@oonacode/win32-x64 0.3.230 → 0.3.232

package/oona.exe

@@ -1,4 +1,4 @@
 [diffend] Oversized file quarantined before diffing.
 name: package/oona.exe
-size: 112295936 bytes
-sha256: 38006815db1ee17abf3ec8651c5c29f7e99994853efff2f87bac4c211753f7eb
+size: 112301056 bytes
+sha256: edd7bbfaa7d9bd906422e1bf68160bc3c19671734b54b2bc7e5e9c72917ce158

package/oonacode-release.json

@@ -1,13 +1,13 @@
 {
     "name":  "oonacode",
-    "version":  "0.3.230",
+    "version":  "0.3.232",
     "channel":  "stable",
     "platform":  "win32",
     "arch":  "x64",
     "mode":  "bun",
     "launcherRuntime":  "bun 1.4.1",
-    "commit":  "db0fe54a",
-    "builtAt":  "2026-10-11T17:16:15.9107668Z",
+    "commit":  "b02fb4a3",
+    "builtAt":  "2026-10-12T13:47:41.2360719Z",
     "engine":  "engine-store",
     "engineVersion":  "0.3.296",
     "siteOrigin":  "https://oonacode.oonak.ai",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@oonacode/win32-x64",
-  "version": "0.3.230",
+  "version": "0.3.232",
   "description": "OonaCode native binaries for Windows x64: oona.exe (CLI + headless host service), the pinned oonaclaude.exe engine sidecar, local web UI assets. Installed automatically as an optional dependency of `oonacode`.",
   "license": "UNLICENSED",
   "homepage": "https://oonacode.oonak.ai",

package/registry.json

@@ -361,6 +361,132 @@
       "docsUrl": "https://developers.openai.com/codex",
       "notes": "The user's OWN ChatGPT plan (Plus, Pro, Team), signed in with OpenAI's device flow as `codex login --device-auth` runs it (docs/codex-engine.md section 6, 2026-09-19): the host asks auth.openai.com for a code, the person types it at auth.openai.com/codex/device, the host polls, exchanges the code with the verifier OpenAI's server minted, and keeps the access + refresh pair in the vault, refreshed two minutes before expiry. The backend a ChatGPT token belongs to is chatgpt.com/backend-api/codex, which speaks ONLY the Responses API (translate.wire = responses) and wants the chatgpt-account-id header (a claim of the id token, kept beside the pair), an originator and the responses=experimental beta - so the engine reaches it through the host's loopback upstream, which translates every request. route=direct: the token never reaches the gateway (guardrail 4) and this entry seeds no server rows. The client id is Codex's PUBLIC one; OpenAI names third-party harnesses (OpenCode, pi) as tolerated users of a ChatGPT plan and the owner accepted that standing on 2026-09-19 - the residual risk is OpenAI changing its mind. Models, context windows and effort levels are what Codex 0.162.1's own model list answered for a ChatGPT account on 2026-10-11 (gpt-6-astra, gpt-5.6-sol and gpt-5.2, listed before, answer 'not supported when using Codex with a ChatGPT account'); each level was sent and answered 200, and minimal and Codex's own ultra are refused by every model. A plan is flat-rate, so no prices. Its x-codex-primary/secondary-used-percent answer headers are the plan's usage."
     },
+    {
+      "id": "openai-codex",
+      "displayName": "OpenAI Codex (ChatGPT plan)",
+      "effortLevels": ["low", "medium", "high", "xhigh", "max"],
+      "kind": "openai-compatible",
+      "route": "gateway",
+      "credentialClass": "api_key",
+      "poolOnly": true,
+      "baseUrl": "https://chatgpt.com/backend-api/codex",
+      "messagesPath": "/responses",
+      "auth": {
+        "header": "authorization-bearer",
+        "extraHeaders": {
+          "originator": "codex_exec",
+          "user-agent": "codex_exec/0.162.1",
+          "OpenAI-Beta": "responses=experimental"
+        }
+      },
+      "defaultModel": "gpt-6-luna",
+      "models": [
+        {
+          "id": "gpt-6-luna",
+          "displayName": "GPT-6 Luna",
+          "contextWindow": 272000,
+          "supportsTools": true,
+          "supportsThinking": true,
+          "supportsVision": true,
+          "effortLevels": ["low", "medium", "high", "xhigh", "max"],
+          "maxOutputTokens": 128000,
+          "notes": "Codex's default (priority 4 in its model list). OpenAI API list price $0.10 / $0.01 cached / $0.50 per 1M (developers.openai.com/api/docs/models/gpt-6-luna, 2026-10-12) — the catalogue cost cells. The Codex backend takes 272k of its 1.05M window."
