@oh-my-pi/pi-ai 3.20.1 → 3.35.0

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

@@ -6,7 +6,7 @@ import type {
 	ResponseInputText,
 	ResponseOutputMessage,
 	ResponseReasoningItem,
-} from "openai/resources/responses/responses.js";
+} from "openai/resources/responses/responses";
 import { calculateCost } from "../models";
 import { getEnvApiKey } from "../stream";
 import type {
@@ -24,6 +24,7 @@ import type {
 } from "../types";
 import { AssistantMessageEventStream } from "../utils/event-stream";
 import { parseStreamingJson } from "../utils/json-parse";
+import { formatErrorMessageWithRetryAfter } from "../utils/retry-after";
 import { sanitizeSurrogates } from "../utils/sanitize-unicode";
 import {
 	CODEX_BASE_URL,
@@ -33,12 +34,9 @@ import {
 	URL_PATHS,
 } from "./openai-codex/constants";
 import { getCodexInstructions } from "./openai-codex/prompts/codex";
-import {
-	type CodexRequestOptions,
-	normalizeModel,
-	type RequestBody,
-	transformRequestBody,
-} from "./openai-codex/request-transformer";
+import { buildCodexPiBridge } from "./openai-codex/prompts/pi-codex-bridge";
+import { buildCodexSystemPrompt } from "./openai-codex/prompts/system-prompt";
+import { type CodexRequestOptions, type RequestBody, transformRequestBody } from "./openai-codex/request-transformer";
 import { parseCodexError, parseCodexSseStream } from "./openai-codex/response-handler";
 import { transformMessages } from "./transorm-messages";
 
@@ -94,6 +92,7 @@ export const streamOpenAICodexResponses: StreamFunction<"openai-codex-responses"
 				model: model.id,
 				input: messages,
 				stream: true,
+				prompt_cache_key: options?.sessionId,
 			};
 
 			if (options?.maxTokens) {
@@ -108,8 +107,15 @@ export const streamOpenAICodexResponses: StreamFunction<"openai-codex-responses"
 				params.tools = convertTools(context.tools);
 			}
 
-			const normalizedModel = normalizeModel(params.model);
-			const codexInstructions = await getCodexInstructions(normalizedModel);
+			const codexInstructions = await getCodexInstructions(params.model);
+			const bridgeText = buildCodexPiBridge(context.tools);
+			const systemPrompt = buildCodexSystemPrompt({
+				codexInstructions,
+				bridgeText,
+				userSystemPrompt: context.systemPrompt,
+			});
+
+			params.instructions = systemPrompt.instructions;
 
 			const codexOptions: CodexRequestOptions = {
 				reasoningEffort: options?.reasoningEffort,
@@ -118,17 +124,14 @@ export const streamOpenAICodexResponses: StreamFunction<"openai-codex-responses"
 				include: options?.include,
 			};
 
-			const transformedBody = await transformRequestBody(
-				params,
-				codexInstructions,
-				codexOptions,
-				options?.codexMode ?? true,
-			);
+			const transformedBody = await transformRequestBody(params, codexOptions, systemPrompt);
 
-			const headers = createCodexHeaders(model.headers, accountId, apiKey, transformedBody.prompt_cache_key);
+			const reasoningEffort = transformedBody.reasoning?.effort ?? null;
+			const headers = createCodexHeaders(model.headers, accountId, apiKey, options?.sessionId);
 			logCodexDebug("codex request", {
 				url,
 				model: params.model,
+				reasoningEffort,
 				headers: redactHeaders(headers),
 			});
 
@@ -149,7 +152,9 @@ export const streamOpenAICodexResponses: StreamFunction<"openai-codex-responses"
 
 			if (!response.ok) {
 				const info = await parseCodexError(response);
-				throw new Error(info.friendlyMessage || info.message);
+				const error = new Error(info.friendlyMessage || info.message);
+				(error as { headers?: Headers }).headers = response.headers;
+				throw error;
 			}
 
 			if (!response.body) {
@@ -340,10 +345,10 @@ export const streamOpenAICodexResponses: StreamFunction<"openai-codex-responses"
 					}
 				} else if (eventType === "error") {
 					const code = (rawEvent as { code?: string }).code || "";
-					const message = (rawEvent as { message?: string }).message || "Unknown error";
-					throw new Error(code ? `Error Code ${code}: ${message}` : message);
+					const message = (rawEvent as { message?: string }).message || "";
+					throw new Error(formatCodexErrorEvent(rawEvent, code, message));
 				} else if (eventType === "response.failed") {
-					throw new Error("Unknown error");
+					throw new Error(formatCodexFailure(rawEvent) ?? "Codex response failed");
 				}
 			}
 
