@kenkaiiii/gg-core 5.93.0 → 5.95.0

package/dist/model-registry.cjs

@@ -57,80 +57,68 @@ __export(model_registry_exports, {
 });
 module.exports = __toCommonJS(model_registry_exports);
 
-// src/oauth/kimi.ts
-var import_node_child_process = require("child_process");
-var import_node_crypto = require("crypto");
-var import_node_fs = require("fs");
-var import_node_os2 = require("os");
-var import_node_path2 = __toESM(require("path"), 1);
-
-// src/paths.ts
-var import_node_path = __toESM(require("path"), 1);
-var import_node_os = __toESM(require("os"), 1);
-
-// src/oauth/kimi.ts
-var DEFAULT_CODING_BASE_URL = "https://api.kimi.com/coding/v1";
-var DEVICE_TIMEOUT_MS = 15 * 60 * 1e3;
-function kimiCodeBaseUrl() {
-  return (process.env.KIMI_CODE_BASE_URL ?? DEFAULT_CODING_BASE_URL).replace(/\/+$/, "");
-}
+// src/kimi-region.ts
+var import_gg_ai = require("@kenkaiiii/gg-ai");
 function isKimiCodingEndpoint(baseUrl) {
   if (typeof baseUrl !== "string" || baseUrl.length === 0) return false;
-  const normalized = baseUrl.replace(/\/+$/, "");
-  return normalized === kimiCodeBaseUrl() || /(^|\.)kimi\.com/i.test(normalized);
+  return /^https:\/\/api\.kimi\.(com|ai)\/coding(\/|$)/i.test(baseUrl);
 }
 
 // src/auth-storage.ts
-var import_promises3 = __toESM(require("fs/promises"), 1);
-var import_node_fs3 = require("fs");
-var import_node_crypto6 = __toESM(require("crypto"), 1);
+var import_promises6 = __toESM(require("fs/promises"), 1);
+var import_node_fs4 = require("fs");
+var import_node_crypto4 = __toESM(require("crypto"), 1);
 
-// src/oauth/anthropic.ts
-var import_node_crypto2 = __toESM(require("crypto"), 1);
+// src/paths.ts
+var import_node_path = __toESM(require("path"), 1);
+var import_node_os = __toESM(require("os"), 1);
 
-// src/claude-code-version.ts
-var import_gg_ai = require("@kenkaiiii/gg-ai");
+// src/claude-code/runtime.ts
+var import_node_child_process = require("child_process");
+var import_node_crypto = __toESM(require("crypto"), 1);
+var import_node_fs2 = __toESM(require("fs"), 1);
+var import_promises2 = __toESM(require("fs/promises"), 1);
+var import_node_path2 = __toESM(require("path"), 1);
 
-// src/oauth/anthropic.ts
-var CLIENT_ID = atob("OWQxYzI1MGEtZTYxYi00NGQ5LTg4ZWQtNTk0NGQxOTYyZjVl");
+// src/claude-code/tar.ts
+var import_node_fs = __toESM(require("fs"), 1);
+var import_promises = __toESM(require("fs/promises"), 1);
+var import_node_zlib = __toESM(require("zlib"), 1);
 
 // src/oauth/openai.ts
 var import_node_http = __toESM(require("http"), 1);
-var import_node_crypto3 = __toESM(require("crypto"), 1);
-
-// src/oauth/gemini.ts
-var import_node_http2 = __toESM(require("http"), 1);
-var import_node_crypto4 = __toESM(require("crypto"), 1);
-var SCOPE = [
-  "https://www.googleapis.com/auth/cloud-platform",
-  "https://www.googleapis.com/auth/userinfo.email",
-  "https://www.googleapis.com/auth/userinfo.profile"
-].join(" ");
-
-// src/oauth/xai.ts
-var DEVICE_TIMEOUT_FALLBACK_MS = 10 * 60 * 1e3;
+var import_node_crypto2 = __toESM(require("crypto"), 1);
+var import_promises3 = __toESM(require("fs/promises"), 1);
+var import_node_path3 = __toESM(require("path"), 1);
+var ISSUER = "https://auth.openai.com";
+var AUTHORIZE_URL = `${ISSUER}/api/accounts/authorize`;
+var TOKEN_URL = `${ISSUER}/api/accounts/oauth/token`;
+var REVOKE_URL = `${ISSUER}/api/accounts/oauth/revoke`;
+var JWKS_URL = `${ISSUER}/.well-known/jwks.json`;
+var CHATGPT_PLAN_SCOPE = "chatgpt.tokens.use.direct";
+var SCOPES = `openid profile email offline_access resource.invoke ${CHATGPT_PLAN_SCOPE}`;
 