+        },
+        {
+          "id": "gpt-5.6-terra",
+          "displayName": "GPT-5.6 Terra",
+          "contextWindow": 272000,
+          "supportsTools": true,
+          "supportsThinking": true,
+          "supportsVision": true,
+          "effortLevels": ["low", "medium", "high", "xhigh", "max"],
+          "maxOutputTokens": 128000,
+          "notes": "OpenAI API list price $2 / $0.20 cached / $12 per 1M (developers.openai.com/api/docs/models/gpt-5.6-terra, 2026-10-12). The Codex backend takes 272k."
+        },
+        {
+          "id": "gpt-5.6-luna",
+          "displayName": "GPT-5.6 Luna",
+          "contextWindow": 272000,
+          "supportsTools": true,
+          "supportsThinking": true,
+          "supportsVision": true,
+          "effortLevels": ["low", "medium", "high", "xhigh", "max"],
+          "maxOutputTokens": 128000,
+          "notes": "OpenAI API list price $0.20 / $0.02 cached / $1.20 per 1M (developers.openai.com/api/docs/models/gpt-5.6-luna, 2026-10-12). The Codex backend takes 272k."
+        },
+        {
+          "id": "gpt-5.5",
+          "displayName": "GPT-5.5",
+          "contextWindow": 272000,
+          "supportsTools": true,
+          "supportsThinking": true,
+          "supportsVision": true,
+          "effortLevels": ["low", "medium", "high", "xhigh"],
+          "maxOutputTokens": 128000,
+          "notes": "Legacy: Codex hides it from its picker, the backend still serves it. No max effort. OpenAI API list price $5 / $0.50 cached / $30 per 1M (developers.openai.com/api/docs/models/gpt-5.5, 2026-10-12)."
+        }
+      ],
+      "modelAliases": {
+        "opus": "gpt-5.6-terra",
+        "sonnet": "gpt-5.6-terra",
+        "haiku": "gpt-6-luna"
+      },
+      "rateLimits": {
+        "maxConcurrent": 4,
+        "requestsPerMinute": 60
+      },
+      "capabilities": {
+        "streaming": true,
+        "tools": true,
+        "thinking": true,
+        "vision": true,
+        "countTokens": false,
+        "promptCaching": true
+      },
+      "translate": {
+        "wire": "responses",
+        "systemRole": "developer",
+        "reasoningEffortFromBudget": true,
+        "supportsParallelToolCalls": true
+      },
+      "phase": 1,
+      "enabled": true,
+      "adminGroup": "openai",
+      "recordLabel": "Subscription",
+      "docsUrl": "https://developers.openai.com/codex",
+      "references": [
+        {
+          "label": "Codex CLI (open source, Apache-2.0)",
+          "url": "https://github.com/openai/codex",
+          "kind": "docs"
+        },
+        {
+          "label": "OpenAI API model pages (the cost cells)",
+          "url": "https://developers.openai.com/api/docs/models",
+          "kind": "pricing"
+        },
+        {
+          "label": "OpenAI terms of use",
+          "url": "https://openai.com/policies/terms-of-use/",
+          "kind": "other"
+        }
+      ],
+      "notes": "OPERATOR POOL ONLY (owner, 2026-10-12: \"in the admin we need to add subscription tab … install codex cli on server like supergrok and use native login flow. but the user requests not goes to codex cli they already translated and proxy to openai\"). The operator's own ChatGPT plan, signed in from Admin → Providers → OpenAI → Subscription by a REAL Codex CLI on the server (`codex login --device-auth` in a private CODEX_HOME; packages/gateway/src/codex/*). The credential row is a POINTER; the access token and the account id are read from Codex's own auth.json per request, and only Codex refreshes them (app-server getAuthStatus refreshToken:true). Chats are NOT run through Codex: the engine's Anthropic request is translated to the Responses wire (@oonacode/translate responses-*.ts) and sent to chatgpt.com/backend-api/codex with Codex's identity headers and chatgpt-account-id. A ChatGPT plan is a consumer subscription and serving other people from it is against OpenAI's terms — the owner's call, as for the Claude and SuperGrok subscriptions. Administered under OpenAI (adminGroup): its record is the OpenAI page's Subscription tab. A user's OWN ChatGPT plan is openai-subscription, direct from their PC. docs/codex-engine.md §8.",
