theclawbay 0.5.0 → 0.5.1

package/dist/lib/clients/codex.d.ts

@@ -4,6 +4,8 @@ export declare const MANAGED_START = "# theclawbay-managed:start";
 export declare const MANAGED_END = "# theclawbay-managed:end";
 export declare const HISTORY_PROVIDER_NEUTRALIZE_SOURCES: Set<string>;
 export declare const HISTORY_PROVIDER_DB_MIGRATE_SOURCES: string[];
+export declare const CODEX_MODEL_CONTEXT_WINDOW = 272000;
+export declare const CODEX_MODEL_AUTO_COMPACT_TOKEN_LIMIT = 220000;
 export type CodexFeatureFlagSnapshot = {
     hadFeaturesTable: boolean;
     appsHadKey: boolean;

package/dist/lib/clients/codex.js

@@ -3,7 +3,7 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
     return (mod && mod.__esModule) ? mod : { "default": mod };
 };
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.HISTORY_PROVIDER_DB_MIGRATE_SOURCES = exports.HISTORY_PROVIDER_NEUTRALIZE_SOURCES = exports.MANAGED_END = exports.MANAGED_START = exports.MIGRATION_STATE_FILE = exports.CODEX_FEATURE_FLAGS_STATE_PATH = void 0;
+exports.CODEX_MODEL_AUTO_COMPACT_TOKEN_LIMIT = exports.CODEX_MODEL_CONTEXT_WINDOW = exports.HISTORY_PROVIDER_DB_MIGRATE_SOURCES = exports.HISTORY_PROVIDER_NEUTRALIZE_SOURCES = exports.MANAGED_END = exports.MANAGED_START = exports.MIGRATION_STATE_FILE = exports.CODEX_FEATURE_FLAGS_STATE_PATH = void 0;
 exports.removeProviderTable = removeProviderTable;
 exports.restoreManagedCodexFeatureFlags = restoreManagedCodexFeatureFlags;
 exports.ensureManagedCodexFeatureFlags = ensureManagedCodexFeatureFlags;
@@ -28,6 +28,14 @@ exports.MANAGED_START = "# theclawbay-managed:start";
 exports.MANAGED_END = "# theclawbay-managed:end";
 exports.HISTORY_PROVIDER_NEUTRALIZE_SOURCES = new Set(["openai", "theclawbay-wan", providers_1.DEFAULT_PROVIDER_ID]);
 exports.HISTORY_PROVIDER_DB_MIGRATE_SOURCES = ["openai", providers_1.DEFAULT_PROVIDER_ID, "theclawbay-wan"];
+// Codex auto-compaction only triggers when it knows the model's context window.
+// In API-key + custom base_url mode there is no server-provided window, so unless
+// we write these, `model_context_window()` is None and compaction never fires —
+// the conversation grows unbounded until it hits the relay's request-size cap.
+// The auto-compact limit (~80% of window) keeps even image-heavy sessions well
+// under the cap while matching Codex's default behavior for gpt-5.x.
+exports.CODEX_MODEL_CONTEXT_WINDOW = 272000;
+exports.CODEX_MODEL_AUTO_COMPACT_TOKEN_LIMIT = 220000;
 function removeProviderTable(source, providerId) {
     const header = `[model_providers.${providerId}]`;
     const lines = source.split(/\r?\n/);
@@ -158,6 +166,10 @@ async function writeCodexConfig(params) {
     next = (0, toml_1.upsertFirstKeyLine)(next, "model_provider", `"${providers_1.DEFAULT_PROVIDER_ID}"`);
     next = (0, toml_1.upsertFirstKeyLine)(next, "model", `"${params.model}"`);
     next = (0, toml_1.upsertFirstKeyLine)(next, "openai_base_url", `"${nativeCodexBaseUrl}"`);
+    // Give Codex the context window it needs to trigger auto-compaction; without
+    // these, custom-base_url sessions never compact and grow until the relay 413s.
+    next = (0, toml_1.upsertFirstKeyLine)(next, "model_context_window", String(exports.CODEX_MODEL_CONTEXT_WINDOW));
+    next = (0, toml_1.upsertFirstKeyLine)(next, "model_auto_compact_token_limit", String(exports.CODEX_MODEL_AUTO_COMPACT_TOKEN_LIMIT));
     next = (0, toml_1.removeTopLevelKeyLineIf)(next, "chatgpt_base_url", urls_1.isTheClawBayChatgptBaseUrl);
     const managedBlock = (0, fsx_1.appendManagedBlock)("", [
         exports.MANAGED_START,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "theclawbay",
-  "version": "0.5.0",
+  "version": "0.5.1",
   "description": "CLI for connecting Codex, Hermes Agent, Gemini-compatible apps, Continue, Cline, GSD, OpenClaw, OpenCode, Kilo, Roo Code, Aider, experimental Trae, and experimental Zo to The Claw Bay.",
   "license": "MIT",
   "repository": {

