@oh-my-pi/pi-ai 13.14.0 → 13.15.2

package/CHANGELOG.md

@@ -2,6 +2,33 @@
 
 ## [Unreleased]
 
+## [13.15.0] - 2026-03-23
+
+### Added
+
+- Added `isUsageLimitError()` to `rate-limit-utils` as a single source of truth for detecting usage/quota limit errors across all providers
+
+### Fixed
+
+- Fixed lazy stream forwarding to properly handle final results from source streams with `result()` methods
+- Fixed lazy stream error handling to convert iterator failures into terminal error results instead of silently failing
+- Fixed `parseRateLimitReason` to recognize "usage limit" in error messages and correctly classify them as `QUOTA_EXHAUSTED`
+- Fixed Codex `fetchWithRetry` retrying 429 responses for `usage_limit_reached` errors for up to 5 minutes instead of returning immediately for credential switching
+- Removed `usage.?limit` from `TRANSIENT_MESSAGE_PATTERN` in retry utils since usage limits are not transient and require credential rotation
+- Fixed `parseRateLimitReason` not recognizing "usage limit" in Codex error messages, causing incorrect fallback to `UNKNOWN` classification instead of `QUOTA_EXHAUSTED`
+
+## [13.14.2] - 2026-03-21
+### Changed
+
+- Updated thinking configuration format from `levels` array to `minLevel` and `maxLevel` properties for improved clarity
+- Corrected context window from 400000 to 272000 tokens for GPT-5.4 mini and nano variants on Codex transport
+- Normalized GPT-5.4 variant priority handling to use parsed variant instead of special-casing raw model IDs
+- Added support for `mini` variant in OpenAI model parsing regex
+
+### Fixed
+
+- Fixed inconsistent thinking level configuration across multiple model definitions
+
 ## [13.14.0] - 2026-03-20
 
 ### Fixed

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-ai",
-	"version": "13.14.0",
+	"version": "13.15.2",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"homepage": "https://github.com/can1357/oh-my-pi",
 	"author": "Can Boluk",
@@ -41,7 +41,7 @@
 		"@aws-sdk/client-bedrock-runtime": "^3",
 		"@bufbuild/protobuf": "^2.11",
 		"@google/genai": "^1.43",
-		"@oh-my-pi/pi-utils": "13.14.0",
+		"@oh-my-pi/pi-utils": "13.15.2",
 		"@sinclair/typebox": "^0.34",
 		"@smithy/node-http-handler": "^4.4",
 		"ajv": "^8.18",

package/src/index.ts

@@ -16,6 +16,7 @@ export * from "./providers/google";
 export * from "./providers/google-gemini-cli";
 export * from "./providers/google-vertex";
 export * from "./providers/kimi";
+export type { OpenAICodexResponsesOptions } from "./providers/openai-codex-responses";
 export * from "./providers/openai-completions";
 export * from "./providers/openai-responses";
 export * from "./providers/synthetic";

package/src/model-thinking.ts

@@ -40,7 +40,13 @@ type SemVer = {
 
 type GeminiKind = "pro" | "flash";
 type AnthropicKind = "opus" | "sonnet";
-type OpenAIVariant = "base" | "codex" | "codex-max" | "codex-mini" | "codex-spark" | "max" | "nano";
+type OpenAIVariant = "base" | "codex" | "codex-max" | "codex-mini" | "codex-spark" | "mini" | "max" | "nano";
+
+const CODEX_GPT_5_4_PRIORITY_BY_VARIANT: Partial<Record<OpenAIVariant, number>> = {
+	base: 0,
+	mini: 1,
+	nano: 2,
+};
 
 interface GeminiModel {
 	family: "gemini";
@@ -299,9 +305,17 @@ function applyOpenAICatalogPolicy(model: ApiModel<Api>, parsedModel: OpenAIModel
 		return;
 	}
 	// GPT-5.4 mini/nano use plain OpenAI IDs on the Codex transport, but Codex still
-	// enforces the lower prompt budget for these variants.
-	if (model.api === "openai-codex-responses" && (model.id === "gpt-5.4-mini" || model.id === "gpt-5.4-nano")) {
-		model.contextWindow = 272000;
+	// enforces the lower prompt budget for these variants. Codex discovery can also
+	// report inconsistent priorities for the GPT-5.4 family, so normalize by parsed
+	// variant instead of special-casing raw model ids.
+	if (model.api === "openai-codex-responses" && semverEqual(parsedModel.version, "5.4")) {
+		const normalizedPriority = CODEX_GPT_5_4_PRIORITY_BY_VARIANT[parsedModel.variant];
+		if (normalizedPriority !== undefined) {
+			model.priority = normalizedPriority;
+		}
+		if (parsedModel.variant === "mini" || parsedModel.variant === "nano") {
+			model.contextWindow = 272000;
+		}
 	}
 }
 
@@ -489,7 +503,7 @@ function parseAnthropicModel(modelId: string): AnthropicModel | null {
 }
 
 function parseOpenAIModel(modelId: string): OpenAIModel | null {
-	const match = /gpt-(\d+(?:\.\d+){0,2})(?:-(codex-spark|codex-mini|codex-max|codex|max|nano))?\b/.exec(modelId);
+	const match = /gpt-(\d+(?:\.\d+){0,2})(?:-(codex-spark|codex-mini|codex-max|codex|mini|max|nano))?\b/.exec(modelId);
 	if (!match) {
 		return null;
 	}