+      "credentialVariants": [
+        {
+          "id": "token-plan",
+          "label": "ChatGPT subscription (Codex on the server)",
+          "header": "authorization-bearer",
+          "operatorOnly": true,
+          "baseUrl": "https://chatgpt.com/backend-api/codex",
+          "codexLogin": {
+            "type": "codex-server",
+            "signInLabel": "Sign in with ChatGPT (Codex on the server)"
+          },
+          "notes": "Signed in by the server's own Codex CLI (version pinned in @oonacode/shared CODEX_TESTED_VERSION, installed at boot from OpenAI's npm package into the codexdata volume, sha512-verified). The identity headers are the ones the real Codex sends (recorded 2026-10-11, docs/codex-engine.md §7) — bump the user-agent version with Codex."
+        }
+      ]
+    },
     {
       "id": "xai",
       "displayName": "xAI (Grok)",
@@ -1358,7 +1484,7 @@
           "operatorOnly": true,
           "baseUrl": "https://daily-cloudcode-pa.googleapis.com",
           "extraHeaders": {
-            "user-agent": "antigravity/cli/1.3.1 (aidev_client; os_type=linux; arch=amd64; cl=994719654; auth_method=consumer)"
+            "user-agent": "antigravity/cli/1.3.3 (aidev_client; os_type=linux; arch=amd64; cl=996823801; auth_method=consumer)"
           },
           "agyLogin": {
             "type": "agy-server",
@@ -1385,6 +1511,144 @@
         }
       ],
       "effortLevels": ["low", "medium", "high"]
+    },
+    {
+      "id": "nvidia",
+      "displayName": "NVIDIA NIM (cloud)",
+      "kind": "openai-compatible",
+      "route": "gateway",
+      "credentialClass": "api_key",
+      "baseUrl": "https://integrate.api.nvidia.com",
+      "messagesPath": "/v1/chat/completions",
+      "modelsPath": "/v1/models",
+      "auth": {
+        "header": "authorization-bearer"
+      },
+      "models": [
+        {
+          "id": "nvidia/nemotron-3-ultra-550b-a55b",
+          "displayName": "Nemotron 3 Ultra 550B (NIM)",
+          "contextWindow": 262144,
+          "supportsTools": true,
+          "supportsThinking": true,
+          "supportsVision": false
+        },
+        {
+          "id": "nvidia/nemotron-3-super-120b-a12b",
+          "displayName": "Nemotron 3 Super 120B (NIM)",
+          "supportsTools": true,
+          "supportsThinking": true,
+          "supportsVision": false
+        },
+        {
+          "id": "nvidia/nemotron-3.5-lightning-30b-a3b",
+          "displayName": "Nemotron 3.5 Lightning 30B (NIM)",
+          "supportsTools": true,
+          "supportsThinking": true,
+          "supportsVision": false
+        },
+        {
+          "id": "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning",
+          "displayName": "Nemotron 3 Nano Omni 30B (NIM)",
+          "contextWindow": 262144,
+          "supportsTools": true,
+          "supportsThinking": true,
+          "supportsVision": true
+        },
+        {
+          "id": "moonshotai/kimi-k3",
+          "displayName": "Kimi K3 (NIM)",
+          "contextWindow": 1048576,
+          "supportsTools": true,
+          "supportsThinking": true,
+          "supportsVision": true,
+          "notes": "NVIDIA's card says a multi-turn tool conversation must send back the earlier reasoning_content; the translator drops thinking blocks on replay, so long tool loops may degrade. Unverified."
