@oonacode/win32-x64 0.3.297 → 0.3.299

package/oona.exe

@@ -1,4 +1,4 @@
 [diffend] Oversized file quarantined before diffing.
 name: package/oona.exe
-size: 114151424 bytes
-sha256: 7341f6bb97faf2d1bb2f66b82270d1e5cb71aba81156682add0b0107a3f376c1
+size: 114185216 bytes
+sha256: 43f44e1818b019c25379fcbc4c3f8aa3b133e0b2a5153110ff830280d3e9dfef

package/oonacode-release.json

@@ -1,13 +1,13 @@
 {
     "name":  "oonacode",
-    "version":  "0.3.297",
+    "version":  "0.3.299",
     "channel":  "stable",
     "platform":  "win32",
     "arch":  "x64",
     "mode":  "bun",
     "launcherRuntime":  "bun 1.4.1",
-    "commit":  "1b28b03c",
-    "builtAt":  "2026-11-02T09:37:42.7001774Z",
+    "commit":  "08daf447",
+    "builtAt":  "2026-11-04T15:49:21.8564698Z",
     "engine":  "engine-store",
     "engineVersion":  "0.3.301",
     "siteOrigin":  "https://oonacode.oonak.ai",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@oonacode/win32-x64",
-  "version": "0.3.297",
+  "version": "0.3.299",
   "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

@@ -332,6 +332,17 @@
           "notes": "Listed by the Codex backend for the operator's ChatGPT account on 2026-10-13 and answered a real turn through the genuine Codex CLI. Served from the person's own ChatGPT plan through the Codex backend; no per-token price - the plan's 5-hour and weekly windows are what it costs. The Codex backend takes 500k of context. Codex ultra effort is refused by the backend.",
           "defaultEffort": "low"
         },
+        {
+          "id": "gpt-6.1-terra",
+          "displayName": "GPT-6.1 Terra",
+          "contextWindow": 500000,
+          "supportsTools": true,
+          "supportsThinking": true,
+          "supportsVision": true,
+          "effortLevels": ["low", "medium", "high", "xhigh", "max"],
+          "notes": "Listed by Codex's own model list for a ChatGPT Plus account (client 0.166.0, 2026-11-04: priority 3, default effort low, 500k of context through Codex). Served from the person's own ChatGPT plan through the Codex backend; no per-token price - the plan's 5-hour and weekly windows are what it costs. Codex ultra effort is listed but refused by the backend, as for every Codex model here.",
+          "defaultEffort": "low"
+        },
         {
           "id": "gpt-6-sol",
           "displayName": "GPT-6 Sol",
@@ -400,7 +411,7 @@
         "signInLabel": "Sign in with ChatGPT"
       },
       "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's own model list answered for a ChatGPT account (0.162.1 on 2026-10-11, 0.163.0 on 2026-10-16); gpt-5.2 (2026-10-11) and gpt-5.5 (2026-10-16, after Codex 0.163.0 dropped it from the list) 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."
+      "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's own model list answered for a ChatGPT account (0.162.1 on 2026-10-11, 0.163.0 on 2026-10-16; Codex 0.166.0's cached list, 2026-11-04, still gives these seven the levels and defaults listed here, lists `ultra` (refused by the backend, so not offered) on every model but the two Lunas, and adds gpt-6.1-terra - low to max, default low, not listed here yet); gpt-5.2 (2026-10-11) and gpt-5.5 (2026-10-16, after Codex 0.163.0 dropped it from the list) 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",
@@ -469,6 +480,18 @@
           "notes": "Listed by the Codex backend for the operator's ChatGPT account on 2026-10-13 and answered a real turn through the genuine Codex CLI. OpenAI API list price $2 / $0.1 cached / $10 per 1M (developers.openai.com/api/docs/models/gpt-6.1-sol, 2026-10-13). The Codex backend takes 500k of context. Codex ultra effort is refused by the backend.",
           "defaultEffort": "low"
         },
+        {
+          "id": "gpt-6.1-terra",
+          "displayName": "GPT-6.1 Terra",
+          "contextWindow": 500000,
+          "supportsTools": true,
+          "supportsThinking": true,
+          "supportsVision": true,
+          "effortLevels": ["low", "medium", "high", "xhigh", "max"],
+          "maxOutputTokens": 128000,
+          "notes": "Listed by Codex's own model list for a ChatGPT Plus account (client 0.166.0, 2026-11-04: priority 3, default effort low). OpenAI API list price $1 / $0.05 cached / $5 per 1M, 1.05M window, 128k output (developers.openai.com/api/docs/models/gpt-6.1-terra, 2026-11-04). The Codex backend takes 500k of context. Codex ultra effort is refused by the backend.",
