merge: sync upstream pi-mono main into sproutclaw

Merge upstream/main (195 commits) while keeping sproutclaw-specific README, webui workflow docs, and restored upstream .pi prompt templates. Co-authored-by: Cursor <cursoragent@cursor.com>
2026-05-22 20:05:31 +08:00
parent 5f7a53ed21 7426ce977d
commit a8daa1d8ff
642 changed files with 18785 additions and 26436 deletions
--- a/packages/ai/src/api-registry.ts
+++ b/packages/ai/src/api-registry.ts
@@ -6,7 +6,7 @@ import type {
 	SimpleStreamOptions,
 	StreamFunction,
 	StreamOptions,
-} from "./types.js";
+} from "./types.ts";

 export type ApiStreamFunction = (
 	model: Model<Api>,
--- a/packages/ai/src/bedrock-provider.ts
+++ b/packages/ai/src/bedrock-provider.ts
@@ -1,4 +1,4 @@
-import { streamBedrock, streamSimpleBedrock } from "./providers/amazon-bedrock.js";
+import { streamBedrock, streamSimpleBedrock } from "./providers/amazon-bedrock.ts";

 export const bedrockProviderModule = {
 	streamBedrock,
--- a/packages/ai/src/cli.ts
+++ b/packages/ai/src/cli.ts
@@ -2,8 +2,8 @@

 import { createInterface } from "node:readline";
 import { existsSync, readFileSync, writeFileSync } from "fs";
-import { getOAuthProvider, getOAuthProviders } from "./utils/oauth/index.js";
-import type { OAuthCredentials, OAuthProviderId } from "./utils/oauth/types.js";
+import { getOAuthProvider, getOAuthProviders } from "./utils/oauth/index.ts";
+import type { OAuthCredentials, OAuthProviderId } from "./utils/oauth/types.ts";

 const AUTH_FILE = "auth.json";
 const PROVIDERS = getOAuthProviders();
--- a/packages/ai/src/env-api-keys.ts
+++ b/packages/ai/src/env-api-keys.ts
@@ -1,4 +1,4 @@
-// NEVER convert to top-level imports - breaks browser/Vite builds (web-ui)
+// NEVER convert to top-level imports - breaks browser/Vite builds
 let _existsSync: typeof import("node:fs").existsSync | null = null;
 let _homedir: typeof import("node:os").homedir | null = null;
 let _join: typeof import("node:path").join | null = null;
@@ -23,7 +23,7 @@ if (typeof process !== "undefined" && (process.versions?.node || process.version
 	});
 }

-import type { KnownProvider } from "./types.js";
+import type { KnownProvider } from "./types.ts";

 let _procEnvCache: Map<string, string> | null = null;

--- a/packages/ai/src/image-models.generated.ts
+++ b/packages/ai/src/image-models.generated.ts
@@ -1,7 +1,7 @@
 // This file is auto-generated by scripts/generate-image-models.ts
 // Do not edit manually - run 'npm run generate-image-models' to update

-import type { ImagesApi, ImagesModel } from "./types.js";
+import type { ImagesApi, ImagesModel } from "./types.ts";

