@gajae-code/ai 0.1.3 → 0.2.1

package/CHANGELOG.md

@@ -2,6 +2,19 @@
 
 ## [Unreleased]
 
+## [0.2.1] - 2026-05-30
+
+### Changed
+
+- Refreshed AI package metadata for the GJC 0.2.1 release.
+
+## [0.2.0] - 2026-05-28
+
+### Fixed
+
+- Fixed OpenAI-compatible base URL handling so configured proxy URLs and inherited environment overrides are respected at model discovery, completions, responses, and streaming call sites.
+- Fixed OpenAI direct-provider feature gates so prompt-cache/session behavior uses the resolved Responses base URL and only treats exact default `api.openai.com` hosts/paths as direct OpenAI.
+
 ## [0.1.3] - 2026-05-28
 
 ### Changed

package/dist/types/providers/openai-request-transform.d.ts

@@ -0,0 +1,4 @@
+import type { FetchImpl, ModelRequestTransform } from "../types";
+export declare function applyOpenAIRequestTransformHeaders(headers: Record<string, string>, transform: ModelRequestTransform | undefined, profileUserAgent: string): Record<string, string>;
+export declare function applyOpenAIRequestTransformBody(params: object, transform: ModelRequestTransform | undefined): void;
+export declare function wrapFetchForOpenAIRequestTransform(baseFetch: FetchImpl, transform: ModelRequestTransform | undefined, profileUserAgent: string): FetchImpl;

package/dist/types/types.d.ts

@@ -689,6 +689,16 @@ export interface VercelGatewayRouting {
     /** List of provider slugs to try in order (e.g., ["anthropic", "openai"]). */
     order?: string[];
 }
+export interface ModelRequestTransform {
+    /** Named request-shaping preset. `openai-proxy` removes OpenAI SDK telemetry headers and uses a generic Gajae-Code User-Agent. */
+    profile?: "openai-proxy";
+    /** Header names to remove from the final outbound request. Case-insensitive. */
+    stripHeaders?: string[];
+    /** Headers to set after stripping; use null to remove a header explicitly. */
+    setHeaders?: Record<string, string | null>;
+    /** Extra request body fields merged after provider defaults; protected core request keys are ignored. */
+    extraBody?: Record<string, unknown>;
+}
 export interface Model<TApi extends Api = any> {
     id: string;
     name: string;
@@ -727,6 +737,10 @@ export interface Model<TApi extends Api = any> {
     preferWebsockets?: boolean;
     /** Preferred model to switch to when context promotion is triggered (model id or provider/id). */
     contextPromotionTarget?: string;
+    /** Provider-facing model id when it differs from the local selector id. */
+    wireModelId?: string;
+    /** Declarative request shaping for OpenAI-compatible proxy providers. */
+    requestTransform?: ModelRequestTransform;
     /** Provider-assigned priority value (lower = higher priority). */
     priority?: number;
     /** Canonical thinking capability metadata for this model. */

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@gajae-code/ai",
-	"version": "0.1.3",
+	"version": "0.2.1",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"homepage": "https://gaebal-gajae.dev",
 	"author": "Yeachan-Heo",
@@ -43,7 +43,7 @@
 	"dependencies": {
 		"@anthropic-ai/sdk": "^0.94.0",
 		"@bufbuild/protobuf": "^2.12.0",
-		"@gajae-code/utils": "0.1.3",
+		"@gajae-code/utils": "0.2.1",
 		"openai": "^6.36.0",
 		"partial-json": "^0.1.7",
 		"zod": "4.4.3"

package/src/provider-models/openai-compat.ts

@@ -1,4 +1,4 @@
-import { $env } from "@gajae-code/utils";
+import { $env, $inheritedEnv } from "@gajae-code/utils";
 import type { ModelManagerOptions } from "../model-manager";
 import { Effort } from "../model-thinking";
 import { getBundledModels } from "../models";
@@ -496,7 +496,11 @@ export interface OpenAIModelManagerConfig {
 