+          "defaultEffort": "low"
+        },
         {
           "id": "gpt-6-sol",
           "displayName": "GPT-6 Sol",
@@ -585,7 +608,7 @@
         "header": "authorization-bearer",
         "extraHeaders": {
           "x-client-id": "tbh:exec",
-          "user-agent": "muse-build/1.4.13 (non-interactive; linux-x86_64; build 39490bd065c800195489caec34473582f7650b54)"
+          "user-agent": "muse-build/1.4.14 (non-interactive; linux-x86_64; build dec7d7497fecdad135ec5835a8161dbc92e55b58)"
         }
       },
       "defaultModel": "muse-spark-1.3",
@@ -641,9 +664,9 @@
           "supportsTools": true,
           "supportsThinking": true,
           "supportsVision": true,
-          "effortLevels": ["low", "medium", "high", "xhigh"],
+          "effortLevels": ["low", "medium", "high", "xhigh", "max"],
           "maxOutputTokens": 128000,
-          "notes": "Meta may use what you send to this model, and its answers, to train its models (the pool reaches it over a Meta subscription that Meta treats as Contributor use) - do not send confidential or personal data. Cheaper: Meta's contributor-mode price $0.10 input / $0.002 cached / $0.20 output per 1M tokens (dev.meta.ai/docs/pricing-rate-limits, 2026-10-13); contributor rate limit 150 RPM / 3M TPM. Contributor mode: Meta may use the content sent for product improvement (the owner accepted this for the pool, 2026-10-13); sent with the model id's -contributor suffix and the x-tbh-contributor-mode header."
+          "notes": "Meta may use what you send to this model, and its answers, to train its models (the pool reaches it over a Meta subscription that Meta treats as Contributor use) - do not send confidential or personal data. Cheaper: Meta's contributor-mode price $0.10 input / $0.002 cached / $0.20 output per 1M tokens (dev.meta.ai/docs/pricing-rate-limits, 2026-10-13); contributor rate limit 150 RPM / 3M TPM. Contributor mode: Meta may use the content sent for product improvement (the owner accepted this for the pool, 2026-10-13); sent with the model id's -contributor suffix and the x-tbh-contributor-mode header. Effort low to max: Meta's live model list (GET /muse-code/models, 2026-11-04) names max for it, the API answered 200 to max (2026-10-14) and the CLI's own `muse model-profile show muse-spark-1.3-contributor --effort max` resolves it; only Meta's reasoning docs page still says max is not available with this id (docs/effort-levels.md)."
         },
         {
           "id": "muse-spark-1.2-contributor",
@@ -723,7 +746,7 @@
           "kind": "other"
         }
       ],
-      "notes": "OPERATOR POOL ONLY (owner, 2026-10-13: \"integrate into oonacode … muse and codex … on the oonacode server in the pool list, all models … implement authentication flow on the server for both … similar to supergrok and codex with translation layer\"). The operator's own Meta Muse Code subscription, signed in from Admin → Providers → Meta Muse by a REAL Muse CLI on the server (`muse login` in a private HOME; packages/gateway/src/muse/*). The credential row is a POINTER; the static api_key a browser sign-in mints is read from Muse's own ~/.config/muse/auth.json per request (there is no refresh token and no expiry). Chats are NOT run through Muse: the engine's Anthropic request is translated to the Responses wire (@oonacode/translate responses-*.ts — the wire muse itself speaks, POST /v1/responses) and sent to api.meta.ai with Muse's identity headers (recorded 2026-10-13 through a loopback recorder). The pool serves the -contributor ids too (owner, 2026-10-13: cheaper; Muse's contributor mode lets Meta use the content for product improvement, accepted for the pool); a request for one carries x-tbh-contributor-mode: 1 like the CLI's, any other model does not. Meta's subscription terms say the credential is for use with Muse Code only; serving other people from it is the owner's call (2026-10-13), as for the Claude, SuperGrok and ChatGPT plans. A user's OWN Muse key is muse-subscription, direct from their PC. The model list is what GET api.meta.ai/muse-code/models answered for the owner's account on 2026-10-13 (muse-image-1.0 is hidden and image-only; the contributor twins are left out). docs/muse-engine.md.",