 export const IMAGE_MODELS = {
 	"openrouter": {
@@ -230,6 +230,171 @@ export const IMAGE_MODELS = {
 	  "cacheWrite": 0
 	}
 		} satisfies ImagesModel<"openrouter-images">,
+		"recraft/recraft-v3": {
+			id: "recraft/recraft-v3",
+			name: "Recraft: Recraft V3",
+			api: "openrouter-images",
+			provider: "openrouter",
+			baseUrl: "https://openrouter.ai/api/v1",
+			input: ["text", "image"],
+			output: ["image"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+		} satisfies ImagesModel<"openrouter-images">,
+		"recraft/recraft-v4": {
+			id: "recraft/recraft-v4",
+			name: "Recraft: Recraft V4",
+			api: "openrouter-images",
+			provider: "openrouter",
+			baseUrl: "https://openrouter.ai/api/v1",
+			input: ["text", "image"],
+			output: ["image"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+		} satisfies ImagesModel<"openrouter-images">,
+		"recraft/recraft-v4-pro": {
+			id: "recraft/recraft-v4-pro",
+			name: "Recraft: Recraft V4 Pro",
+			api: "openrouter-images",
+			provider: "openrouter",
+			baseUrl: "https://openrouter.ai/api/v1",
+			input: ["text", "image"],
+			output: ["image"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+		} satisfies ImagesModel<"openrouter-images">,
+		"recraft/recraft-v4-pro-vector": {
+			id: "recraft/recraft-v4-pro-vector",
+			name: "Recraft: Recraft V4 Pro Vector",
+			api: "openrouter-images",
+			provider: "openrouter",
+			baseUrl: "https://openrouter.ai/api/v1",
+			input: ["text", "image"],
+			output: ["image"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+		} satisfies ImagesModel<"openrouter-images">,
+		"recraft/recraft-v4-vector": {
+			id: "recraft/recraft-v4-vector",
+			name: "Recraft: Recraft V4 Vector",
+			api: "openrouter-images",
+			provider: "openrouter",
+			baseUrl: "https://openrouter.ai/api/v1",
+			input: ["text", "image"],
+			output: ["image"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+		} satisfies ImagesModel<"openrouter-images">,
+		"recraft/recraft-v4.1": {
+			id: "recraft/recraft-v4.1",
+			name: "Recraft: Recraft V4.1",
+			api: "openrouter-images",
+			provider: "openrouter",
+			baseUrl: "https://openrouter.ai/api/v1",
+			input: ["text", "image"],
+			output: ["image"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+		} satisfies ImagesModel<"openrouter-images">,
+		"recraft/recraft-v4.1-pro": {
+			id: "recraft/recraft-v4.1-pro",
+			name: "Recraft: Recraft V4.1 Pro",
+			api: "openrouter-images",
+			provider: "openrouter",
+			baseUrl: "https://openrouter.ai/api/v1",
+			input: ["text", "image"],
+			output: ["image"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+		} satisfies ImagesModel<"openrouter-images">,
+		"recraft/recraft-v4.1-pro-vector": {
+			id: "recraft/recraft-v4.1-pro-vector",
+			name: "Recraft: Recraft V4.1 Pro Vector",
+			api: "openrouter-images",
+			provider: "openrouter",
+			baseUrl: "https://openrouter.ai/api/v1",
+			input: ["text", "image"],
+			output: ["image"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+		} satisfies ImagesModel<"openrouter-images">,
+		"recraft/recraft-v4.1-utility": {
+			id: "recraft/recraft-v4.1-utility",
+			name: "Recraft: Recraft V4.1 Utility",
+			api: "openrouter-images",
+			provider: "openrouter",
+			baseUrl: "https://openrouter.ai/api/v1",
+			input: ["text", "image"],
+			output: ["image"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+		} satisfies ImagesModel<"openrouter-images">,
+		"recraft/recraft-v4.1-utility-pro": {
+			id: "recraft/recraft-v4.1-utility-pro",
+			name: "Recraft: Recraft V4.1 Utility Pro",
+			api: "openrouter-images",
+			provider: "openrouter",
+			baseUrl: "https://openrouter.ai/api/v1",
+			input: ["text", "image"],
+			output: ["image"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+		} satisfies ImagesModel<"openrouter-images">,
+		"recraft/recraft-v4.1-vector": {
+			id: "recraft/recraft-v4.1-vector",
+			name: "Recraft: Recraft V4.1 Vector",
+			api: "openrouter-images",
+			provider: "openrouter",
+			baseUrl: "https://openrouter.ai/api/v1",
+			input: ["text", "image"],
+			output: ["image"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+		} satisfies ImagesModel<"openrouter-images">,
 		"sourceful/riverflow-v2-fast": {
 			id: "sourceful/riverflow-v2-fast",
 			name: "Sourceful: Riverflow V2 Fast",
@@ -305,5 +470,20 @@ export const IMAGE_MODELS = {
 	  "cacheWrite": 0
 	}
 		} satisfies ImagesModel<"openrouter-images">,
+		"x-ai/grok-imagine-image-quality": {
+			id: "x-ai/grok-imagine-image-quality",
+			name: "xAI: Grok Imagine Image Quality",
+			api: "openrouter-images",
+			provider: "openrouter",
+			baseUrl: "https://openrouter.ai/api/v1",
+			input: ["text", "image"],
+			output: ["image"],
+			cost: {
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+		} satisfies ImagesModel<"openrouter-images">,
 	},
 } as const satisfies Record<string, Record<string, ImagesModel<ImagesApi>>>;
--- a/packages/ai/src/image-models.ts
+++ b/packages/ai/src/image-models.ts
@@ -1,5 +1,5 @@
-import { IMAGE_MODELS } from "./image-models.generated.js";
-import type { ImagesApi, ImagesModel, KnownImagesProvider } from "./types.js";
+import { IMAGE_MODELS } from "./image-models.generated.ts";
+import type { ImagesApi, ImagesModel, KnownImagesProvider } from "./types.ts";

 const imageModelRegistry: Map<string, Map<string, ImagesModel<ImagesApi>>> = new Map();