 export function openaiModelManagerOptions(config?: OpenAIModelManagerConfig): ModelManagerOptions<"openai-responses"> {
 	const apiKey = config?.apiKey;
-	const baseUrl = config?.baseUrl?.trim() || $env.OPENAI_BASE_URL?.trim() || OPENAI_DEFAULT_BASE_URL;
+	const baseUrl =
+		config?.baseUrl?.trim() ||
+		$inheritedEnv("OPENAI_BASE_URL") ||
+		$env.OPENAI_BASE_URL?.trim() ||
+		OPENAI_DEFAULT_BASE_URL;
 	const references = createBundledReferenceMap<"openai-responses">("openai");
 	return {
 		providerId: "openai",

package/src/providers/google-gemini-headers.ts

@@ -4,7 +4,7 @@
  * GeminiCLI/VERSION/MODEL (PLATFORM; ARCH; SURFACE)
  */
 export function getGeminiCliUserAgent(modelId = "gemini-3.1-pro-preview"): string {
-	const version = process.env.PI_AI_GEMINI_CLI_VERSION || "0.35.3";
+	const version = process.env.PI_AI_GEMINI_CLI_VERSION || "0.44.1";
 	const platform = process.platform === "win32" ? "win32" : process.platform;
 	const arch = process.arch === "x64" ? "x64" : process.arch;
 	return `GeminiCLI/${version}/${modelId} (${platform}; ${arch}; terminal)`;

package/src/providers/openai-completions.ts

@@ -1,4 +1,4 @@
-import { $env, extractHttpStatusFromError } from "@gajae-code/utils";
+import { $env, $inheritedEnv, extractHttpStatusFromError } from "@gajae-code/utils";
 import OpenAI from "openai";
 import type {
 	ChatCompletionAssistantMessageParam,
@@ -66,20 +66,35 @@ import {
 	resolveGitHubCopilotBaseUrl,
 } from "./github-copilot-headers";
 import { detectOpenAICompat, type ResolvedOpenAICompat, resolveOpenAICompat } from "./openai-completions-compat";
+import {
+	applyOpenAIRequestTransformBody,
+	applyOpenAIRequestTransformHeaders,
+	wrapFetchForOpenAIRequestTransform,
+} from "./openai-request-transform";
 import { createInitialResponsesAssistantMessage } from "./openai-responses-shared";
 import { transformMessages } from "./transform-messages";
 import { joinTextWithImagePlaceholder, NON_VISION_IMAGE_PLACEHOLDER } from "./vision-guard";
 
 const OPENAI_DEFAULT_BASE_URL = "https://api.openai.com/v1";
+const OPENAI_DEFAULT_BASE_URL_HOST = "api.openai.com";
+
+function isDefaultOpenAIBaseUrl(baseUrl: string): boolean {
+	try {
+		const url = new URL(baseUrl);
+		return url.hostname === OPENAI_DEFAULT_BASE_URL_HOST && (url.pathname === "" || url.pathname === "/v1");
+	} catch {
+		return baseUrl === OPENAI_DEFAULT_BASE_URL;
+	}
+}
 
 function resolveOpenAIProviderBaseUrl(
 	baseUrl: string | undefined,
 	authCredentialType: "api_key" | "oauth" | undefined,
 ): string {
 	if (authCredentialType === "oauth") return OPENAI_DEFAULT_BASE_URL;
-	const envBaseUrl = $env.OPENAI_BASE_URL?.trim();
+	const envBaseUrl = $inheritedEnv("OPENAI_BASE_URL") ?? $env.OPENAI_BASE_URL?.trim();
 	const configuredBaseUrl = baseUrl?.trim();
-	if (envBaseUrl && (!configuredBaseUrl || configuredBaseUrl.toLowerCase().includes("api.openai.com"))) {
+	if (envBaseUrl && (!configuredBaseUrl || isDefaultOpenAIBaseUrl(configuredBaseUrl))) {
 		return envBaseUrl;
 	}
 	return configuredBaseUrl || envBaseUrl || OPENAI_DEFAULT_BASE_URL;
@@ -946,6 +961,7 @@ async function createClient(
 	if (model.provider === "kimi-code") {
 		headers = { ...getKimiCommonHeaders(), ...headers };
 	}
+	headers = applyOpenAIRequestTransformHeaders(headers, model.requestTransform, `Gajae-Code/${packageJson.version}`);
 	let copilotPremiumRequests: number | undefined;
 