+      "notes": "OPERATOR POOL ONLY (owner, 2026-10-13: \"integrate into oonacode … muse and codex … on the oonacode server in the pool list, all models … implement authentication flow on the server for both … similar to supergrok and codex with translation layer\"). The operator's own Meta Muse Code subscription, signed in from Admin → Providers → Meta Muse by a REAL Muse CLI on the server (`muse login` in a private HOME; packages/gateway/src/muse/*). The credential row is a POINTER; the static api_key a browser sign-in mints is read from Muse's own ~/.config/muse/auth.json per request (there is no refresh token and no expiry). Chats are NOT run through Muse: the engine's Anthropic request is translated to the Responses wire (@oonacode/translate responses-*.ts — the wire muse itself speaks, POST /v1/responses) and sent to api.meta.ai with Muse's identity headers (recorded 2026-10-13 through a loopback recorder). The pool serves the -contributor ids (owner, 2026-10-13: cheaper; Muse's contributor mode lets Meta use the content for product improvement, accepted for the pool) and, since 2026-11-04, ONLY those four (the standard ids are in the catalogue but not pooled); every request carries x-tbh-contributor-mode like the CLI's: 1 with a -contributor id, 0 with any other (measured on Muse 1.4.14: contributor mode appends -contributor to an id only when Meta's model list holds the twin). Meta's subscription terms say the credential is for use with Muse Code only; serving other people from it is the owner's call (2026-10-13), as for the Claude, SuperGrok and ChatGPT plans. A user's OWN Muse key is muse-subscription, direct from their PC. The model list is what GET api.meta.ai/muse-code/models answered for the owner's account on 2026-10-13 (muse-image-1.0 is hidden and image-only and is not listed here; the four contributor twins are listed below). docs/muse-engine.md (section 11, \"The setup as it stands\").",
       "credentialVariants": [
         {
           "id": "token-plan",
@@ -754,7 +777,7 @@
         "header": "authorization-bearer",
         "extraHeaders": {
           "x-client-id": "tbh:exec",
-          "user-agent": "muse-build/1.4.13 (non-interactive; linux-x86_64; build 39490bd065c800195489caec34473582f7650b54)"
+          "user-agent": "muse-build/1.4.14 (non-interactive; linux-x86_64; build dec7d7497fecdad135ec5835a8161dbc92e55b58)"
         }
       },
       "models": [
@@ -809,9 +832,9 @@
           "supportsTools": true,
           "supportsThinking": true,
           "supportsVision": true,
-          "effortLevels": ["low", "medium", "high", "xhigh"],
+          "effortLevels": ["low", "medium", "high", "xhigh", "max"],
           "maxOutputTokens": 128000,
-          "notes": "Served from the person's own Muse subscription key through Meta's Responses API (api.meta.ai/v1/responses); the plan's 5-hour windows are what it costs. Cheaper: Meta's contributor-mode price $0.10 input / $0.002 cached / $0.20 output per 1M tokens (dev.meta.ai/docs/pricing-rate-limits, 2026-10-13); contributor rate limit 150 RPM / 3M TPM. Contributor mode: Meta may use the content sent for product improvement (the owner accepted this for the pool, 2026-10-13); sent with the model id's -contributor suffix and the x-tbh-contributor-mode header."
+          "notes": "Served from the person's own Muse subscription key through Meta's Responses API (api.meta.ai/v1/responses); the plan's 5-hour windows are what it costs. Cheaper: Meta's contributor-mode price $0.10 input / $0.002 cached / $0.20 output per 1M tokens (dev.meta.ai/docs/pricing-rate-limits, 2026-10-13); contributor rate limit 150 RPM / 3M TPM. Contributor mode: Meta may use the content sent for product improvement (the owner accepted this for the pool, 2026-10-13); sent with the model id's -contributor suffix and the x-tbh-contributor-mode header. Effort low to max: Meta's live model list (GET /muse-code/models, 2026-11-04) names max for it, the API answered 200 to max (2026-10-14) and the CLI's own `muse model-profile show muse-spark-1.3-contributor --effort max` resolves it; only Meta's reasoning docs page still says max is not available with this id (docs/effort-levels.md)."