--- a/packages/ai/src/images-api-registry.ts
+++ b/packages/ai/src/images-api-registry.ts
@@ -1,4 +1,4 @@
-import type { AssistantImages, ImagesApi, ImagesContext, ImagesFunction, ImagesModel, ImagesOptions } from "./types.js";
+import type { AssistantImages, ImagesApi, ImagesContext, ImagesFunction, ImagesModel, ImagesOptions } from "./types.ts";

 export type ImagesApiFunction = (
 	model: ImagesModel<ImagesApi>,
--- a/packages/ai/src/images.ts
+++ b/packages/ai/src/images.ts
@@ -1,7 +1,7 @@
-import "./providers/images/register-builtins.js";
+import "./providers/images/register-builtins.ts";

-import { getImagesApiProvider } from "./images-api-registry.js";
-import type { AssistantImages, ImagesApi, ImagesContext, ImagesModel, ProviderImagesOptions } from "./types.js";
+import { getImagesApiProvider } from "./images-api-registry.ts";
+import type { AssistantImages, ImagesApi, ImagesContext, ImagesModel, ProviderImagesOptions } from "./types.ts";

 function resolveImagesApiProvider(api: ImagesApi) {
 	const provider = getImagesApiProvider(api);
--- a/packages/ai/src/index.ts
+++ b/packages/ai/src/index.ts
@@ -1,34 +1,34 @@
 export type { Static, TSchema } from "typebox";
 export { Type } from "typebox";

-export * from "./api-registry.js";
-export * from "./env-api-keys.js";
-export * from "./image-models.js";
-export * from "./images.js";
-export * from "./images-api-registry.js";
-export * from "./models.js";
-export type { BedrockOptions, BedrockThinkingDisplay } from "./providers/amazon-bedrock.js";
-export type { AnthropicEffort, AnthropicOptions, AnthropicThinkingDisplay } from "./providers/anthropic.js";
-export type { AzureOpenAIResponsesOptions } from "./providers/azure-openai-responses.js";
-export * from "./providers/faux.js";
-export type { GoogleOptions } from "./providers/google.js";
-export type { GoogleThinkingLevel } from "./providers/google-shared.js";
-export type { GoogleVertexOptions } from "./providers/google-vertex.js";
-export * from "./providers/images/register-builtins.js";
-export type { MistralOptions } from "./providers/mistral.js";
+export * from "./api-registry.ts";
+export * from "./env-api-keys.ts";
+export * from "./image-models.ts";
+export * from "./images.ts";
+export * from "./images-api-registry.ts";
+export * from "./models.ts";
+export type { BedrockOptions, BedrockThinkingDisplay } from "./providers/amazon-bedrock.ts";
+export type { AnthropicEffort, AnthropicOptions, AnthropicThinkingDisplay } from "./providers/anthropic.ts";
+export type { AzureOpenAIResponsesOptions } from "./providers/azure-openai-responses.ts";
+export * from "./providers/faux.ts";
+export type { GoogleOptions } from "./providers/google.ts";
+export type { GoogleThinkingLevel } from "./providers/google-shared.ts";
+export type { GoogleVertexOptions } from "./providers/google-vertex.ts";
+export * from "./providers/images/register-builtins.ts";
+export type { MistralOptions } from "./providers/mistral.ts";
 export type {
 	OpenAICodexResponsesOptions,
 	OpenAICodexWebSocketDebugStats,
-} from "./providers/openai-codex-responses.js";
-export type { OpenAICompletionsOptions } from "./providers/openai-completions.js";
-export type { OpenAIResponsesOptions } from "./providers/openai-responses.js";
-export * from "./providers/register-builtins.js";
-export * from "./session-resources.js";
-export * from "./stream.js";
-export * from "./types.js";
-export * from "./utils/diagnostics.js";
-export * from "./utils/event-stream.js";
-export * from "./utils/json-parse.js";
+} from "./providers/openai-codex-responses.ts";
+export type { OpenAICompletionsOptions } from "./providers/openai-completions.ts";
+export type { OpenAIResponsesOptions } from "./providers/openai-responses.ts";
+export * from "./providers/register-builtins.ts";
+export * from "./session-resources.ts";
+export * from "./stream.ts";
+export * from "./types.ts";
+export * from "./utils/diagnostics.ts";
+export * from "./utils/event-stream.ts";
+export * from "./utils/json-parse.ts";
 export type {
 	OAuthAuthInfo,
 	OAuthCredentials,
@@ -40,7 +40,7 @@ export type {
 	OAuthProviderInterface,
 	OAuthSelectOption,
 	OAuthSelectPrompt,
-} from "./utils/oauth/types.js";
-export * from "./utils/overflow.js";
-export * from "./utils/typebox-helpers.js";
-export * from "./utils/validation.js";
+} from "./utils/oauth/types.ts";
+export * from "./utils/overflow.ts";
+export * from "./utils/typebox-helpers.ts";
+export * from "./utils/validation.ts";
--- a/packages/ai/src/models.generated.ts
+++ b/packages/ai/src/models.generated.ts
--- a/packages/ai/src/models.ts
+++ b/packages/ai/src/models.ts
@@ -1,5 +1,5 @@
-import { MODELS } from "./models.generated.js";
-import type { Api, KnownProvider, Model, ModelThinkingLevel, Usage } from "./types.js";
+import { MODELS } from "./models.generated.ts";
+import type { Api, KnownProvider, Model, ModelThinkingLevel, Usage } from "./types.ts";

