@gajae-code/ai 0.16.4 → 0.16.6

package/CHANGELOG.md

@@ -2,6 +2,13 @@
 
 ## [Unreleased]
 
+## [0.16.6] - 2026-09-07
+
+## [0.16.5] - 2026-09-07
+
+- Documented `GJC_OPENAI_CODE_WEBSOCKET_V2` as a switch that enables a websocket v2 path. No code read it under that name, under the legacy `PI_CODEX_WEBSOCKET_V2`, or under the `PI_OPENAI_CODE_WEBSOCKET_V2` the historical entry records; the v2 beta header has been unconditional for websocket transport. The documentation row is removed rather than reintroducing a knob, and the test that claimed to gate on it no longer writes an environment variable nothing reads.
+- Maintenance reasoning now fails closed for Anthropic models routed through an unverified custom endpoint and for raw reasoning-enabled models without thinking metadata. This prevents unsupported thinking controls and avoids a synchronous missing-metadata crash before provider wire transformation.
+
 ## [0.16.4] - 2026-09-05
 
 ### Added

package/dist/types/types.d.ts

@@ -256,12 +256,12 @@ export interface StreamOptions {
      * Return undefined to keep the payload unchanged.
      * The `scope` parameter carries the per-attempt identity for execution attribution.
      */
-    onPayload?: (payload: unknown, model?: Model<Api>, scope?: AttemptScopeRef) => unknown | undefined | Promise<unknown | undefined>;
+    onPayload?: (payload: unknown, model?: Model<Api>, scope?: AttemptScopeRef, signal?: AbortSignal) => unknown | undefined | Promise<unknown | undefined>;
     /**
      * Optional callback for provider response metadata after headers are received.
      * The `scope` parameter carries the per-attempt identity for execution attribution.
      */
-    onResponse?: (response: ProviderResponseMetadata, model?: Model<Api>, scope?: AttemptScopeRef) => void | Promise<void>;
+    onResponse?: (response: ProviderResponseMetadata, model?: Model<Api>, scope?: AttemptScopeRef, signal?: AbortSignal) => void | Promise<void>;
     /**
      * Internal dispatch-admission hook. Providers invoke this immediately before
      * submitting an outbound request; stream forwarding retains a first-response

package/dist/types/utils/provider-response.d.ts

@@ -3,4 +3,5 @@ export declare function normalizeProviderResponse(response: Response, requestId?
 export declare function notifyProviderResponse(options: {
     onResponse?: StreamOptions["onResponse"];
     attemptScope?: AttemptScopeRef;
+    signal?: AbortSignal;
 } | undefined, response: Response, model?: Model<Api>, requestId?: string | null, metadata?: Record<string, unknown>): Promise<void>;

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@gajae-code/ai",
-	"version": "0.16.4",
+	"version": "0.16.6",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"homepage": "https://gajae-code.com",
 	"author": "Yeachan-Heo and Gajae Code Contributors",
@@ -40,8 +40,8 @@
 	"dependencies": {
 		"@anthropic-ai/sdk": "^0.94.0",
 		"@bufbuild/protobuf": "^2.12.0",
-		"@gajae-code/natives": "0.16.4",
-		"@gajae-code/utils": "0.16.4",
+		"@gajae-code/natives": "0.16.6",
+		"@gajae-code/utils": "0.16.6",
 		"openai": "^6.36.0",
 		"partial-json": "^0.1.7",
 		"zod": "4.4.3"

package/src/model-thinking.ts

@@ -242,7 +242,6 @@ export function modelSupportsReasoningControl<TApi extends Api>(
 	resolvedBaseUrl?: string,
 ): boolean {
 	if (!model.reasoning) return false;
-
 	if (model.api === "openai-completions") {
 		const completionsModel = model as ApiModel<"openai-completions">;
 		const explicitSupport = completionsModel.compat?.supportsReasoningEffort;
@@ -366,6 +365,18 @@ export function clampThinkingLevelForModel<TApi extends Api>(
 	if (!modelSupportsReasoningControl(model) || requested === undefined) {
 		return undefined;
 	}
+	if (model.api === "anthropic-messages" && model.provider === "anthropic") {
+		const baseUrl = model.baseUrl;
+		try {
+			const url = new URL(baseUrl || "https://api.anthropic.com");
+			if (url.protocol !== "https:" || url.hostname !== "api.anthropic.com") return undefined;
+		} catch {
+			return undefined;
+		}
+	}
+	if (!model.thinking) {
+		return undefined;
+	}
 
 	const levels = getSupportedEfforts(model);
 	if (levels.includes(requested)) {

package/src/providers/amazon-bedrock.ts

@@ -220,7 +220,7 @@ export const streamBedrock: StreamFunction<"bedrock-converse-stream"> = (
 				if (tc.any || tc.tool) additionalModelRequestFields = undefined;
 			}
 