         },
         {
           "id": "muse-spark-1.2-contributor",
@@ -1733,7 +1756,7 @@
           "supportsVision": true,
           "inputPricePerM": 0.75,
           "outputPricePerM": 3.75,
-          "notes": "Google lists one model per level — gemini-3.8-flash-high, -medium and -low (fetchAvailableModels, 2026-10-03: 1,048,576 context / 65,536 output each); OonaCode shows ONE model and the chat’s effort picker fills the level (upstreamId template; xhigh and max read as high; no effort = high). Cost = Google’s Gemini API list price for gemini-3.8-flash: $0.75 in / $0.075 cached / $3.75 out per 1M through 2026-12-31, DOUBLING to $1.50 / $0.15 / $7.50 from 2027-01-01 (ai.google.dev/gemini-api/docs/pricing, updated 2026-10-01) — move the cost cells on the model page then."
+          "notes": "Google lists one model per level — gemini-3.8-flash-high, -medium and -low (fetchAvailableModels, 2026-10-03: 1,048,576 context / 65,536 output each); OonaCode shows ONE model and the chat’s effort picker fills the level (upstreamId template; xhigh and max read as high; no effort = high; agy 1.5.0 accepts --effort low, medium or high only and rejects xhigh and max for every model). Asked as agy 1.5.0 asks it: thoughts included and the id's level as thinkingLevel LOW / MEDIUM / HIGH, no budget (agy up to 1.3.x sent Low 1000, Medium 4000, High -1). Cost = Google’s Gemini API list price for gemini-3.8-flash: $0.75 in / $0.075 cached / $3.75 out per 1M through 2026-12-31, DOUBLING to $1.50 / $0.15 / $7.50 from 2027-01-01 (ai.google.dev/gemini-api/docs/pricing, updated 2026-10-01) — move the cost cells on the model page then."
         },
         {
           "id": "gemini-4-argon",
@@ -1746,7 +1769,7 @@
           "supportsVision": true,
           "inputPricePerM": 2,
           "outputPricePerM": 10,
-          "notes": "Google lists one model per level - gemini-4-argon-high, -medium and -low (fetchAvailableModels on the operator's Ultra account, 2026-10-13: 1,048,576 context / 65,536 output each; model enums M417 / M418 / M419); OonaCode shows ONE model and the chat's effort picks the level. Asked like the Google-hosted Claude models: a dynamic thinking budget (-1) AND the thinkingLevel (recorded from agy 1.3.3, three real turns plus a tool call). Ultra accounts only: a non-Ultra personal account does not list it. Google's Gemini API list price is INTRODUCTORY $2 / $0.10 cached / $10 per 1M through 2026-11-30 and $4 / $0.20 / $20 from 2026-12-01 - move the catalogue cells then."
+          "notes": "Google lists one model per level - gemini-4-argon-high, -medium and -low (fetchAvailableModels on the operator's Ultra account, 2026-10-13: 1,048,576 context / 65,536 output each; model enums M417 / M418 / M419); OonaCode shows ONE model and the chat's effort picks the level. Asked like the Google-hosted Claude models: thoughts included and the id's level as thinkingLevel, no budget (agy 1.4.0 and 1.5.0; agy 1.3.3 sent a dynamic budget of -1 beside it). Ultra accounts only: a non-Ultra personal account does not list it. Google's Gemini API list price is INTRODUCTORY $2 / $0.10 cached / $10 per 1M through 2026-11-30 and $4 / $0.20 / $20 from 2026-12-01 - move the catalogue cells then."
         },
         {
           "id": "google-claude-opus-5-5",
@@ -1760,7 +1783,7 @@
           "inputPricePerM": 4,
           "outputPricePerM": 20,
           "effortLevels": ["low", "medium", "high"],
-          "notes": "PRIVATE (owner, 2026-10-07): in no list — served only when this exact id is asked (/model <id>, the picker’s Custom model… row, oona -m <id>); never the pool entry, /v1/models or the API list (its catalogue row is unlisted). Claude through the operator’s Antigravity sign-in under its OWN id, so the pool never folds it with Anthropic’s lanes. Takes chats up to 500,000 tokens, and the gateway declines above that ×1.1 (migration 0227; it was 200,000 from 0177 and 100,000 before, while Google’s prompt cache was broken) - a bigger prompt, and any PDF, is declined before any upstream call), effort low / medium / high (higher reads as high; the id is a template filled from the chat’s effort). Google lists 1,000,000 / 128,000 for every 5.5 level (fetchAvailableModels, 2026-10-06) but its prompt cache barely applies (28k of a 417k prompt) and it meters by token cost, so the ceiling is the pool’s own. Thinking is asked as agy 1.3.0 asks it (budget 0 plus the id’s level). Cost = Anthropic’s list price ($4 / $0.20 cached / $20 per 1M) — Google publishes none."