 // src/file-lock.ts
-var import_promises = __toESM(require("fs/promises"), 1);
-var import_promises2 = require("timers/promises");
+var import_promises4 = __toESM(require("fs/promises"), 1);
+var import_promises5 = require("timers/promises");
 
 // src/logger.ts
-var import_node_fs2 = __toESM(require("fs"), 1);
-var import_node_path3 = __toESM(require("path"), 1);
-var import_node_crypto5 = require("crypto");
+var import_node_fs3 = __toESM(require("fs"), 1);
+var import_node_path4 = __toESM(require("path"), 1);
+var import_node_crypto3 = require("crypto");
 var import_gg_ai2 = require("@kenkaiiii/gg-ai");
 var MAX_BYTES = 10 * 1024 * 1024;
 var MIN_HEADROOM_BYTES = 64 * 1024;
 
 // src/hidden-providers.ts
-var HIDDEN_PROVIDERS = ["gemini"];
+var HIDDEN_PROVIDERS = [];
 function isHiddenProvider(provider) {
   return HIDDEN_PROVIDERS.includes(provider);
 }
 
 // src/auth-storage.ts
 var import_gg_ai3 = require("@kenkaiiii/gg-ai");
-var MOONSHOT_OAUTH_KEY = "moonshot-oauth";
+var KIMI_CODE_KEY = "kimi-code";
 var XIAOMI_CREDITS_KEY = "xiaomi-credits";
 var GEMINI_API_KEY = "gemini-api";
 var LOCAL_CREDENTIAL_LIFETIME_MS = 100 * 365 * 24 * 60 * 60 * 1e3;
@@ -155,7 +143,8 @@ var CODEX_FAST_BILLING = {
   detail: "Codex Fast mode uses your ChatGPT plan limits at 2.5x the standard rate (purchased credits at 2x).",
   paid: false
 };
-var GEMINI_AUTH_KEYS = [GEMINI_API_KEY, "gemini"];
+var GEMINI_AUTH_KEYS = [GEMINI_API_KEY];
+var KIMI_AUTH_KEYS = [KIMI_CODE_KEY, "moonshot"];
 var MODELS = [
   // ── Anthropic ──────────────────────────────────────────
   // NOTE: Claude Mythos 5.1 (`claude-mythos-5-1`, 2026-09-01) is kept commented
@@ -285,14 +274,13 @@ var MODELS = [
   {
     // GPT-6 Astra — "Our most capable model for complex, demanding work."
     // (Codex catalog priority 2, listed for every ChatGPT plan, requires a
-    // Codex client >= 0.153.0 — see CODEX_CLIENT_VERSION). Same split as 5.6:
-    // 1.05M on the public Responses API, 272K on the ChatGPT OAuth route
+    // Codex client >= 0.153.0). Same split as 5.6:
+    // 1.05M on the public Responses API, 272K on the ChatGPT plan route
     // (openai/codex models.json, `gpt-6-astra`). Reasoning ladder low → medium
     // → high → xhigh → max → ultra; `ultra` is the Codex orchestration preset
     // (multi_agent v2) and is Codex-only — the public API tops out at `max`.
-    // Note: through a plain API key OpenAI requires the Responses API for tool
-    // calling on Astra, so the Chat Completions path is text-only; the OAuth
-    // Codex route is the supported way to use it as an agent.
+    // OpenAI requires the Responses API for tool calling on Astra, so both
+    // the ChatGPT plan and API keys ride gg-ai's Responses transport.
     id: "gpt-6-astra",
     name: "GPT-6 Astra",
     provider: "openai",
@@ -326,18 +314,16 @@ var MODELS = [
     billing: CODEX_FAST_BILLING
   },
   // GPT-6 Sol + Luna — released 2026-09-22 below Astra, replacing the whole
-  // GPT-5.6 family (Sol/Terra/Luna; there is no GPT-6 Terra — OpenAI's Codex
+  // GPT-5.6 family (Sol/Terra/Luna; there was no GPT-6 Terra — OpenAI's Codex
   // catalog upgrades 5.6 Terra to 6 Sol). GPT-6.1 Sol (2026-09-29) then replaced
-  // GPT-6 Sol. Same window split as Astra: 1.05M on the public Responses API,
