@tuan_son.dinh/gsd 2.6.2 → 2.7.0

package/packages/pi-ai/src/providers/openai-responses.ts

@@ -0,0 +1,262 @@
+import OpenAI from "openai";
+import type { ResponseCreateParamsStreaming } from "openai/resources/responses/responses.js";
+import { getEnvApiKey } from "../env-api-keys.js";
+import { supportsXhigh } from "../models.js";
+import type {
+	Api,
+	AssistantMessage,
+	CacheRetention,
+	Context,
+	Model,
+	SimpleStreamOptions,
+	StreamFunction,
+	StreamOptions,
+	Usage,
+} from "../types.js";
+import { AssistantMessageEventStream } from "../utils/event-stream.js";
+import { buildCopilotDynamicHeaders, hasCopilotVisionInput } from "./github-copilot-headers.js";
+import { convertResponsesMessages, convertResponsesTools, processResponsesStream } from "./openai-responses-shared.js";
+import { buildBaseOptions, clampReasoning } from "./simple-options.js";
+
+const OPENAI_TOOL_CALL_PROVIDERS = new Set(["openai", "openai-codex", "opencode"]);
+
+/**
+ * Resolve cache retention preference.
+ * Defaults to "short" and uses PI_CACHE_RETENTION for backward compatibility.
+ */
+function resolveCacheRetention(cacheRetention?: CacheRetention): CacheRetention {
+	if (cacheRetention) {
+		return cacheRetention;
+	}
+	if (typeof process !== "undefined" && process.env.PI_CACHE_RETENTION === "long") {
+		return "long";
+	}
+	return "short";
+}
+
+/**
+ * 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 {
+	if (cacheRetention !== "long") {
+		return undefined;
+	}
+	if (baseUrl.includes("api.openai.com")) {
+		return "24h";
+	}
+	return undefined;
+}
+
+// OpenAI Responses-specific options
+export interface OpenAIResponsesOptions extends StreamOptions {
+	reasoningEffort?: "minimal" | "low" | "medium" | "high" | "xhigh";
+	reasoningSummary?: "auto" | "detailed" | "concise" | null;
+	serviceTier?: ResponseCreateParamsStreaming["service_tier"];
+}
+
+/**
+ * Generate function for OpenAI Responses API
+ */
+export const streamOpenAIResponses: StreamFunction<"openai-responses", OpenAIResponsesOptions> = (
+	model: Model<"openai-responses">,
+	context: Context,
+	options?: OpenAIResponsesOptions,
+): AssistantMessageEventStream => {
+	const stream = new AssistantMessageEventStream();
+
+	// Start async processing
+	(async () => {
+		const output: AssistantMessage = {
+			role: "assistant",
+			content: [],
+			api: model.api as Api,
+			provider: model.provider,
+			model: model.id,
+			usage: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+				totalTokens: 0,
+				cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
+			},
+			stopReason: "stop",
+			timestamp: Date.now(),
+		};
+
+		try {
+			// Create OpenAI client
+			const apiKey = options?.apiKey || getEnvApiKey(model.provider) || "";
+			const client = createClient(model, context, apiKey, options?.headers);
+			let params = buildParams(model, context, options);
+			const nextParams = await options?.onPayload?.(params, model);
+			if (nextParams !== undefined) {
+				params = nextParams as ResponseCreateParamsStreaming;
+			}
+			const openaiStream = await client.responses.create(
+				params,
+				options?.signal ? { signal: options.signal } : undefined,
+			);
+			stream.push({ type: "start", partial: output });
+
+			await processResponsesStream(openaiStream, output, stream, model, {
+				serviceTier: options?.serviceTier,
+				applyServiceTierPricing,
+			});
+
+			if (options?.signal?.aborted) {
+				throw new Error("Request was aborted");