+          "notes": "PRIVATE (owner, 2026-10-07): in no list — served only when this exact id is asked (/model <id>, the picker’s Custom model… row, oona -m <id>); never the pool entry, /v1/models or the API list (its catalogue row is unlisted). Claude through the operator’s Antigravity sign-in under its OWN id, so the pool never folds it with Anthropic’s lanes. Takes chats up to 500,000 tokens, and the gateway declines above that ×1.1 (migration 0227; it was 200,000 from 0177 and 100,000 before, while Google’s prompt cache was broken) - a bigger prompt, and any PDF, is declined before any upstream call), effort low / medium / high (higher reads as high; the id is a template filled from the chat’s effort). Google lists 1,000,000 / 128,000 for every 5.5 level (fetchAvailableModels, 2026-10-06) but its prompt cache barely applies (28k of a 417k prompt) and it meters by token cost, so the ceiling is the pool’s own. Thinking is asked as agy 1.5.0 asks it (thoughts included and the id’s level as thinkingLevel LOW / MEDIUM / HIGH, no budget; agy 1.3.0 to 1.3.3 also sent a budget of 0). Cost = Anthropic’s list price ($4 / $0.20 cached / $20 per 1M) — Google publishes none."
         },
         {
           "id": "google-claude-sonnet-5-5",
@@ -1774,7 +1797,7 @@
           "inputPricePerM": 2,
           "outputPricePerM": 10,
           "effortLevels": ["low", "medium", "high"],
-          "notes": "PRIVATE (owner, 2026-10-07): in no list — served only when this exact id is asked (/model <id>, the picker’s Custom model… row, oona -m <id>); never the pool entry, /v1/models or the API list (its catalogue row is unlisted). Claude through the operator’s Antigravity sign-in under its OWN id, so the pool never folds it with Anthropic’s lanes. Takes chats up to 500,000 tokens, and the gateway declines above that ×1.1 (migration 0227; it was 200,000 from 0177 and 100,000 before, while Google’s prompt cache was broken) - a bigger prompt, and any PDF, is declined before any upstream call), effort low / medium / high (higher reads as high; the id is a template filled from the chat’s effort). Google lists 1,000,000 / 128,000 for every 5.5 level (fetchAvailableModels, 2026-10-06) but its prompt cache barely applies (28k of a 417k prompt) and it meters by token cost, so the ceiling is the pool’s own. Thinking is asked as agy 1.3.0 asks it (budget 0 plus the id’s level). Cost = Anthropic’s list price ($2 / $0.20 cached / $10 per 1M) — Google publishes none."
+          "notes": "PRIVATE (owner, 2026-10-07): in no list — served only when this exact id is asked (/model <id>, the picker’s Custom model… row, oona -m <id>); never the pool entry, /v1/models or the API list (its catalogue row is unlisted). Claude through the operator’s Antigravity sign-in under its OWN id, so the pool never folds it with Anthropic’s lanes. Takes chats up to 500,000 tokens, and the gateway declines above that ×1.1 (migration 0227; it was 200,000 from 0177 and 100,000 before, while Google’s prompt cache was broken) - a bigger prompt, and any PDF, is declined before any upstream call), effort low / medium / high (higher reads as high; the id is a template filled from the chat’s effort). Google lists 1,000,000 / 128,000 for every 5.5 level (fetchAvailableModels, 2026-10-06) but its prompt cache barely applies (28k of a 417k prompt) and it meters by token cost, so the ceiling is the pool’s own. Thinking is asked as agy 1.5.0 asks it (thoughts included and the id’s level as thinkingLevel LOW / MEDIUM / HIGH, no budget; agy 1.3.0 to 1.3.3 also sent a budget of 0). Cost = Anthropic’s list price ($2 / $0.20 cached / $10 per 1M) — Google publishes none."
         }
       ],
       "defaultModel": "gemini-3.8-flash",
@@ -1828,7 +1851,7 @@
           "operatorOnly": true,
           "baseUrl": "https://daily-cloudcode-pa.googleapis.com",
           "extraHeaders": {
-            "user-agent": "antigravity/cli/1.4.7 (aidev_client; os_type=linux; arch=amd64; cl=1006008713; auth_method=consumer)"
+            "user-agent": "antigravity/cli/1.5.0 (aidev_client; os_type=linux; arch=amd64; cl=1013440158; auth_method=consumer)"
           },
           "agyLogin": {
             "type": "agy-server",