 	let baseUrl =
@@ -1003,7 +1019,14 @@ async function createClient(
 		},
 		baseFetch.preconnect ? { preconnect: baseFetch.preconnect } : {},
 	);
-	const debugFetch = onSseEvent ? wrapFetchForSseDebug(wrappedFetch, event => onSseEvent(event, model)) : wrappedFetch;
+	const transformedFetch = wrapFetchForOpenAIRequestTransform(
+		wrappedFetch,
+		model.requestTransform,
+		`Gajae-Code/${packageJson.version}`,
+	);
+	const debugFetch = onSseEvent
+		? wrapFetchForSseDebug(transformedFetch, event => onSseEvent(event, model))
+		: transformedFetch;
 	// Bound HTTP request timeout to roughly the first-event watchdog window.
 	// The OpenAI SDK's default is 10 minutes per attempt × `maxRetries`, which
 	// turns a stalled-before-headers fetch into a multi-minute hang invisible
@@ -1068,11 +1091,12 @@ function buildParams(
 	const effectiveMaxTokens = options?.maxTokens ?? (isKimi ? model.maxTokens : undefined);
 
 	const requestModelId =
-		model.provider === "fireworks"
+		model.wireModelId ??
+		(model.provider === "fireworks"
 			? toFireworksWireModelId(model.id)
 			: model.provider === "firepass"
 				? toFirepassWireModelId(model.id)
-				: model.id;
+				: model.id);
 	const params: OpenAICompletionsParams = {
 		model: requestModelId,
 		messages,
@@ -1250,6 +1274,7 @@ function buildParams(
 	if (compat.extraBody) {
 		Object.assign(params, compat.extraBody);
 	}
+	applyOpenAIRequestTransformBody(params, model.requestTransform);
 