+			}
+
+			if (output.stopReason === "aborted" || output.stopReason === "error") {
+				throw new Error("An unknown error occurred");
+			}
+
+			stream.push({ type: "done", reason: output.stopReason, message: output });
+			stream.end();
+		} catch (error) {
+			for (const block of output.content) delete (block as { index?: number }).index;
+			output.stopReason = options?.signal?.aborted ? "aborted" : "error";
+			output.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);
+			stream.push({ type: "error", reason: output.stopReason, error: output });
+			stream.end();
+		}
+	})();
+
+	return stream;
+};
+
+export const streamSimpleOpenAIResponses: StreamFunction<"openai-responses", SimpleStreamOptions> = (
+	model: Model<"openai-responses">,
+	context: Context,
+	options?: SimpleStreamOptions,
+): AssistantMessageEventStream => {
+	const apiKey = options?.apiKey || getEnvApiKey(model.provider);
+	if (!apiKey) {
+		throw new Error(`No API key for provider: ${model.provider}`);
+	}
+
+	const base = buildBaseOptions(model, options, apiKey);
+	const reasoningEffort = supportsXhigh(model) ? options?.reasoning : clampReasoning(options?.reasoning);
+
+	return streamOpenAIResponses(model, context, {
+		...base,
+		reasoningEffort,
+	} satisfies OpenAIResponsesOptions);
+};
+
+function createClient(
+	model: Model<"openai-responses">,
+	context: Context,
+	apiKey?: string,
+	optionsHeaders?: Record<string, string>,
+) {
+	if (!apiKey) {
+		if (!process.env.OPENAI_API_KEY) {
+			throw new Error(
+				"OpenAI API key is required. Set OPENAI_API_KEY environment variable or pass it as an argument.",
+			);
+		}
+		apiKey = process.env.OPENAI_API_KEY;
+	}
+
+	const headers = { ...model.headers };
+	if (model.provider === "github-copilot") {
+		const hasImages = hasCopilotVisionInput(context.messages);
+		const copilotHeaders = buildCopilotDynamicHeaders({
+			messages: context.messages,
+			hasImages,
+		});
+		Object.assign(headers, copilotHeaders);
+	}
+
+	// Merge options headers last so they can override defaults
+	if (optionsHeaders) {
+		Object.assign(headers, optionsHeaders);
+	}
+
+	return new OpenAI({
+		apiKey,
+		baseURL: model.baseUrl,
+		dangerouslyAllowBrowser: true,
+		defaultHeaders: headers,
+	});
+}
+
+function buildParams(model: Model<"openai-responses">, context: Context, options?: OpenAIResponsesOptions) {
+	const messages = convertResponsesMessages(model, context, OPENAI_TOOL_CALL_PROVIDERS);
+
+	const cacheRetention = resolveCacheRetention(options?.cacheRetention);
+	const params: ResponseCreateParamsStreaming = {
+		model: model.id,
+		input: messages,
+		stream: true,
+		prompt_cache_key: cacheRetention === "none" ? undefined : options?.sessionId,
+		prompt_cache_retention: getPromptCacheRetention(model.baseUrl, cacheRetention),
+		store: false,
+	};
+
+	if (options?.maxTokens) {
+		params.max_output_tokens = options?.maxTokens;
+	}
+
+	if (options?.temperature !== undefined) {
+		params.temperature = options?.temperature;
+	}
+
+	if (options?.serviceTier !== undefined) {
+		params.service_tier = options.serviceTier;
+	}
+
+	if (context.tools) {
+		params.tools = convertResponsesTools(context.tools);
+	}
+
+	if (model.reasoning) {
+		if (options?.reasoningEffort || options?.reasoningSummary) {
+			params.reasoning = {
+				effort: options?.reasoningEffort || "medium",
+				summary: options?.reasoningSummary || "auto",
+			};
+			params.include = ["reasoning.encrypted_content"];