-			const commandInput: ConverseStreamRequest = {
+			let commandInput: ConverseStreamRequest = {
 				messages: convertMessages(context, model, cacheRetention),
 				system: buildSystemPrompt(context.systemPrompt, model, cacheRetention),
 				inferenceConfig: {
@@ -231,7 +231,15 @@ export const streamBedrock: StreamFunction<"bedrock-converse-stream"> = (
 				toolConfig,
 				additionalModelRequestFields,
 			};
-			options?.onPayload?.(commandInput, model, options?.attemptScope);
+			const replacementPayload = await options?.onPayload?.(
+				commandInput,
+				model,
+				options?.attemptScope,
+				options?.signal,
+			);
+			if (replacementPayload !== undefined) {
+				commandInput = replacementPayload as typeof commandInput;
+			}
 
 			const host = `bedrock-runtime.${region}.amazonaws.com`;
 			const url = `https://${host}/model/${encodeURIComponent(model.id)}/converse-stream`;

package/src/providers/anthropic.ts

@@ -1999,7 +1999,12 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 				if (dropFastMode) {
 					dropAnthropicFastMode(nextParams);
 				}
-				const replacementPayload = await options?.onPayload?.(nextParams, model, options?.attemptScope);
+				const replacementPayload = await options?.onPayload?.(
+					nextParams,
+					model,
+					options?.attemptScope,
+					options?.signal,
+				);
 				if (replacementPayload !== undefined) {
 					nextParams = replacementPayload as typeof nextParams;
 				}

package/src/providers/azure-openai-responses.ts

@@ -129,9 +129,12 @@ export const streamAzureOpenAIResponses: StreamFunction<"azure-openai-responses"
 			const apiKey = options?.apiKey || getEnvApiKey(model.provider) || "";
 			const client = createClient(model, apiKey, options);
 			const { baseUrl } = resolveAzureConfig(model, options);
-			const params = buildParams(model, context, options, deploymentName, baseUrl);
+			let params = buildParams(model, context, options, deploymentName, baseUrl);
 			const idleTimeoutMs = options?.streamIdleTimeoutMs ?? getOpenAIStreamIdleTimeoutMs();
-			options?.onPayload?.(params, model, options?.attemptScope);
+			const replacementPayload = await options?.onPayload?.(params, model, options?.attemptScope, options?.signal);
+			if (replacementPayload !== undefined) {
+				params = replacementPayload as typeof params;
+			}
 			rawRequestDump = {
 				provider: model.provider,
 				api: output.api,

package/src/providers/cursor.ts

@@ -3575,7 +3575,7 @@ async function buildGrpcRequest(
 
 	if (options?.onPayload) {
 		const payload = toJson(AgentRunRequestSchema, runRequest);
-		const replacement = await options.onPayload(payload, model, options.attemptScope);
+		const replacement = await options.onPayload(payload, model, options.attemptScope, options.signal);
 		if (replacement !== undefined) {
 			runRequest = fromJson(AgentRunRequestSchema, replacement as JsonValue);
 		}

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

@@ -354,7 +354,12 @@ export const streamGoogleGeminiCli: StreamFunction<"google-gemini-cli"> = (
 			const endpoints = baseUrl ? [baseUrl] : isAntigravity ? ANTIGRAVITY_ENDPOINT_FALLBACKS : [DEFAULT_ENDPOINT];
 
 			let requestBody = buildRequest(model, context, projectId, options, isAntigravity);
-			const replacementPayload = await options?.onPayload?.(requestBody, model, options?.attemptScope);
+			const replacementPayload = await options?.onPayload?.(
+				requestBody,
+				model,
+				options?.attemptScope,
+				options?.signal,
+			);
 			if (replacementPayload !== undefined) {
 				requestBody = replacementPayload as typeof requestBody;
 			}

package/src/providers/google-shared.ts

@@ -899,7 +899,7 @@ export function streamGoogleGenAI<T extends "google-generative-ai" | "google-ver
 		try {
 			const plan = await prepare();
 			let params = plan.params;
-			const replacement = await options?.onPayload?.(params, model, options?.attemptScope);
+			const replacement = await options?.onPayload?.(params, model, options?.attemptScope, options?.signal);
 			if (replacement !== undefined) {
 				params = replacement as GenerateContentParameters;
 			}