 	return { params, toolStrictMode };
 }

package/src/providers/openai-request-transform.ts

@@ -0,0 +1,135 @@
+import type { FetchImpl, ModelRequestTransform } from "../types";
+
+const PROTECTED_EXTRA_BODY_KEYS = new Set([
+	"model",
+	"messages",
+	"input",
+	"instructions",
+	"stream",
+	"stream_options",
+	"store",
+	"max_tokens",
+	"max_completion_tokens",
+	"max_output_tokens",
+	"temperature",
+	"top_p",
+	"presence_penalty",
+	"frequency_penalty",
+	"reasoning",
+	"reasoning_effort",
+	"prompt_cache_key",
+	"prompt_cache_retention",
+	"service_tier",
+	"stop",
+	"tools",
+	"tool_choice",
+	"parallel_tool_calls",
+]);
+
+const OPENAI_PROXY_STRIP_HEADERS = [
+	"x-stainless-arch",
+	"x-stainless-async",
+	"x-stainless-lang",
+	"x-stainless-os",
+	"x-stainless-package-version",
+	"x-stainless-retry-count",
+	"x-stainless-runtime",
+	"x-stainless-runtime-version",
+	"x-stainless-timeout",
+	"x-stainless-helper-method",
+	"openai-organization",
+	"openai-project",
+] as const;
+
+function resolveRequestTransform(
+	transform: ModelRequestTransform | undefined,
+	profileUserAgent: string,
+): ModelRequestTransform | undefined {
+	if (!transform) return undefined;
+	const profileTransform: ModelRequestTransform =
+		transform.profile === "openai-proxy"
+			? {
+					stripHeaders: [...OPENAI_PROXY_STRIP_HEADERS],
+					setHeaders: { "User-Agent": profileUserAgent },
+				}
+			: {};
+	return {
+		...profileTransform,
+		...transform,
+		stripHeaders: transform.stripHeaders ?? profileTransform.stripHeaders,
+		setHeaders: transform.setHeaders
+			? { ...(profileTransform.setHeaders ?? {}), ...transform.setHeaders }
+			: profileTransform.setHeaders,
+		extraBody: transform.extraBody,
+	};
+}
+
+function deleteHeaders(headers: Headers, names: readonly string[] | undefined): void {
+	for (const name of names ?? []) {
+		headers.delete(name);
+	}
+}
+
+function setHeaders(headers: Headers, values: Record<string, string | null> | undefined): void {
+	for (const [name, value] of Object.entries(values ?? {})) {
+		if (value === null) {
+			headers.delete(name);
+		} else {
+			headers.set(name, value);
+		}
+	}
+}
+
+function transformHeaders(
+	headers: RequestInit["headers"] | undefined,
+	transform: ModelRequestTransform | undefined,
+): Headers {
+	const result = new Headers(headers);
+	deleteHeaders(result, transform?.stripHeaders);
+	setHeaders(result, transform?.setHeaders);
+	return result;
+}
+
+export function applyOpenAIRequestTransformHeaders(
+	headers: Record<string, string>,
+	transform: ModelRequestTransform | undefined,
+	profileUserAgent: string,
+): Record<string, string> {
+	const resolved = resolveRequestTransform(transform, profileUserAgent);
+	if (!resolved) return headers;
+	return Object.fromEntries(transformHeaders(headers, resolved).entries());
+}
+
+export function applyOpenAIRequestTransformBody(params: object, transform: ModelRequestTransform | undefined): void {
+	if (!transform?.extraBody) return;
+	const body = params as Record<string, unknown>;
+	for (const [key, value] of Object.entries(transform.extraBody)) {
+		if (!PROTECTED_EXTRA_BODY_KEYS.has(key) && !(key in body)) {
+			body[key] = value;
+		}
+	}
+}
+
+export function wrapFetchForOpenAIRequestTransform(
+	baseFetch: FetchImpl,
+	transform: ModelRequestTransform | undefined,
+	profileUserAgent: string,
+): FetchImpl {
+	const resolved = resolveRequestTransform(transform, profileUserAgent);
+	if (!resolved) return baseFetch;
+	return Object.assign(
+		async (input: string | URL | Request, init?: RequestInit): Promise<Response> => {
+			if (input instanceof Request) {
+				const request = new Request(input, init);
+				deleteHeaders(request.headers, resolved.stripHeaders);
+				setHeaders(request.headers, resolved.setHeaders);
+				return baseFetch(request);
+			}
+			return baseFetch(input, {
+				...init,
+				headers: transformHeaders(init?.headers, resolved),
+			});
+		},
+		baseFetch.preconnect ? { preconnect: baseFetch.preconnect } : {},
+	);
+}