 const modelRegistry: Map<string, Map<string, Model<Api>>> = new Map();

--- a/packages/ai/src/oauth.ts
+++ b/packages/ai/src/oauth.ts
@@ -1 +1 @@
-export * from "./utils/oauth/index.js";
+export * from "./utils/oauth/index.ts";
--- a/packages/ai/src/providers/amazon-bedrock.ts
+++ b/packages/ai/src/providers/amazon-bedrock.ts
@@ -20,8 +20,9 @@ import {
 	type ToolConfiguration,
 	ToolResultStatus,
 } from "@aws-sdk/client-bedrock-runtime";
+import { NodeHttpHandler } from "@smithy/node-http-handler";
 import type { DocumentType } from "@smithy/types";
-import { calculateCost } from "../models.js";
+import { calculateCost } from "../models.ts";
 import type {
 	Api,
 	AssistantMessage,
@@ -39,12 +40,13 @@ import type {
 	Tool,
 	ToolCall,
 	ToolResultMessage,
-} from "../types.js";
-import { AssistantMessageEventStream } from "../utils/event-stream.js";
-import { parseStreamingJson } from "../utils/json-parse.js";
-import { sanitizeSurrogates } from "../utils/sanitize-unicode.js";
-import { adjustMaxTokensForThinking, buildBaseOptions, clampReasoning } from "./simple-options.js";
-import { transformMessages } from "./transform-messages.js";
+} from "../types.ts";
+import { AssistantMessageEventStream } from "../utils/event-stream.ts";
+import { parseStreamingJson } from "../utils/json-parse.ts";
+import { createHttpProxyAgentsForTarget } from "../utils/node-http-proxy.ts";
+import { sanitizeSurrogates } from "../utils/sanitize-unicode.ts";
+import { adjustMaxTokensForThinking, buildBaseOptions, clampReasoning } from "./simple-options.ts";
+import { transformMessages } from "./transform-messages.ts";

 export type BedrockThinkingDisplay = "summarized" | "omitted";

@@ -156,30 +158,15 @@ export const streamBedrock: StreamFunction<"bedrock-converse-stream", BedrockOpt
 				};
 			}