+  // GPT-6 Sol, and GPT-6.1 Terra (2026-10-26) sits between Sol and Luna. Same window split as Astra: 1.05M on the public Responses API,
   // 272K on the Codex route; 128K output, text+image input, freeform
   // apply_patch, responses-lite transport. The GPT-5.6 ids and `gpt-6-sol` are
   // retired — a saved session on one falls back to the provider default on
   // next start.
   {
     // GPT-6.1 Sol — "Latest workhorse model for coding and everyday work."
-    // (Codex priority 1, default low). The catalog says Codex client >= 0.153.0,
-    // but the ChatGPT backend only serves it from 0.159.0 (see
-    // CODEX_CLIENT_VERSION). $2/$10 MTok, cached input $0.10. Ladder low →
+    // (Codex priority 1, default low). $2/$10 MTok, cached input $0.10. Ladder low →
     // medium → high → xhigh → max → ultra; ultra is the Codex orchestration
     // preset (xhigh effort on the wire + proactive local subagent delegation).
     id: "gpt-6.1-sol",
@@ -370,6 +356,40 @@ var MODELS = [
     maxThinkingLevel: "ultra",
     billing: CODEX_FAST_BILLING
   },
+  {
+    // GPT-6.1 Terra (Codex priority 7, default low, client >= 0.153.0). $1/$5
+    // MTok, cached input $0.05. Unlike Sol/Astra the Codex route gets a 500K
+    // window (catalog `context_window`). Same ladder and Ultra preset as Sol.
+    id: "gpt-6.1-terra",
+    name: "GPT-6.1 Terra",
+    provider: "openai",
+    contextWindow: 105e4,
+    codexContextWindow: 5e5,
+    maxOutputTokens: 128e3,
+    supportsThinking: true,
+    defaultThinkingLevel: "low",
+    supportsImages: true,
+    supportsVideo: false,
+    costTier: "medium",
+    maxThinkingLevel: "ultra"
+  },
+  {
+    // Codex Fast mode for 6.1 Terra ("1.5x speed") — see `gpt-6-astra-fast`
+    // for the wire mapping and billing.
+    id: "gpt-6.1-terra-fast",
+    name: "GPT-6.1 Terra Fast",
+    provider: "openai",
+    contextWindow: 105e4,
+    codexContextWindow: 5e5,
+    maxOutputTokens: 128e3,
+    supportsThinking: true,
+    defaultThinkingLevel: "low",
+    supportsImages: true,
+    supportsVideo: false,
+    costTier: "medium",
+    maxThinkingLevel: "ultra",
+    billing: CODEX_FAST_BILLING
+  },
   {
     // Luna — "Fast and affordable model for easier tasks." (Codex priority 4,
     // default medium, needs a Codex client >= 0.155.0). $0.10/$0.50 MTok.
@@ -490,8 +510,8 @@ var MODELS = [
     maxThinkingLevel: "high",
     authStorageKeys: GEMINI_AUTH_KEYS
   },
-  // Keep 3.1 Flash Lite first for the working OAuth default and fast-model routing.
-  // New GA models are opt-in; Code Assist access varies by account. Gemini 3.7
+  // Keep 3.1 Flash Lite first for the default and fast-model routing.
+  // Gemini 3.7
   // Flash is retired: Google deprecated it on 2026-10-08 and routes every
   // `gemini-3.7-flash` request to 3.8 Flash (ai.google.dev/gemini-api/docs/
   // changelog), so a saved session on it falls back to the provider default.
@@ -524,9 +544,7 @@ var MODELS = [
     authStorageKeys: GEMINI_AUTH_KEYS
   },
   {
-    // Wire name `gemini-3-flash` — the Code Assist (OAuth) backend rejects the
-    // display string `gemini-3.5-flash` with a 404, so gemini-cli keeps this
-    // alternative name (SECONDARY_GEMINI_3_5_FLASH_MODEL) for that endpoint.