package/src/providers/openai-responses.ts

@@ -1,10 +1,11 @@
-import { $env, extractHttpStatusFromError, structuredCloneJSON } from "@gajae-code/utils";
+import { $env, $inheritedEnv, extractHttpStatusFromError, structuredCloneJSON } from "@gajae-code/utils";
 import OpenAI from "openai";
 import type {
 	Tool as OpenAITool,
 	ResponseCreateParamsStreaming,
 	ResponseInput,
 } from "openai/resources/responses/responses";
+import packageJson from "../../package.json" with { type: "json" };
 import { getEnvApiKey } from "../stream";
 import type {
 	AssistantMessage,
@@ -50,6 +51,11 @@ import {
 	resolveGitHubCopilotBaseUrl,
 } from "./github-copilot-headers";
 import { compactGrammarDefinition } from "./grammar";
+import {
+	applyOpenAIRequestTransformBody,
+	applyOpenAIRequestTransformHeaders,
+	wrapFetchForOpenAIRequestTransform,
+} from "./openai-request-transform";
 import {
 	appendResponsesToolResultMessages,
 	applyCommonResponsesSamplingParams,
@@ -69,11 +75,11 @@ import { transformMessages } from "./transform-messages";
  * Get prompt cache retention based on cacheRetention and base URL.
  * Only applies to direct OpenAI API calls (api.openai.com).
  */
-function getPromptCacheRetention(baseUrl: string, cacheRetention: CacheRetention): "24h" | undefined {
+function getPromptCacheRetention(baseUrl: string | undefined, cacheRetention: CacheRetention): "24h" | undefined {
 	if (cacheRetention !== "long") {
 		return undefined;
 	}
-	if (baseUrl.includes("api.openai.com")) {
+	if (baseUrl && isDefaultOpenAIBaseUrl(baseUrl)) {
 		return "24h";
 	}
 	return undefined;
@@ -103,15 +109,34 @@ const OPENAI_RESPONSES_PROVIDER_SESSION_STATE_PREFIX = "openai-responses:";
 const OPENAI_RESPONSES_FIRST_EVENT_TIMEOUT_MESSAGE =
 	"OpenAI responses stream timed out while waiting for the first event";
 const OPENAI_DEFAULT_BASE_URL = "https://api.openai.com/v1";
+const OPENAI_DEFAULT_BASE_URL_HOST = "api.openai.com";
+
+function isDefaultOpenAIBaseUrl(baseUrl: string): boolean {
+	try {
+		const url = new URL(baseUrl);
+		return url.hostname === OPENAI_DEFAULT_BASE_URL_HOST && (url.pathname === "" || url.pathname === "/v1");
+	} catch {
+		return baseUrl === OPENAI_DEFAULT_BASE_URL;
+	}
+}
+
+function isOpenAIHostBaseUrl(baseUrl: string): boolean {
+	try {
+		const url = new URL(baseUrl);
+		return url.hostname === OPENAI_DEFAULT_BASE_URL_HOST;
+	} catch {
+		return baseUrl.toLowerCase().startsWith(OPENAI_DEFAULT_BASE_URL);
+	}
+}
 
 function resolveOpenAIProviderBaseUrl(
 	baseUrl: string | undefined,
 	authCredentialType: "api_key" | "oauth" | undefined,
 ): string {
 	if (authCredentialType === "oauth") return OPENAI_DEFAULT_BASE_URL;
-	const envBaseUrl = $env.OPENAI_BASE_URL?.trim();
+	const envBaseUrl = $inheritedEnv("OPENAI_BASE_URL") ?? $env.OPENAI_BASE_URL?.trim();
 	const configuredBaseUrl = baseUrl?.trim();
-	if (envBaseUrl && (!configuredBaseUrl || configuredBaseUrl.toLowerCase().includes("api.openai.com"))) {
+	if (envBaseUrl && (!configuredBaseUrl || isDefaultOpenAIBaseUrl(configuredBaseUrl))) {
 		return envBaseUrl;
 	}
 	return configuredBaseUrl || envBaseUrl || OPENAI_DEFAULT_BASE_URL;
@@ -344,11 +369,18 @@ function createClient(
 	}
 	const rawApiKey = apiKey;
 
-	const headers = { ...(model.headers ?? {}), ...(extraHeaders ?? {}) };
+	const headers = applyOpenAIRequestTransformHeaders(
+		{ ...(model.headers ?? {}), ...(extraHeaders ?? {}) },
+		model.requestTransform,
+		`Gajae-Code/${packageJson.version}`,
+	);
 	let copilotPremiumRequests: number | undefined;
 