package/theclawbay-supported-models.json

@@ -45,9 +45,9 @@
     "note": "Google's flagship coding and reasoning model through CliRelay.",
     "tone": "coral",
     "provider": "google",
-    "inputPer1M": 1.25,
-    "cachedInputPer1M": 0.125,
-    "outputPer1M": 10.0
+    "inputPer1M": 1.16,
+    "cachedInputPer1M": 0.12,
+    "outputPer1M": 9.25
   },
   {
     "id": "gemini-2.5-flash",
@@ -55,9 +55,9 @@
     "note": "Balanced Gemini path for fast multimodal and agentic work.",
     "tone": "ink",
     "provider": "google",
-    "inputPer1M": 0.3,
-    "cachedInputPer1M": 0.03,
-    "outputPer1M": 2.5
+    "inputPer1M": 0.89,
+    "cachedInputPer1M": 0.09,
+    "outputPer1M": 7.4
   },
   {
     "id": "gemini-2.5-flash-lite",
@@ -65,9 +65,9 @@
     "note": "Lowest-cost Gemini path for lightweight coding and tooling loops.",
     "tone": "sea",
     "provider": "google",
-    "inputPer1M": 0.1,
-    "cachedInputPer1M": 0.01,
-    "outputPer1M": 0.4
+    "inputPer1M": 0.3,
+    "cachedInputPer1M": 0.03,
+    "outputPer1M": 1.19
   },
   {
     "id": "gemini-3-pro-preview",
@@ -75,9 +75,9 @@
     "note": "Preview Gemini model for heavier multimodal and coding sessions.",
     "tone": "coral",
     "provider": "google",
-    "inputPer1M": 2.0,
-    "cachedInputPer1M": 0.2,
-    "outputPer1M": 12.0
+    "inputPer1M": 1.85,
+    "cachedInputPer1M": 0.19,
+    "outputPer1M": 11.1
   },
   {
     "id": "gemini-3.1-pro-preview",
@@ -85,9 +85,9 @@
     "note": "Newest Gemini Pro preview with stronger agentic and coding behavior.",
     "tone": "coral",
     "provider": "google",
-    "inputPer1M": 2.0,
-    "cachedInputPer1M": 0.2,
-    "outputPer1M": 12.0
+    "inputPer1M": 1.85,
+    "cachedInputPer1M": 0.02,
+    "outputPer1M": 11.1
   },
   {
     "id": "gemini-3-flash-preview",
@@ -95,9 +95,9 @@
     "note": "Faster Gemini preview tuned for responsive high-volume tasks.",
     "tone": "ink",
     "provider": "google",
-    "inputPer1M": 0.5,
+    "inputPer1M": 0.47,
     "cachedInputPer1M": 0.05,
-    "outputPer1M": 3.0
+    "outputPer1M": 2.78
   },
   {
     "id": "deepseek-v4-flash",