+		} else {
+			if (model.name.startsWith("gpt-5")) {
+				// Jesus Christ, see https://community.openai.com/t/need-reasoning-false-option-for-gpt-5/1351588/7
+				messages.push({
+					role: "developer",
+					content: [
+						{
+							type: "input_text",
+							text: "# Juice: 0 !important",
+						},
+					],
+				});
+			}
+		}
+	}
+
+	return params;
+}
+
+function getServiceTierCostMultiplier(serviceTier: ResponseCreateParamsStreaming["service_tier"] | undefined): number {
+	switch (serviceTier) {
+		case "flex":
+			return 0.5;
+		case "priority":
+			return 2;
+		default:
+			return 1;
+	}
+}
+
+function applyServiceTierPricing(usage: Usage, serviceTier: ResponseCreateParamsStreaming["service_tier"] | undefined) {
+	const multiplier = getServiceTierCostMultiplier(serviceTier);
+	if (multiplier === 1) return;
+
+	usage.cost.input *= multiplier;
+	usage.cost.output *= multiplier;
+	usage.cost.cacheRead *= multiplier;
+	usage.cost.cacheWrite *= multiplier;
+	usage.cost.total = usage.cost.input + usage.cost.output + usage.cost.cacheRead + usage.cost.cacheWrite;
+}

package/packages/pi-ai/src/providers/register-builtins.ts

@@ -0,0 +1,186 @@
+import { clearApiProviders, registerApiProvider } from "../api-registry.js";
+import type { AssistantMessage, AssistantMessageEvent, Context, Model, SimpleStreamOptions } from "../types.js";
+import { AssistantMessageEventStream } from "../utils/event-stream.js";
+import type { BedrockOptions } from "./amazon-bedrock.js";
+import { streamAnthropic, streamSimpleAnthropic } from "./anthropic.js";
+import { streamAzureOpenAIResponses, streamSimpleAzureOpenAIResponses } from "./azure-openai-responses.js";
+import { streamGoogle, streamSimpleGoogle } from "./google.js";
+import { streamGoogleGeminiCli, streamSimpleGoogleGeminiCli } from "./google-gemini-cli.js";
+import { streamGoogleVertex, streamSimpleGoogleVertex } from "./google-vertex.js";
+import { streamMistral, streamSimpleMistral } from "./mistral.js";
+import { streamOpenAICodexResponses, streamSimpleOpenAICodexResponses } from "./openai-codex-responses.js";
+import { streamOpenAICompletions, streamSimpleOpenAICompletions } from "./openai-completions.js";
+import { streamOpenAIResponses, streamSimpleOpenAIResponses } from "./openai-responses.js";
+
+interface BedrockProviderModule {
+	streamBedrock: (
+		model: Model<"bedrock-converse-stream">,
+		context: Context,
+		options?: BedrockOptions,
+	) => AsyncIterable<AssistantMessageEvent>;
+	streamSimpleBedrock: (
+		model: Model<"bedrock-converse-stream">,
+		context: Context,
+		options?: SimpleStreamOptions,
+	) => AsyncIterable<AssistantMessageEvent>;
+}
+
+type DynamicImport = (specifier: string) => Promise<unknown>;
+
+const dynamicImport: DynamicImport = (specifier) => import(specifier);
+const BEDROCK_PROVIDER_SPECIFIER = "./amazon-" + "bedrock.js";
+
+let bedrockProviderModuleOverride: BedrockProviderModule | undefined;
+
+export function setBedrockProviderModule(module: BedrockProviderModule): void {
+	bedrockProviderModuleOverride = module;
+}
+
+async function loadBedrockProviderModule(): Promise<BedrockProviderModule> {
+	if (bedrockProviderModuleOverride) {
+		return bedrockProviderModuleOverride;
+	}
+	const module = await dynamicImport(BEDROCK_PROVIDER_SPECIFIER);
+	return module as BedrockProviderModule;
+}
+
+function forwardStream(target: AssistantMessageEventStream, source: AsyncIterable<AssistantMessageEvent>): void {