@@ -352,7 +357,7 @@ export const streamOpenAICodexResponses: StreamFunction<"openai-codex-responses"
 			}
 
 			if (output.stopReason === "aborted" || output.stopReason === "error") {
-				throw new Error("An unknown error occurred");
+				throw new Error("Codex response failed");
 			}
 
 			stream.push({ type: "done", reason: output.stopReason, message: output });
@@ -360,7 +365,7 @@ export const streamOpenAICodexResponses: StreamFunction<"openai-codex-responses"
 		} 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);
+			output.errorMessage = formatErrorMessageWithRetryAfter(error);
 			stream.push({ type: "error", reason: output.stopReason, error: output });
 			stream.end();
 		}
@@ -406,11 +411,11 @@ function logCodexDebug(message: string, details?: Record<string, unknown>): void
 
 function redactHeaders(headers: Headers): Record<string, string> {
 	const redacted: Record<string, string> = {};
-	headers.forEach((value, key) => {
+	for (const [key, value] of headers.entries()) {
 		const lower = key.toLowerCase();
 		if (lower === "authorization") {
 			redacted[key] = "Bearer [redacted]";
-			return;
+			continue;
 		}
 		if (
 			lower.includes("account") ||
@@ -419,10 +424,10 @@ function redactHeaders(headers: Headers): Record<string, string> {
 			lower === "cookie"
 		) {
 			redacted[key] = "[redacted]";
-			return;
+			continue;
 		}
 		redacted[key] = value;
-	});
+	}
 	return redacted;
 }
 
@@ -617,3 +622,68 @@ function mapStopReason(status: string | undefined): StopReason {
 			return "stop";
 	}
 }
+
+function asRecord(value: unknown): Record<string, unknown> | null {
+	if (value && typeof value === "object") {
+		return value as Record<string, unknown>;
+	}
+	return null;
+}
+
+function getString(value: unknown): string | undefined {
+	return typeof value === "string" ? value : undefined;
+}
+
+function truncate(text: string, limit: number): string {
+	if (text.length <= limit) return text;
+	return `${text.slice(0, limit)}...[truncated ${text.length - limit}]`;
+}
+
+function formatCodexFailure(rawEvent: Record<string, unknown>): string | null {
+	const response = asRecord(rawEvent.response);
+	const error = asRecord(rawEvent.error) ?? (response ? asRecord(response.error) : null);
+
+	const message = getString(error?.message) ?? getString(rawEvent.message) ?? getString(response?.message);
+	const code = getString(error?.code) ?? getString(error?.type) ?? getString(rawEvent.code);
+	const status = getString(response?.status) ?? getString(rawEvent.status);
+
+	const meta: string[] = [];
+	if (code) meta.push(`code=${code}`);
+	if (status) meta.push(`status=${status}`);
+
+	if (message) {
+		const metaText = meta.length ? ` (${meta.join(", ")})` : "";
+		return `Codex response failed: ${message}${metaText}`;
+	}
+
+	if (meta.length) {
+		return `Codex response failed (${meta.join(", ")})`;
+	}
+
+	try {
+		return `Codex response failed: ${truncate(JSON.stringify(rawEvent), 800)}`;
+	} catch {
+		return "Codex response failed";
+	}
+}
+
+function formatCodexErrorEvent(rawEvent: Record<string, unknown>, code: string, message: string): string {
+	const detail = formatCodexFailure(rawEvent);
+	if (detail) {
+		return detail.replace("response failed", "error event");
+	}
+
+	const meta: string[] = [];
+	if (code) meta.push(`code=${code}`);
+	if (message) meta.push(`message=${message}`);
+
+	if (meta.length > 0) {
+		return `Codex error event (${meta.join(", ")})`;
+	}
+
+	try {
+		return `Codex error event: ${truncate(JSON.stringify(rawEvent), 800)}`;
+	} catch {
+		return "Codex error event";
+	}
+}

package/src/providers/openai-completions.ts

@@ -7,7 +7,7 @@ import type {
 	ChatCompletionContentPartText,
 	ChatCompletionMessageParam,
 	ChatCompletionToolMessageParam,
-} from "openai/resources/chat/completions.js";
+} from "openai/resources/chat/completions";
 import { calculateCost } from "../models";
 import { getEnvApiKey } from "../stream";
 import type {
@@ -26,6 +26,7 @@ import type {
 } from "../types";
 import { AssistantMessageEventStream } from "../utils/event-stream";
 import { parseStreamingJson } from "../utils/json-parse";
+import { formatErrorMessageWithRetryAfter } from "../utils/retry-after";
 import { sanitizeSurrogates } from "../utils/sanitize-unicode";
 import { transformMessages } from "./transorm-messages";
 
