSourceHeaders(): Record<string, string> {
+	const platform = process.platform;
+	const arch = process.arch;
+	const appVersion = normalizePrintableHeaderValue(ZCODE_APP_VERSION);
+	const releaseChannel = normalizePrintableHeaderValue(ZCODE_RELEASE_CHANNEL);
+	let locale: string | undefined;
+	let timezone: string | undefined;
+	try {
+		const resolved = Intl.DateTimeFormat().resolvedOptions();
+		locale = normalizePrintableHeaderValue(resolved.locale);
+		timezone = normalizePrintableHeaderValue(resolved.timeZone);
+	} catch {}
+	const osVersion = normalizePrintableHeaderValue(os.version());
+	const headers: Record<string, string> = {
+		"User-Agent": `ZCode/${appVersion ?? "unknown"}`,
+		"HTTP-Referer": "https://zcode.z.ai",
+		"X-Title": "Z Code@electron",
+		"X-Platform": `${platform}-${arch}`,
+		"X-Client-Language": locale ?? "unknown",
+		"X-Client-Timezone": timezone ?? "unknown",
+		"X-Os-Category": normalizeOsCategory(platform),
+		"X-ZCode-Agent": "glm",
+	};
+	if (appVersion) headers["X-ZCode-App-Version"] = appVersion;
+	if (releaseChannel) headers["X-Release-Channel"] = releaseChannel;
+	if (osVersion) headers["X-Os-Version"] = osVersion;
+	return headers;
+}
+
 export function buildAnthropicHeaders(options: AnthropicHeaderOptions): Record<string, string> {
 	const oauthToken = options.isOAuth ?? isAnthropicOAuthToken(options.apiKey);
 	const extraBetas = options.extraBetas ?? [];
@@ -197,6 +266,9 @@ export function buildAnthropicHeaders(options: AnthropicHeaderOptions): Record<s
 		};
 	} else if (!isAnthropicApiBaseUrl(options.baseUrl)) {
 		const incomingUserAgent = getHeaderCaseInsensitive(options.modelHeaders, "User-Agent");
+		// ZCode merges its source headers LAST for GLM providers (`withZCodeSourceHeaders`
+		// → `{ ...base, ...extra, ...source }`), so they win over any incoming User-Agent.
+		const zcodeSourceHeaders = options.zcodeSourceHeaders ? buildZCodeSourceHeaders() : undefined;
 		return {
 			...modelHeaders,
 			Accept: acceptHeader,
@@ -204,6 +276,7 @@ export function buildAnthropicHeaders(options: AnthropicHeaderOptions): Record<s
 			...sharedHeaders,
 			"Anthropic-Beta": betaHeader,
 			...(incomingUserAgent ? { "User-Agent": incomingUserAgent } : {}),
+			...(zcodeSourceHeaders ?? {}),
 		};
 	} else {
 		return {
@@ -678,6 +751,12 @@ function resolveAnthropicBaseUrl(model: Model<"anthropic-messages">, apiKey?: st
 	if (model.provider === "github-copilot") {
 		return normalizeAnthropicBaseUrl(resolveGitHubCopilotBaseUrl(model.baseUrl, apiKey) ?? model.baseUrl);
 	}
+	// glm-zcode logs in via ZCode's OAuth but auto-provisions a real Z.AI API key and
+	// calls api.z.ai directly (no zcode.z.ai gateway, no captcha). Pin the base so dynamic
+	// discovery / stale bundled catalogs / model cache can't redirect it elsewhere.
+	if (model.provider === "glm-zcode") {
+		return normalizeAnthropicBaseUrl(process.env.ZCODE_PLAN_ANTHROPIC_BASE_URL) ?? "https://api.z.ai/api/anthropic";
+	}
 	if (model.provider === "anthropic" && isFoundryEnabled()) {
 		const foundryBaseUrl = normalizeAnthropicBaseUrl($env.FOUNDRY_BASE_URL);
 		if (foundryBaseUrl) {
@@ -1697,6 +1776,7 @@ export function buildAnthropicClientOptions(args: AnthropicClientOptionsArgs): A
 		stream,
 		modelHeaders: mergeHeaders(model.headers, foundryCustomHeaders, headers, dynamicHeaders),
 		isCloudflareAiGateway: model.provider === "cloudflare-ai-gateway",
+		zcodeSourceHeaders: model.provider === "glm-zcode",
 	});
 
 	if (model.provider === "cloudflare-ai-gateway") {
@@ -2169,7 +2249,7 @@ function buildParams(
  * See: https://github.com/jaybeyond/Sayknow_CLI/issues/814
  */
 function isZaiAnthropicEndpoint(model: Model<"anthropic-messages">): boolean {
-	if (model.provider === "zai") return true;
+	if (model.provider === "zai" || model.provider === "glm-zcode") return true;
 	const baseUrl = model.baseUrl;
 	if (!baseUrl) return false;
 	try {
@@ -2187,7 +2267,7 @@ function isZaiAnthropicEndpoint(model: Model<"anthropic-messages">): boolean {
  */
 function isNonSigningAnthropicEndpoint(model: Model<"anthropic-messages">): boolean {
 	// Known non-signing providers
-	if (model.provider === "zai" || model.provider === "deepseek") return true;
+	if (model.provider === "zai" || model.provider === "glm-zcode" || model.provider === "deepseek") return true;
 	const baseUrl = model.baseUrl;
 	if (!baseUrl) return false;
 	try {

package/src/stream.ts

@@ -88,6 +88,7 @@ const serviceProviderMap: Record<string, KeyResolver> = {
 	kilo: "KILO_API_KEY",
 	"vercel-ai-gateway": "AI_GATEWAY_API_KEY",
 	zai: "ZAI_API_KEY",
+	"glm-zcode": "GLM_ZCODE_API_KEY",
 	mistral: "MISTRAL_API_KEY",
 	minimax: "MINIMAX_API_KEY",
 	"minimax-code": "MINIMAX_CODE_API_KEY",

package/src/types.ts

@@ -122,6 +122,7 @@ export type KnownProvider =
 	| "kilo"
 	| "vercel-ai-gateway"
 	| "zai"
+	| "glm-zcode"
 	| "mistral"
 	| "minimax"
 	| "opencode-go"