-			if (
-				process.env.HTTP_PROXY ||
-				process.env.HTTPS_PROXY ||
-				process.env.NO_PROXY ||
-				process.env.http_proxy ||
-				process.env.https_proxy ||
-				process.env.no_proxy
-			) {
-				const nodeHttpHandler = await import("@smithy/node-http-handler");
-				const proxyAgent = await import("proxy-agent");
-
-				const agent = new proxyAgent.ProxyAgent();
-
+			const proxyAgents = createHttpProxyAgentsForTarget(model.baseUrl);
+			if (proxyAgents) {
 				// Bedrock runtime uses NodeHttp2Handler by default since v3.798.0, which is based
 				// on `http2` module and has no support for http agent.
-				// Use NodeHttpHandler to support http agent.
-				config.requestHandler = new nodeHttpHandler.NodeHttpHandler({
-					httpAgent: agent,
-					httpsAgent: agent,
-				});
+				// Use NodeHttpHandler to support HTTP(S) proxy agents.
+				config.requestHandler = new NodeHttpHandler(proxyAgents);
 			} else if (process.env.AWS_BEDROCK_FORCE_HTTP1 === "1") {
 				// Some custom endpoints require HTTP/1.1 instead of HTTP/2
-				const nodeHttpHandler = await import("@smithy/node-http-handler");
-				config.requestHandler = new nodeHttpHandler.NodeHttpHandler();
+				config.requestHandler = new NodeHttpHandler();
 			}
 		} else {
 			// Non-Node environment (browser): fall back to us-east-1 since
@@ -196,12 +183,13 @@ export const streamBedrock: StreamFunction<"bedrock-converse-stream", BedrockOpt
 		try {
 			const client = new BedrockRuntimeClient(config);
 			const cacheRetention = resolveCacheRetention(options.cacheRetention);
+			const inferenceMaxTokens = options.maxTokens ?? (isAnthropicClaudeModel(model) ? model.maxTokens : undefined);
 			let commandInput = {
 				modelId: model.id,
 				messages: convertMessages(context, model, cacheRetention),
 				system: buildSystemPrompt(context.systemPrompt, model, cacheRetention),
 				inferenceConfig: {
-					...(options.maxTokens !== undefined && { maxTokens: options.maxTokens }),
+					...(inferenceMaxTokens !== undefined && { maxTokens: inferenceMaxTokens }),
 					...(options.temperature !== undefined && { temperature: options.temperature }),
 				},
 				toolConfig: convertToolConfig(context.tools, options.toolChoice),
@@ -327,8 +315,10 @@ export const streamSimpleBedrock: StreamFunction<"bedrock-converse-stream", Simp
 			} satisfies BedrockOptions);
 		}

+		// Undefined means the caller did not request an output cap; let the helper use the model cap.
+		// Do not coerce to 0 here, or the thinking budget would become the entire maxTokens value.
 		const adjusted = adjustMaxTokensForThinking(
-			base.maxTokens || 0,
+			base.maxTokens,
 			model.maxTokens,
 			options.reasoning,
 			options.thinkingBudgets,
@@ -631,24 +621,31 @@ function convertMessages(
 		const m = transformedMessages[i];

 		switch (m.role) {
-			case "user":
+			case "user": {
+				const content: ContentBlock[] = [];
+				if (typeof m.content === "string") {
+					content.push({ text: sanitizeSurrogates(m.content) });
+				} else {
+					for (const c of m.content) {
+						switch (c.type) {
+							case "text":
+								content.push({ text: sanitizeSurrogates(c.text) });
+								break;
+							case "image":
+								content.push({ image: createImageBlock(c.mimeType, c.data) });
+								break;
+							default:
+								continue;
+						}
+					}
+				}
+				if (content.length === 0) continue;
 				result.push({
 					role: ConversationRole.USER,
-					content:
-						typeof m.content === "string"
-							? [{ text: sanitizeSurrogates(m.content) }]
-							: m.content.map((c) => {
-									switch (c.type) {
-										case "text":
-											return { text: sanitizeSurrogates(c.text) };
-										case "image":
-											return { image: createImageBlock(c.mimeType, c.data) };
-										default:
-											throw new Error("Unknown user content type");
-									}
-								}),
+					content,
 				});
 				break;
+			}
 			case "assistant": {
 				// Skip assistant messages with empty content (e.g., from aborted requests)
 				// Bedrock rejects messages with empty content arrays
@@ -699,7 +696,7 @@ function convertMessages(
 							}
 							break;
 						default:
-							throw new Error("Unknown assistant content type");
+							continue;
 					}
 				}
 				// Skip if all content blocks were filtered out
@@ -758,7 +755,7 @@ function convertMessages(
 				break;
 			}
 			default:
-				throw new Error("Unknown message role");
+				continue;
 		}
 	}