@@ -196,34 +197,44 @@ export const streamOpenAICompletions: StreamFunction<"openai-completions"> = (
 
 					// Some endpoints return reasoning in reasoning_content (llama.cpp),
 					// or reasoning (other openai compatible endpoints)
+					// Use the first non-empty reasoning field to avoid duplication
+					// (e.g., chutes.ai returns both reasoning_content and reasoning with same content)
 					const reasoningFields = ["reasoning_content", "reasoning", "reasoning_text"];
+					let foundReasoningField: string | null = null;
 					for (const field of reasoningFields) {
 						if (
 							(choice.delta as any)[field] !== null &&
 							(choice.delta as any)[field] !== undefined &&
 							(choice.delta as any)[field].length > 0
 						) {
-							if (!currentBlock || currentBlock.type !== "thinking") {
-								finishCurrentBlock(currentBlock);
-								currentBlock = {
-									type: "thinking",
-									thinking: "",
-									thinkingSignature: field,
-								};
-								output.content.push(currentBlock);
-								stream.push({ type: "thinking_start", contentIndex: blockIndex(), partial: output });
+							if (!foundReasoningField) {
+								foundReasoningField = field;
+								break;
 							}
+						}
+					}
 
-							if (currentBlock.type === "thinking") {
-								const delta = (choice.delta as any)[field];
-								currentBlock.thinking += delta;
-								stream.push({
-									type: "thinking_delta",
-									contentIndex: blockIndex(),
-									delta,
-									partial: output,
-								});
-							}
+					if (foundReasoningField) {
+						if (!currentBlock || currentBlock.type !== "thinking") {
+							finishCurrentBlock(currentBlock);
+							currentBlock = {
+								type: "thinking",
+								thinking: "",
+								thinkingSignature: foundReasoningField,
+							};
+							output.content.push(currentBlock);
+							stream.push({ type: "thinking_start", contentIndex: blockIndex(), partial: output });
+						}
+
+						if (currentBlock.type === "thinking") {
+							const delta = (choice.delta as any)[foundReasoningField];
+							currentBlock.thinking += delta;
+							stream.push({
+								type: "thinking_delta",
+								contentIndex: blockIndex(),
+								delta,
+								partial: output,
+							});
 						}
 					}
 
@@ -296,7 +307,7 @@ export const streamOpenAICompletions: StreamFunction<"openai-completions"> = (
 		} catch (error) {
 			for (const block of output.content) delete (block as any).index;
 			output.stopReason = options?.signal?.aborted ? "aborted" : "error";
-			output.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);
+			output.errorMessage = formatErrorMessageWithRetryAfter(error);
 			stream.push({ type: "error", reason: output.stopReason, error: output });
 			stream.end();
 		}
@@ -480,10 +491,8 @@ function convertMessages(
 			const nonEmptyThinkingBlocks = thinkingBlocks.filter((b) => b.thinking && b.thinking.trim().length > 0);
 			if (nonEmptyThinkingBlocks.length > 0) {
 				if (compat.requiresThinkingAsText) {
-					// Convert thinking blocks to text with <thinking> delimiters
-					const thinkingText = nonEmptyThinkingBlocks
-						.map((b) => `<thinking>\n${b.thinking}\n</thinking>`)
-						.join("\n");
+					// Convert thinking blocks to plain text (no tags to avoid model mimicking them)
+					const thinkingText = nonEmptyThinkingBlocks.map((b) => b.thinking).join("\n\n");
 					const textContent = assistantMsg.content as Array<{ type: "text"; text: string }> | null;
 					if (textContent) {
 						textContent.unshift({ type: "text", text: thinkingText });
@@ -633,8 +642,7 @@ function detectCompatFromUrl(baseUrl: string): Required<OpenAICompat> {
 		baseUrl.includes("cerebras.ai") ||
 		baseUrl.includes("api.x.ai") ||
 		baseUrl.includes("mistral.ai") ||
-		baseUrl.includes("chutes.ai") ||
-		baseUrl.includes("localhost");
+		baseUrl.includes("chutes.ai");
 