+        },
+        {
+          "id": "z-ai/glm-5.3",
+          "displayName": "GLM-5.3 (NIM)",
+          "contextWindow": 1048576,
+          "supportsTools": true,
+          "supportsThinking": true,
+          "supportsVision": false
+        },
+        {
+          "id": "z-ai/glm-5.3-flash",
+          "displayName": "GLM-5.3 Flash (NIM)",
+          "contextWindow": 1048576,
+          "supportsTools": true,
+          "supportsThinking": true,
+          "supportsVision": true
+        },
+        {
+          "id": "deepseek-ai/deepseek-v4.1-flash",
+          "displayName": "DeepSeek V4.1 Flash (NIM)",
+          "supportsTools": true,
+          "supportsThinking": true,
+          "supportsVision": false,
+          "notes": "NVIDIA's card was not readable (404), so no context window is set: the engine assumes 200k."
+        },
+        {
+          "id": "poolside/laguna-xs-2.1",
+          "displayName": "Poolside Laguna XS 2.1 (NIM)",
+          "contextWindow": 262144,
+          "supportsTools": true,
+          "supportsThinking": true,
+          "supportsVision": false
+        },
+        {
+          "id": "meta/muse-glimmer-30b",
+          "displayName": "Meta Muse Glimmer 30B (NIM)",
+          "contextWindow": 131072,
+          "supportsTools": true,
+          "supportsThinking": true,
+          "supportsVision": true
+        },
+        {
+          "id": "google/gemma-4-31b-it",
+          "displayName": "Gemma 4 31B (NIM)",
+          "contextWindow": 262144,
+          "supportsTools": true,
+          "supportsThinking": false,
+          "supportsVision": true,
+          "notes": "Thinking is switched by a token in the system prompt, which the translator does not send, so it runs without."
+        },
+        {
+          "id": "openai/gpt-oss-20b",
+          "displayName": "GPT-OSS 20B (NIM)",
+          "contextWindow": 128000,
+          "supportsTools": true,
+          "supportsThinking": true,
+          "supportsVision": false
+        }
+      ],
+      "defaultModel": "nvidia/nemotron-3-ultra-550b-a55b",
+      "modelAliases": {
+        "opus": "nvidia/nemotron-3-ultra-550b-a55b",
+        "sonnet": "nvidia/nemotron-3-ultra-550b-a55b",
+        "haiku": "nvidia/nemotron-3-super-120b-a12b"
+      },
+      "capabilities": {
+        "streaming": true,
+        "tools": true,
+        "thinking": true,
+        "vision": false,
+        "countTokens": false,
+        "promptCaching": false
+      },
+      "translate": {
+        "maxTokensParam": "max_tokens",
+        "reasoningEffortFromBudget": false,
+        "reasoningContentField": "reasoning_content",
+        "supportsParallelToolCalls": false,
+        "systemRole": "system",
+        "streamIncludeUsage": true
+      },
+      "userBaseUrl": true,
+      "phase": 1,
+      "enabled": true,
+      "notes": "A user's OWN key only (owner, 2026-10-12: \"bring back the nvidia for local only, i have apikey want to use it now\"). The host translates it on the PC (engine/local-upstream.ts) and the key never reaches the gateway; the server never calls NVIDIA (SERVER_EXCLUDED_PROVIDER_IDS: no pool rows, no admin page). Models are NVIDIA's current hosted catalogue that call tools (build.nvidia.com, 2026-10-12); a context window is set only where the model card states one, otherwise the engine's 200k default applies. Left out: embeddings, safety filters, parsers, DiffusionGemma and the two Llama 3.2 vision models (no tool calling).",
+      "docsUrl": "https://docs.api.nvidia.com/nim/"
     }
   ]
 }