+	(async () => {
+		for await (const event of source) {
+			target.push(event);
+		}
+		target.end();
+	})();
+}
+
+function createLazyLoadErrorMessage(model: Model<"bedrock-converse-stream">, error: unknown): AssistantMessage {
+	return {
+		role: "assistant",
+		content: [],
+		api: "bedrock-converse-stream",
+		provider: model.provider,
+		model: model.id,
+		usage: {
+			input: 0,
+			output: 0,
+			cacheRead: 0,
+			cacheWrite: 0,
+			totalTokens: 0,
+			cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
+		},
+		stopReason: "error",
+		errorMessage: error instanceof Error ? error.message : String(error),
+		timestamp: Date.now(),
+	};
+}
+
+function streamBedrockLazy(
+	model: Model<"bedrock-converse-stream">,
+	context: Context,
+	options?: BedrockOptions,
+): AssistantMessageEventStream {
+	const outer = new AssistantMessageEventStream();
+
+	loadBedrockProviderModule()
+		.then((module) => {
+			const inner = module.streamBedrock(model, context, options);
+			forwardStream(outer, inner);
+		})
+		.catch((error) => {
+			const message = createLazyLoadErrorMessage(model, error);
+			outer.push({ type: "error", reason: "error", error: message });
+			outer.end(message);
+		});
+
+	return outer;
+}
+
+function streamSimpleBedrockLazy(
+	model: Model<"bedrock-converse-stream">,
+	context: Context,
+	options?: SimpleStreamOptions,
+): AssistantMessageEventStream {
+	const outer = new AssistantMessageEventStream();
+
+	loadBedrockProviderModule()
+		.then((module) => {
+			const inner = module.streamSimpleBedrock(model, context, options);
+			forwardStream(outer, inner);
+		})
+		.catch((error) => {
+			const message = createLazyLoadErrorMessage(model, error);
+			outer.push({ type: "error", reason: "error", error: message });
+			outer.end(message);
+		});
+
+	return outer;
+}
+
+export function registerBuiltInApiProviders(): void {
+	registerApiProvider({
+		api: "anthropic-messages",
+		stream: streamAnthropic,
+		streamSimple: streamSimpleAnthropic,
+	});
+
+	registerApiProvider({
+		api: "openai-completions",
+		stream: streamOpenAICompletions,
+		streamSimple: streamSimpleOpenAICompletions,
+	});
+
+	registerApiProvider({
+		api: "mistral-conversations",
+		stream: streamMistral,
+		streamSimple: streamSimpleMistral,
+	});
+
+	registerApiProvider({
+		api: "openai-responses",
+		stream: streamOpenAIResponses,
+		streamSimple: streamSimpleOpenAIResponses,
+	});
+
+	registerApiProvider({
+		api: "azure-openai-responses",
+		stream: streamAzureOpenAIResponses,
+		streamSimple: streamSimpleAzureOpenAIResponses,
+	});
+
+	registerApiProvider({
+		api: "openai-codex-responses",
+		stream: streamOpenAICodexResponses,
+		streamSimple: streamSimpleOpenAICodexResponses,
+	});
+
+	registerApiProvider({
+		api: "google-generative-ai",
+		stream: streamGoogle,
+		streamSimple: streamSimpleGoogle,
+	});
+
+	registerApiProvider({
+		api: "google-gemini-cli",
+		stream: streamGoogleGeminiCli,
+		streamSimple: streamSimpleGoogleGeminiCli,
+	});
+
+	registerApiProvider({
+		api: "google-vertex",
+		stream: streamGoogleVertex,
+		streamSimple: streamSimpleGoogleVertex,
+	});
+
+	registerApiProvider({
+		api: "bedrock-converse-stream",
+		stream: streamBedrockLazy,
+		streamSimple: streamSimpleBedrockLazy,
+	});
+}
+
+export function resetApiProviders(): void {
+	clearApiProviders();
+	registerBuiltInApiProviders();
+}
+
+registerBuiltInApiProviders();