 	const useMaxTokens = baseUrl.includes("mistral.ai") || baseUrl.includes("chutes.ai");
 

package/src/providers/openai-responses.ts

@@ -9,7 +9,7 @@ import type {
 	ResponseInputText,
 	ResponseOutputMessage,
 	ResponseReasoningItem,
-} from "openai/resources/responses/responses.js";
+} from "openai/resources/responses/responses";
 import { calculateCost } from "../models";
 import { getEnvApiKey } from "../stream";
 import type {
@@ -27,6 +27,7 @@ import type {
 } from "../types";
 import { AssistantMessageEventStream } from "../utils/event-stream";
 import { parseStreamingJson } from "../utils/json-parse";
+import { formatErrorMessageWithRetryAfter } from "../utils/retry-after";
 import { sanitizeSurrogates } from "../utils/sanitize-unicode";
 import { transformMessages } from "./transorm-messages";
 
@@ -303,7 +304,7 @@ export const streamOpenAIResponses: StreamFunction<"openai-responses"> = (
 		} catch (error) {
 			for (const block of output.content) delete (block as any).index;
 			output.stopReason = options?.signal?.aborted ? "aborted" : "error";
-			output.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);
+			output.errorMessage = formatErrorMessageWithRetryAfter(error);
 			stream.push({ type: "error", reason: output.stopReason, error: output });
 			stream.end();
 		}

package/src/providers/transorm-messages.ts

@@ -45,12 +45,13 @@ export function transformMessages<TApi extends Api>(messages: Message[], model:
 				assistantMsg.api !== model.api;
 
 			// Transform message from different provider/model
-			const transformedContent = assistantMsg.content.map((block) => {
+			const transformedContent = assistantMsg.content.flatMap((block) => {
 				if (block.type === "thinking") {
-					// Convert thinking block to text block with <thinking> tags
+					// Skip empty thinking blocks, convert others to plain text
+					if (!block.thinking || block.thinking.trim() === "") return [];
 					return {
 						type: "text" as const,
-						text: `<thinking>\n${block.thinking}\n</thinking>`,
+						text: block.thinking,
 					};
 				}
 				// Normalize tool call IDs for github-copilot cross-API switches

package/src/stream.ts

@@ -21,8 +21,9 @@ import type {
 	KnownProvider,
 	Model,
 	OptionsForApi,
-	ReasoningEffort,
 	SimpleStreamOptions,
+	ThinkingBudgets,
+	ThinkingLevel,
 } from "./types";
 
 const VERTEX_ADC_CREDENTIALS_PATH = join(homedir(), ".config", "gcloud", "application_default_credentials.json");
@@ -64,7 +65,6 @@ export function getEnvApiKey(provider: any): string | undefined {
 		if (hasCredentials && hasProject && hasLocation) {
 			return "<authenticated>";
 		}
-		return undefined;
 	}
 
 	const envMap: Record<string, string> = {
@@ -76,6 +76,7 @@ export function getEnvApiKey(provider: any): string | undefined {
 		openrouter: "OPENROUTER_API_KEY",
 		zai: "ZAI_API_KEY",
 		mistral: "MISTRAL_API_KEY",
+		opencode: "OPENCODE_API_KEY",
 	};
 
 	const envVar = envMap[provider];
@@ -178,10 +179,11 @@ function mapOptionsForApi<TApi extends Api>(
 		maxTokens: options?.maxTokens || Math.min(model.maxTokens, 32000),
 		signal: options?.signal,
 		apiKey: apiKey || options?.apiKey,
+		sessionId: options?.sessionId,
 	};
 
 	// Helper to clamp xhigh to high for providers that don't support it
-	const clampReasoning = (effort: ReasoningEffort | undefined) => (effort === "xhigh" ? "high" : effort);
+	const clampReasoning = (effort: ThinkingLevel | undefined) => (effort === "xhigh" ? "high" : effort);
 
 	switch (model.api) {
 		case "anthropic-messages": {
@@ -192,15 +194,17 @@ function mapOptionsForApi<TApi extends Api>(
 