 	let baseUrl =
 		model.provider === "openai" ? resolveOpenAIProviderBaseUrl(model.baseUrl, authCredentialType) : model.baseUrl;
+	if (model.provider === "openai" && !baseUrl) {
+		baseUrl = OPENAI_DEFAULT_BASE_URL;
+	}
 	if (model.provider === "github-copilot") {
 		apiKey = parseGitHubCopilotApiKey(rawApiKey).accessToken;
 		const hasImages = hasCopilotVisionInput(context.messages);
@@ -363,11 +395,16 @@ function createClient(
 		copilotPremiumRequests = copilot.premiumRequests;
 		baseUrl = resolveGitHubCopilotBaseUrl(model.baseUrl, rawApiKey) ?? model.baseUrl;
 	}
-	if (sessionId && model.provider === "openai" && (baseUrl ?? "").toLowerCase().includes("api.openai.com")) {
+	if (sessionId && model.provider === "openai" && (!model.baseUrl || (baseUrl && isDefaultOpenAIBaseUrl(baseUrl)))) {
 		headers.session_id ??= sessionId;
 		headers["x-client-request-id"] ??= sessionId;
 	}
 	const baseFetch = fetchOverride ?? fetch;
+	const transformedFetch = wrapFetchForOpenAIRequestTransform(
+		baseFetch,
+		model.requestTransform,
+		`Gajae-Code/${packageJson.version}`,
+	);
 	return {
 		client: new OpenAI({
 			apiKey,
@@ -375,7 +412,9 @@ function createClient(
 			dangerouslyAllowBrowser: true,
 			maxRetries: 5,
 			defaultHeaders: headers,
-			fetch: onSseEvent ? wrapFetchForSseDebug(baseFetch, event => onSseEvent(event, model)) : baseFetch,
+			fetch: onSseEvent
+				? wrapFetchForSseDebug(transformedFetch, event => onSseEvent(event, model))
+				: transformedFetch,
 		}),
 		copilotPremiumRequests,
 		baseUrl,
@@ -411,7 +450,9 @@ function buildParams(
 	const systemPrompts = normalizeSystemPrompts(context.systemPrompt);
 	let systemInstructions: string | undefined;
 	if (systemPrompts.length > 0) {
-		const needsDeveloperRole = model.reasoning && supportsDeveloperRole(resolvedBaseUrl ?? model);
+		const needsDeveloperRole =
+			model.reasoning &&
+			((model.provider === "openai" && !model.baseUrl) || supportsDeveloperRole(resolvedBaseUrl || model));
 		if (needsDeveloperRole) {
 			// Reasoning models on known OpenAI-compatible endpoints require the
 			// `developer` role. Send all system prompts inline in `input`.
@@ -429,12 +470,14 @@ function buildParams(
 	const cacheRetention = resolveCacheRetention(options?.cacheRetention);
 	const promptCacheKey = getOpenAIResponsesCacheSessionId(options);
 	const params: OpenAIResponsesSamplingParams = {
-		model: model.id,
+		model: model.wireModelId ?? model.id,
 		input: messages,
 		instructions: systemInstructions,
 		stream: true,
 		prompt_cache_key: promptCacheKey,
-		prompt_cache_retention: promptCacheKey ? getPromptCacheRetention(model.baseUrl, cacheRetention) : undefined,
+		prompt_cache_retention: promptCacheKey
+			? getPromptCacheRetention(resolvedBaseUrl || model.baseUrl, cacheRetention)
+			: undefined,
 		store: false,
 		stream_options: model.provider === "openai" ? { include_obfuscation: false } : undefined,
 	};
@@ -464,6 +507,7 @@ function buildParams(
 	applyResponsesReasoningParams(params, model, options, messages, effort =>
 		mapReasoningEffort(effort as NonNullable<OpenAIResponsesOptions["reasoning"]>, model.compat?.reasoningEffortMap),
 	);
+	applyOpenAIRequestTransformBody(params, model.requestTransform);
 
 	return { conversationMessages, params };
 }
@@ -482,24 +526,28 @@ function isAzureOpenAIBaseUrl(baseUrl: string): boolean {
 function supportsStrictMode(model: Model<"openai-responses">): boolean {
 	if (model.provider === "openai" || model.provider === "azure" || model.provider === "github-copilot") return true;
 