package/packages/pi-ai/src/providers/simple-options.ts

@@ -0,0 +1,46 @@
+import type { Api, Model, SimpleStreamOptions, StreamOptions, ThinkingBudgets, ThinkingLevel } from "../types.js";
+
+export function buildBaseOptions(model: Model<Api>, options?: SimpleStreamOptions, apiKey?: string): StreamOptions {
+	return {
+		temperature: options?.temperature,
+		maxTokens: options?.maxTokens || Math.min(model.maxTokens, 32000),
+		signal: options?.signal,
+		apiKey: apiKey || options?.apiKey,
+		cacheRetention: options?.cacheRetention,
+		sessionId: options?.sessionId,
+		headers: options?.headers,
+		onPayload: options?.onPayload,
+		maxRetryDelayMs: options?.maxRetryDelayMs,
+		metadata: options?.metadata,
+	};
+}
+
+export function clampReasoning(effort: ThinkingLevel | undefined): Exclude<ThinkingLevel, "xhigh"> | undefined {
+	return effort === "xhigh" ? "high" : effort;
+}
+
+export function adjustMaxTokensForThinking(
+	baseMaxTokens: number,
+	modelMaxTokens: number,
+	reasoningLevel: ThinkingLevel,
+	customBudgets?: ThinkingBudgets,
+): { maxTokens: number; thinkingBudget: number } {
+	const defaultBudgets: ThinkingBudgets = {
+		minimal: 1024,
+		low: 2048,
+		medium: 8192,
+		high: 16384,
+	};
+	const budgets = { ...defaultBudgets, ...customBudgets };
+
+	const minOutputTokens = 1024;
+	const level = clampReasoning(reasoningLevel)!;
+	let thinkingBudget = budgets[level]!;
+	const maxTokens = Math.min(baseMaxTokens + thinkingBudget, modelMaxTokens);
+
+	if (maxTokens <= thinkingBudget) {
+		thinkingBudget = Math.max(0, maxTokens - minOutputTokens);
+	}
+
+	return { maxTokens, thinkingBudget };
+}

package/packages/pi-ai/src/providers/transform-messages.ts

@@ -0,0 +1,172 @@
+import type { Api, AssistantMessage, Message, Model, ToolCall, ToolResultMessage } from "../types.js";
+
+/**
+ * Normalize tool call ID for cross-provider compatibility.
+ * OpenAI Responses API generates IDs that are 450+ chars with special characters like `|`.
+ * Anthropic APIs require IDs matching ^[a-zA-Z0-9_-]+$ (max 64 chars).
+ */
+export function transformMessages<TApi extends Api>(
+	messages: Message[],
+	model: Model<TApi>,
+	normalizeToolCallId?: (id: string, model: Model<TApi>, source: AssistantMessage) => string,
+): Message[] {
+	// Build a map of original tool call IDs to normalized IDs
+	const toolCallIdMap = new Map<string, string>();
+
+	// First pass: transform messages (thinking blocks, tool call ID normalization)
+	const transformed = messages.map((msg) => {
+		// User messages pass through unchanged
+		if (msg.role === "user") {
+			return msg;
+		}
+
+		// Handle toolResult messages - normalize toolCallId if we have a mapping
+		if (msg.role === "toolResult") {
+			const normalizedId = toolCallIdMap.get(msg.toolCallId);
+			if (normalizedId && normalizedId !== msg.toolCallId) {
+				return { ...msg, toolCallId: normalizedId };
+			}
+			return msg;
+		}
+
+		// Assistant messages need transformation check
+		if (msg.role === "assistant") {
+			const assistantMsg = msg as AssistantMessage;
+			const isSameModel =
+				assistantMsg.provider === model.provider &&
+				assistantMsg.api === model.api &&
+				assistantMsg.model === model.id;
+
+			const transformedContent = assistantMsg.content.flatMap((block) => {
+				if (block.type === "thinking") {
+					// Redacted thinking is opaque encrypted content, only valid for the same model.
+					// Drop it for cross-model to avoid API errors.