 			// Claude requires max_tokens > thinking.budget_tokens
 			// So we need to ensure maxTokens accounts for both thinking and output
-			const anthropicBudgets = {
+			const defaultBudgets: ThinkingBudgets = {
 				minimal: 1024,
 				low: 2048,
 				medium: 8192,
 				high: 16384,
 			};
+			const budgets = { ...defaultBudgets, ...options?.thinkingBudgets };
 
 			const minOutputTokens = 1024;
-			let thinkingBudget = anthropicBudgets[clampReasoning(options.reasoning)!];
+			const level = clampReasoning(options.reasoning)!;
+			let thinkingBudget = budgets[level]!;
 			// Caller's maxTokens is the desired output; add thinking budget on top, capped at model limit
 			const maxTokens = Math.min((base.maxTokens || 0) + thinkingBudget, model.maxTokens);
 
@@ -261,7 +265,7 @@ function mapOptionsForApi<TApi extends Api>(
 				...base,
 				thinking: {
 					enabled: true,
-					budgetTokens: getGoogleBudget(googleModel, effort),
+					budgetTokens: getGoogleBudget(googleModel, effort, options?.thinkingBudgets),
 				},
 			} satisfies GoogleOptions;
 		}
@@ -287,15 +291,16 @@ function mapOptionsForApi<TApi extends Api>(
 			// Models using thinkingBudget (Gemini 2.x, Claude via Antigravity)
 			// Claude requires max_tokens > thinking.budget_tokens
 			// So we need to ensure maxTokens accounts for both thinking and output
-			const budgets: Record<ClampedReasoningEffort, number> = {
+			const defaultBudgets: ThinkingBudgets = {
 				minimal: 1024,
 				low: 2048,
 				medium: 8192,
 				high: 16384,
 			};
+			const budgets = { ...defaultBudgets, ...options?.thinkingBudgets };
 
 			const minOutputTokens = 1024;
-			let thinkingBudget = budgets[effort];
+			let thinkingBudget = budgets[effort]!;
 			// Caller's maxTokens is the desired output; add thinking budget on top, capped at model limit
 			const maxTokens = Math.min((base.maxTokens || 0) + thinkingBudget, model.maxTokens);
 
@@ -316,22 +321,20 @@ function mapOptionsForApi<TApi extends Api>(
 
 		case "google-vertex": {
 			// Explicitly disable thinking when reasoning is not specified
-			// This is needed because Gemini has "dynamic thinking" enabled by default
 			if (!options?.reasoning) {
 				return { ...base, thinking: { enabled: false } } satisfies GoogleVertexOptions;
 			}
 
-			const googleModel = model as Model<"google-vertex">;
+			const vertexModel = model as Model<"google-vertex">;
 			const effort = clampReasoning(options.reasoning)!;
+			const geminiModel = vertexModel as unknown as Model<"google-generative-ai">;
 
-			// Gemini 3 models use thinkingLevel exclusively instead of thinkingBudget.
-			// https://ai.google.dev/gemini-api/docs/thinking#set-budget
-			if (isGemini3ProModel(googleModel) || isGemini3FlashModel(googleModel)) {
+			if (isGemini3ProModel(geminiModel) || isGemini3FlashModel(geminiModel)) {
 				return {
 					...base,
 					thinking: {
 						enabled: true,
-						level: getGemini3ThinkingLevel(effort, googleModel),
+						level: getGemini3ThinkingLevel(effort, geminiModel),
 					},
 				} satisfies GoogleVertexOptions;
 			}
@@ -340,7 +343,7 @@ function mapOptionsForApi<TApi extends Api>(
 				...base,
 				thinking: {
 					enabled: true,
-					budgetTokens: getGoogleBudget(googleModel, effort),
+					budgetTokens: getGoogleBudget(geminiModel, effort, options?.thinkingBudgets),
 				},
 			} satisfies GoogleVertexOptions;
 		}
@@ -353,21 +356,21 @@ function mapOptionsForApi<TApi extends Api>(
 	}
 }
 
-type ClampedReasoningEffort = Exclude<ReasoningEffort, "xhigh">;
+type ClampedThinkingLevel = Exclude<ThinkingLevel, "xhigh">;
 
-function isGemini3ProModel(model: Model<"google-generative-ai"> | Model<"google-vertex">): boolean {
+function isGemini3ProModel(model: Model<"google-generative-ai">): boolean {
 	// Covers gemini-3-pro, gemini-3-pro-preview, and possible other prefixed ids in the future
 	return model.id.includes("3-pro");
 }
 