-	const baseUrl = model.baseUrl.toLowerCase();
+	const baseUrl = model.baseUrl;
+	const lowerBaseUrl = baseUrl.toLowerCase();
 	return (
-		baseUrl.includes("api.openai.com") ||
-		baseUrl.includes(".openai.azure.com") ||
-		baseUrl.includes("models.inference.ai.azure.com")
+		isDefaultOpenAIBaseUrl(baseUrl) ||
+		lowerBaseUrl.includes(".openai.azure.com") ||
+		lowerBaseUrl.includes("models.inference.ai.azure.com")
 	);
 }
 
 export function supportsDeveloperRole(modelOrBaseUrl: Pick<Model, "provider" | "baseUrl"> | string): boolean {
-	const baseUrl =
-		typeof modelOrBaseUrl === "string" ? modelOrBaseUrl.toLowerCase() : (modelOrBaseUrl.baseUrl ?? "").toLowerCase();
+	const baseUrl = typeof modelOrBaseUrl === "string" ? modelOrBaseUrl : (modelOrBaseUrl.baseUrl ?? "");
+	if (typeof modelOrBaseUrl !== "string" && modelOrBaseUrl.provider === "openai" && !baseUrl) {
+		return true;
+	}
+	const lowerBaseUrl = baseUrl.toLowerCase();
 	return (
-		baseUrl.includes("api.openai.com") ||
-		baseUrl.includes(".openai.azure.com") ||
-		baseUrl.includes("azure.com/openai") ||
-		baseUrl.includes("models.inference.ai.azure.com") ||
-		baseUrl.includes("githubcopilot.com") ||
-		baseUrl.includes("copilot-api.")
+		isOpenAIHostBaseUrl(baseUrl) ||
+		lowerBaseUrl.includes(".openai.azure.com") ||
+		lowerBaseUrl.includes("azure.com/openai") ||
+		lowerBaseUrl.includes("models.inference.ai.azure.com") ||
+		lowerBaseUrl.includes("githubcopilot.com") ||
+		lowerBaseUrl.includes("copilot-api.")
 	);
 }
 

package/src/stream.ts

@@ -1,7 +1,7 @@
 import * as fs from "node:fs";
 import * as os from "node:os";
 import * as path from "node:path";
-import { $env, $pickenv, extractHttpStatusFromError } from "@gajae-code/utils";
+import { $env, $inheritedEnv, $pickenv, extractHttpStatusFromError } from "@gajae-code/utils";
 import { getCustomApi } from "./api-registry";
 import type { Effort } from "./model-thinking";
 import {
@@ -77,7 +77,7 @@ type KeyResolver = string | (() => string | undefined);
 
 const serviceProviderMap: Record<string, KeyResolver> = {
 	"alibaba-coding-plan": "ALIBABA_CODING_PLAN_API_KEY",
-	openai: "OPENAI_API_KEY",
+	openai: () => $inheritedEnv("OPENAI_API_KEY") ?? $env.OPENAI_API_KEY,
 	google: "GEMINI_API_KEY",
 	groq: "GROQ_API_KEY",
 	cerebras: "CEREBRAS_API_KEY",

package/src/types.ts

@@ -823,6 +823,18 @@ export interface VercelGatewayRouting {
 }
 
 // Model interface for the unified model system
+
+export interface ModelRequestTransform {
+	/** Named request-shaping preset. `openai-proxy` removes OpenAI SDK telemetry headers and uses a generic Gajae-Code User-Agent. */
+	profile?: "openai-proxy";
+	/** Header names to remove from the final outbound request. Case-insensitive. */
+	stripHeaders?: string[];
+	/** Headers to set after stripping; use null to remove a header explicitly. */
+	setHeaders?: Record<string, string | null>;
+	/** Extra request body fields merged after provider defaults; protected core request keys are ignored. */
+	extraBody?: Record<string, unknown>;
+}
+
 export interface Model<TApi extends Api = any> {
 	id: string;
 	name: string;
@@ -861,6 +873,10 @@ export interface Model<TApi extends Api = any> {
 	preferWebsockets?: boolean;
 	/** Preferred model to switch to when context promotion is triggered (model id or provider/id). */
 	contextPromotionTarget?: string;
+	/** Provider-facing model id when it differs from the local selector id. */
+	wireModelId?: string;
+	/** Declarative request shaping for OpenAI-compatible proxy providers. */
+	requestTransform?: ModelRequestTransform;
 	/** Provider-assigned priority value (lower = higher priority). */
 	priority?: number;
 	/** Canonical thinking capability metadata for this model. */