+					if (block.redacted) {
+						return isSameModel ? block : [];
+					}
+					// For same model: keep thinking blocks with signatures (needed for replay)
+					// even if the thinking text is empty (OpenAI encrypted reasoning)
+					if (isSameModel && block.thinkingSignature) return block;
+					// Skip empty thinking blocks, convert others to plain text
+					if (!block.thinking || block.thinking.trim() === "") return [];
+					if (isSameModel) return block;
+					return {
+						type: "text" as const,
+						text: block.thinking,
+					};
+				}
+
+				if (block.type === "text") {
+					if (isSameModel) return block;
+					return {
+						type: "text" as const,
+						text: block.text,
+					};
+				}
+
+				if (block.type === "toolCall") {
+					const toolCall = block as ToolCall;
+					let normalizedToolCall: ToolCall = toolCall;
+
+					if (!isSameModel && toolCall.thoughtSignature) {
+						normalizedToolCall = { ...toolCall };
+						delete (normalizedToolCall as { thoughtSignature?: string }).thoughtSignature;
+					}
+
+					if (!isSameModel && normalizeToolCallId) {
+						const normalizedId = normalizeToolCallId(toolCall.id, model, assistantMsg);
+						if (normalizedId !== toolCall.id) {
+							toolCallIdMap.set(toolCall.id, normalizedId);
+							normalizedToolCall = { ...normalizedToolCall, id: normalizedId };
+						}
+					}
+
+					return normalizedToolCall;
+				}
+
+				return block;
+			});
+
+			return {
+				...assistantMsg,
+				content: transformedContent,
+			};
+		}
+		return msg;
+	});
+
+	// Second pass: insert synthetic empty tool results for orphaned tool calls
+	// This preserves thinking signatures and satisfies API requirements
+	const result: Message[] = [];
+	let pendingToolCalls: ToolCall[] = [];
+	let existingToolResultIds = new Set<string>();
+
+	for (let i = 0; i < transformed.length; i++) {
+		const msg = transformed[i];
+
+		if (msg.role === "assistant") {
+			// If we have pending orphaned tool calls from a previous assistant, insert synthetic results now
+			if (pendingToolCalls.length > 0) {
+				for (const tc of pendingToolCalls) {
+					if (!existingToolResultIds.has(tc.id)) {
+						result.push({
+							role: "toolResult",
+							toolCallId: tc.id,
+							toolName: tc.name,
+							content: [{ type: "text", text: "No result provided" }],
+							isError: true,
+							timestamp: Date.now(),
+						} as ToolResultMessage);
+					}
+				}
+				pendingToolCalls = [];
+				existingToolResultIds = new Set();
+			}
+
+			// Skip errored/aborted assistant messages entirely.
+			// These are incomplete turns that shouldn't be replayed:
+			// - May have partial content (reasoning without message, incomplete tool calls)
+			// - Replaying them can cause API errors (e.g., OpenAI "reasoning without following item")
+			// - The model should retry from the last valid state
+			const assistantMsg = msg as AssistantMessage;
+			if (assistantMsg.stopReason === "error" || assistantMsg.stopReason === "aborted") {
+				continue;
+			}
+
+			// Track tool calls from this assistant message
+			const toolCalls = assistantMsg.content.filter((b) => b.type === "toolCall") as ToolCall[];
+			if (toolCalls.length > 0) {
+				pendingToolCalls = toolCalls;
+				existingToolResultIds = new Set();
+			}
+
+			result.push(msg);
+		} else if (msg.role === "toolResult") {
+			existingToolResultIds.add(msg.toolCallId);
+			result.push(msg);
+		} else if (msg.role === "user") {
+			// User message interrupts tool flow - insert synthetic results for orphaned calls
+			if (pendingToolCalls.length > 0) {
+				for (const tc of pendingToolCalls) {
+					if (!existingToolResultIds.has(tc.id)) {
+						result.push({
+							role: "toolResult",
+							toolCallId: tc.id,
+							toolName: tc.name,
+							content: [{ type: "text", text: "No result provided" }],
+							isError: true,
+							timestamp: Date.now(),
+						} as ToolResultMessage);
+					}
+				}
+				pendingToolCalls = [];
+				existingToolResultIds = new Set();
+			}
+			result.push(msg);
+		} else {
+			result.push(msg);
+		}
+	}
+
+	return result;
+}