-function isGemini3FlashModel(model: Model<"google-generative-ai"> | Model<"google-vertex">): boolean {
+function isGemini3FlashModel(model: Model<"google-generative-ai">): boolean {
 	// Covers gemini-3-flash, gemini-3-flash-preview, and possible other prefixed ids in the future
 	return model.id.includes("3-flash");
 }
 
 function getGemini3ThinkingLevel(
-	effort: ClampedReasoningEffort,
-	model: Model<"google-generative-ai"> | Model<"google-vertex">,
+	effort: ClampedThinkingLevel,
+	model: Model<"google-generative-ai">,
 ): GoogleThinkingLevel {
 	if (isGemini3ProModel(model)) {
 		// Gemini 3 Pro only supports LOW/HIGH (for now)
@@ -393,7 +396,7 @@ function getGemini3ThinkingLevel(
 	}
 }
 
-function getGeminiCliThinkingLevel(effort: ClampedReasoningEffort, modelId: string): GoogleThinkingLevel {
+function getGeminiCliThinkingLevel(effort: ClampedThinkingLevel, modelId: string): GoogleThinkingLevel {
 	if (modelId.includes("3-pro")) {
 		// Gemini 3 Pro only supports LOW/HIGH (for now)
 		switch (effort) {
@@ -419,12 +422,18 @@ function getGeminiCliThinkingLevel(effort: ClampedReasoningEffort, modelId: stri
 }
 
 function getGoogleBudget(
-	model: Model<"google-generative-ai"> | Model<"google-vertex">,
-	effort: ClampedReasoningEffort,
+	model: Model<"google-generative-ai">,
+	effort: ClampedThinkingLevel,
+	customBudgets?: ThinkingBudgets,
 ): number {
+	// Custom budgets take precedence if provided for this level
+	if (customBudgets?.[effort] !== undefined) {
+		return customBudgets[effort]!;
+	}
+
 	// See https://ai.google.dev/gemini-api/docs/thinking#set-budget
 	if (model.id.includes("2.5-pro")) {
-		const budgets: Record<ClampedReasoningEffort, number> = {
+		const budgets: Record<ClampedThinkingLevel, number> = {
 			minimal: 128,
 			low: 2048,
 			medium: 8192,
@@ -435,7 +444,7 @@ function getGoogleBudget(
 
 	if (model.id.includes("2.5-flash")) {
 		// Covers 2.5-flash-lite as well
-		const budgets: Record<ClampedReasoningEffort, number> = {
+		const budgets: Record<ClampedThinkingLevel, number> = {
 			minimal: 128,
 			low: 2048,
 			medium: 8192,

package/src/types.ts

@@ -54,10 +54,19 @@ export type KnownProvider =
 	| "cerebras"
 	| "openrouter"
 	| "zai"
-	| "mistral";
+	| "mistral"
+	| "opencode";
 export type Provider = KnownProvider | string;
 
-export type ReasoningEffort = "minimal" | "low" | "medium" | "high" | "xhigh";
+export type ThinkingLevel = "minimal" | "low" | "medium" | "high" | "xhigh";
+
+/** Token budgets for each thinking level (token-based providers only) */
+export interface ThinkingBudgets {
+	minimal?: number;
+	low?: number;
+	medium?: number;
+	high?: number;
+}
 
 // Base options all providers share
 export interface StreamOptions {
@@ -65,11 +74,19 @@ export interface StreamOptions {
 	maxTokens?: number;
 	signal?: AbortSignal;
 	apiKey?: string;
+	/**
+	 * Optional session identifier for providers that support session-based caching.
+	 * Providers can use this to enable prompt caching, request routing, or other
+	 * session-aware features. Ignored by providers that don't support it.
+	 */
+	sessionId?: string;
 }
 
 // Unified options with reasoning passed to streamSimple() and completeSimple()
 export interface SimpleStreamOptions extends StreamOptions {
-	reasoning?: ReasoningEffort;
+	reasoning?: ThinkingLevel;
+	/** Custom token budgets for thinking levels (token-based providers only) */
+	thinkingBudgets?: ThinkingBudgets;
 }
 
 // Generic StreamFunction with typed options
@@ -146,7 +163,7 @@ export interface ToolResultMessage<TDetails = any> {
 	toolName: string;
 	content: (TextContent | ImageContent)[]; // Supports text and images
 	details?: TDetails;
-	isError?: boolean;
+	isError: boolean;
 	timestamp: number; // Unix timestamp in milliseconds
 }