package/src/providers/kiro-api-key.ts

@@ -611,8 +611,11 @@ export const streamKiroApiKey: StreamFunction<"kiro-codewhisperer-stream"> = (
 			const configuredBaseUrl = model.baseUrl;
 			const usesExplicitBaseUrl = Boolean(configuredBaseUrl) && !isRegionDerivedKiroApiBaseUrl(configuredBaseUrl);
 			const endpoint = configuredBaseUrl || kiroApiBaseUrl(kiroApiRegion(options));
-			const request = buildApiKeyRequest(model, context, options);
-			options?.onPayload?.(request, model, options?.attemptScope);
+			let request = buildApiKeyRequest(model, context, options);
+			const replacementPayload = await options?.onPayload?.(request, model, options?.attemptScope, options?.signal);
+			if (replacementPayload !== undefined) {
+				request = replacementPayload;
+			}
 
 			const response = await fetch(endpoint, {
 				method: "POST",

package/src/providers/kiro-codewhisperer.ts

@@ -193,11 +193,19 @@ export const streamKiroCodeWhisperer: StreamFunction<"kiro-codewhisperer-stream"
 
 			// Build request
 			const conversationState = buildConversationState(context, model, options);
-			const requestBody: GenerateAssistantResponseRequest = {
+			let requestBody: GenerateAssistantResponseRequest = {
 				conversationState,
 			};
 
-			options?.onPayload?.(requestBody, model, options?.attemptScope);
+			const replacementPayload = await options?.onPayload?.(
+				requestBody,
+				model,
+				options?.attemptScope,
+				options?.signal,
+			);
+			if (replacementPayload !== undefined) {
+				requestBody = replacementPayload as typeof requestBody;
+			}
 
 			const host = `${STREAMING_SERVICE_NAME}.${region}.amazonaws.com`;
 			const url = `https://${host}/`;

package/src/providers/mock.ts

@@ -339,6 +339,7 @@ async function runMock(
 				},
 				model,
 				options.attemptScope,
+				options.signal,
 			);
 		} catch (err) {
 			stream.fail(err);

package/src/providers/ollama.ts

@@ -407,7 +407,7 @@ export const streamOllama: StreamFunction<"ollama-chat"> = (
 			const baseUrl = normalizeBaseUrl(model.baseUrl);
 			let body = createChatBody(model, context, options);
 			const sentForcedToolChoice = body.tool_choice === "required";
-			const replacementPayload = await options.onPayload?.(body, model, options?.attemptScope);
+			const replacementPayload = await options.onPayload?.(body, model, options?.attemptScope, options?.signal);
 			if (replacementPayload !== undefined) {
 				body = replacementPayload as typeof body;
 			}

package/src/providers/openai-codex-responses.ts

@@ -717,8 +717,16 @@ async function buildCodexRequestContext(
 	const baseUrl = model.baseUrl || CODEX_BASE_URL;
 	const url = resolveCodexResponsesUrl(baseUrl);
 	const promptCacheKey = normalizeOpenAIResponsesPromptCacheKey(options?.sessionId);
-	const transformedBody = await buildTransformedCodexRequestBody(model, context, options);
-	options?.onPayload?.(transformedBody, model, options?.attemptScope);
+	let transformedBody = await buildTransformedCodexRequestBody(model, context, options);
+	const replacementPayload = await options?.onPayload?.(
+		transformedBody,
+		model,
+		options?.attemptScope,
+		options?.signal,
+	);
+	if (replacementPayload !== undefined) {
+		transformedBody = replacementPayload as typeof transformedBody;
+	}
 
 	const requestHeaders = { ...(model.headers ?? {}), ...(options?.headers ?? {}) };
 	const rawRequestDump: RawHttpRequestDump = {

package/src/providers/openai-completions.ts

@@ -624,15 +624,24 @@ export const streamOpenAICompletions: StreamFunction<"openai-completions"> = (
 			const createCompletionsStream = async (toolStrictModeOverride?: ToolStrictModeOverride) => {
 				clearCapturedErrorResponse();
 				const effectiveToolStrictModeOverride = disableStrictTools ? "none" : toolStrictModeOverride;
-				const { params, toolStrictMode } = buildParams(
+				const { params: builtParams, toolStrictMode } = buildParams(
 					model,
 					context,
 					options,
 					baseUrl,
 					effectiveToolStrictModeOverride,
 				);
+				let params = builtParams;
 				appliedToolStrictMode = toolStrictMode;
-				options?.onPayload?.(params, undefined, options?.attemptScope);
+				const replacementPayload = await options?.onPayload?.(
+					params,
+					undefined,
+					options?.attemptScope,
+					options?.signal,
+				);
+				if (replacementPayload !== undefined) {
+					params = replacementPayload as typeof params;
+				}
 				rawRequestDump = {
 					provider: model.provider,
 					api: output.api,