+    // Wire name `gemini-3-flash`, the id Google serves 3.5 Flash under.
     id: "gemini-3-flash",
     name: "Gemini 3.5 Flash",
     provider: "gemini",
@@ -541,9 +559,7 @@ var MODELS = [
     authStorageKeys: GEMINI_AUTH_KEYS
   },
   {
-    // Gemini 3.1 Pro is public preview — gated behind Code Assist preview
-    // enablement, so free/personal OAuth accounts 404 on it (see
-    // ACCOUNT_GATED_MODELS in gg-ai's gemini provider).
+    // Gemini 3.1 Pro is public preview.
     id: "gemini-3.1-pro-preview",
     name: "Gemini 3.1 Pro (Preview)",
     provider: "gemini",
@@ -560,8 +576,8 @@ var MODELS = [
   // ── Moonshot (Kimi) ────────────────────────────────────
   // K3 is Kimi's 2.8T-parameter flagship for long-horizon coding, knowledge
   // work, and deep reasoning. Its effort ladder is server-declared as
-  // low/high/max on both the public API (default max) and the Kimi For Coding
-  // OAuth endpoint (default high); thinking can also be fully disabled.
+  // low/high/max on both the public API (default max) and the Kimi Code plan
+  // endpoint (default high); thinking can also be fully disabled.
   {
     id: "kimi-k3",
     name: "Kimi K3",
@@ -575,13 +591,14 @@ var MODELS = [
     supportsVideo: true,
     maxVideoBytes: 100 * 1024 * 1024,
     costTier: "high",
-    maxThinkingLevel: "max"
+    maxThinkingLevel: "max",
+    authStorageKeys: KIMI_AUTH_KEYS
   },
-  // K2.8 Preview (2026-09-11) is served only on the Kimi For Coding OAuth
-  // endpoint, under its rolling `kimi-for-coding` id (live /models, 2026-09-28:
+  // K2.8 Preview (2026-09-11) is served only on the Kimi Code plan endpoint,
+  // under its rolling `kimi-for-coding` id (live /models, 2026-09-28:
   // display_name "K2.8 Preview", 1M context, image + video input, efforts
-  // low/high/max default max). The public API-key endpoint does not serve it,
-  // so it resolves from the Kimi sign-in credential only.
+  // low/high/max default max). The public Moonshot API does not serve it, so
+  // it resolves from a Kimi Code plan key only.
   {
     id: "kimi-for-coding",
     name: "Kimi K2.8 Preview",
@@ -594,7 +611,7 @@ var MODELS = [
     maxVideoBytes: 100 * 1024 * 1024,
     costTier: "medium",
     maxThinkingLevel: "max",
-    authStorageKeys: [MOONSHOT_OAUTH_KEY]
+    authStorageKeys: [KIMI_CODE_KEY]
   },
   // K2.7 Code is requested by its pinned id (not the `kimi-for-coding` alias
   // that moved to K2.8), so it stays the real K2.7 on both endpoints.
@@ -609,7 +626,8 @@ var MODELS = [
     supportsVideo: true,
     maxVideoBytes: 100 * 1024 * 1024,
     costTier: "medium",
-    maxThinkingLevel: "high"
+    maxThinkingLevel: "high",
+    authStorageKeys: KIMI_AUTH_KEYS
   },
   // K2.7 Code HighSpeed — the same model as K2.7 Code served at ~180 tok/s
   // (up to ~260 on short contexts), 256K context (platform.kimi.ai model list,
@@ -628,7 +646,8 @@ var MODELS = [
     supportsVideo: true,
     maxVideoBytes: 100 * 1024 * 1024,
     costTier: "medium",
-    maxThinkingLevel: "high"
+    maxThinkingLevel: "high",
+    authStorageKeys: KIMI_AUTH_KEYS
   },
   // ── Z.AI (GLM) ─────────────────────────────────────────
   // Two GLM entries, both live on the coding endpoint (verified against its