package/src/providers/openai-responses.ts

@@ -395,9 +395,17 @@ export const streamOpenAIResponses: StreamFunction<"openai-responses"> = (
 			);
 			const premiumRequestsTotal = copilotPremiumRequests;
 			const providerSessionState = getOpenAIResponsesProviderSessionState(model, options?.providerSessionState);
-			const { params } = buildParams(model, context, options, providerSessionState, cacheRetention, baseUrl);
+			let { params } = buildParams(model, context, options, providerSessionState, cacheRetention, baseUrl);
 			const idleTimeoutMs = options?.streamIdleTimeoutMs ?? getOpenAIStreamIdleTimeoutMs(model.provider, model.id);
-			options?.onPayload?.(params, undefined, options?.attemptScope);
+			const replacementPayload = await options?.onPayload?.(
+				params,
+				undefined,
+				options?.attemptScope,
+				options?.signal,
+			);
+			if (replacementPayload !== undefined) {
+				params = replacementPayload as typeof params;
+			}
 			rawRequestDump = {
 				provider: model.provider,
 				api: output.api,

package/src/types.d.ts

@@ -256,12 +256,12 @@ export interface StreamOptions {
      * Return undefined to keep the payload unchanged.
      * The `scope` parameter carries the per-attempt identity for execution attribution.
      */
-    onPayload?: (payload: unknown, model?: Model<Api>, scope?: AttemptScopeRef) => unknown | undefined | Promise<unknown | undefined>;
+    onPayload?: (payload: unknown, model?: Model<Api>, scope?: AttemptScopeRef, signal?: AbortSignal) => unknown | undefined | Promise<unknown | undefined>;
     /**
      * Optional callback for provider response metadata after headers are received.
      * The `scope` parameter carries the per-attempt identity for execution attribution.
      */
-    onResponse?: (response: ProviderResponseMetadata, model?: Model<Api>, scope?: AttemptScopeRef) => void | Promise<void>;
+    onResponse?: (response: ProviderResponseMetadata, model?: Model<Api>, scope?: AttemptScopeRef, signal?: AbortSignal) => void | Promise<void>;
     /**
      * Internal dispatch-admission hook. Providers invoke this immediately before
      * submitting an outbound request; stream forwarding retains a first-response

package/src/types.ts

@@ -464,6 +464,7 @@ export interface StreamOptions {
 		payload: unknown,
 		model?: Model<Api>,
 		scope?: AttemptScopeRef,
+		signal?: AbortSignal,
 	) => unknown | undefined | Promise<unknown | undefined>;
 	/**
 	 * Optional callback for provider response metadata after headers are received.
@@ -473,6 +474,7 @@ export interface StreamOptions {
 		response: ProviderResponseMetadata,
 		model?: Model<Api>,
 		scope?: AttemptScopeRef,
+		signal?: AbortSignal,
 	) => void | Promise<void>;
 	/**
 	 * Internal dispatch-admission hook. Providers invoke this immediately before

package/src/utils/provider-response.d.ts

@@ -3,4 +3,5 @@ export declare function normalizeProviderResponse(response: Response, requestId?
 export declare function notifyProviderResponse(options: {
     onResponse?: StreamOptions["onResponse"];
     attemptScope?: AttemptScopeRef;
+    signal?: AbortSignal;
 } | undefined, response: Response, model?: Model<Api>, requestId?: string | null, metadata?: Record<string, unknown>): Promise<void>;

package/src/utils/provider-response.ts

@@ -19,12 +19,19 @@ export function normalizeProviderResponse(
 }
 
 export async function notifyProviderResponse(
-	options: { onResponse?: StreamOptions["onResponse"]; attemptScope?: AttemptScopeRef } | undefined,
+	options:
+		| { onResponse?: StreamOptions["onResponse"]; attemptScope?: AttemptScopeRef; signal?: AbortSignal }
+		| undefined,
 	response: Response,
 	model?: Model<Api>,
 	requestId?: string | null,
 	metadata?: Record<string, unknown>,
 ): Promise<void> {
 	if (!options?.onResponse) return;
-	await options.onResponse(normalizeProviderResponse(response, requestId, metadata), model, options.attemptScope);
+	await options.onResponse(
+		normalizeProviderResponse(response, requestId, metadata),
+		model,
+		options.attemptScope,
+		options.signal,
+	);
 }