--- a/packages/ai/src/providers/anthropic.ts
+++ b/packages/ai/src/providers/anthropic.ts
@@ -6,8 +6,8 @@ import type {
 	MessageParam,
 	RawMessageStreamEvent,
 } from "@anthropic-ai/sdk/resources/messages.js";
-import { getEnvApiKey } from "../env-api-keys.js";
-import { calculateCost } from "../models.js";
+import { getEnvApiKey } from "../env-api-keys.ts";
+import { calculateCost } from "../models.ts";
 import type {
 	AnthropicMessagesCompat,
 	Api,
@@ -26,16 +26,16 @@ import type {
 	Tool,
 	ToolCall,
 	ToolResultMessage,
-} from "../types.js";
-import { AssistantMessageEventStream } from "../utils/event-stream.js";
-import { headersToRecord } from "../utils/headers.js";
-import { parseJsonWithRepair, parseStreamingJson } from "../utils/json-parse.js";
-import { sanitizeSurrogates } from "../utils/sanitize-unicode.js";
+} from "../types.ts";
+import { AssistantMessageEventStream } from "../utils/event-stream.ts";
+import { headersToRecord } from "../utils/headers.ts";
+import { parseJsonWithRepair, parseStreamingJson } from "../utils/json-parse.ts";
+import { sanitizeSurrogates } from "../utils/sanitize-unicode.ts";

-import { resolveCloudflareBaseUrl } from "./cloudflare.js";
-import { buildCopilotDynamicHeaders, hasCopilotVisionInput } from "./github-copilot-headers.js";
-import { adjustMaxTokensForThinking, buildBaseOptions } from "./simple-options.js";
-import { transformMessages } from "./transform-messages.js";
+import { resolveCloudflareBaseUrl } from "./cloudflare.ts";
+import { buildCopilotDynamicHeaders, hasCopilotVisionInput } from "./github-copilot-headers.ts";
+import { adjustMaxTokensForThinking, buildBaseOptions } from "./simple-options.ts";
+import { transformMessages } from "./transform-messages.ts";

 /**
 * Resolve cache retention preference.
@@ -165,9 +165,16 @@ const FINE_GRAINED_TOOL_STREAMING_BETA = "fine-grained-tool-streaming-2025-05-14
 const INTERLEAVED_THINKING_BETA = "interleaved-thinking-2025-05-14";

 function getAnthropicCompat(model: Model<"anthropic-messages">): Required<AnthropicMessagesCompat> {
+	// Auto-detect session affinity and cache control support from provider
+	const isFireworks = model.provider === "fireworks";
+	const isCloudflareAiGatewayAnthropic =
+		model.provider === "cloudflare-ai-gateway" && model.baseUrl.includes("anthropic");
 	return {
-		supportsEagerToolInputStreaming: model.compat?.supportsEagerToolInputStreaming ?? true,
-		supportsLongCacheRetention: model.compat?.supportsLongCacheRetention ?? true,
+		supportsEagerToolInputStreaming: model.compat?.supportsEagerToolInputStreaming ?? !isFireworks,
+		supportsLongCacheRetention: model.compat?.supportsLongCacheRetention ?? !isFireworks,
+		sendSessionAffinityHeaders:
+			model.compat?.sendSessionAffinityHeaders ?? !!(isFireworks || isCloudflareAiGatewayAnthropic),
+		supportsCacheControlOnTools: model.compat?.supportsCacheControlOnTools ?? !isFireworks,
 	};
 }

@@ -463,6 +470,9 @@ export const streamAnthropic: StreamFunction<"anthropic-messages", AnthropicOpti
 					});
 				}

+				const cacheRetention = options?.cacheRetention ?? resolveCacheRetention();
+				const cacheSessionId = cacheRetention === "none" ? undefined : options?.sessionId;
+
 				const created = createClient(
 					model,
 					apiKey,
@@ -470,6 +480,7 @@ export const streamAnthropic: StreamFunction<"anthropic-messages", AnthropicOpti
 					shouldUseFineGrainedToolStreamingBeta(model, context),
 					options?.headers,
 					copilotDynamicHeaders,
+					cacheSessionId,
 				);
 				client = created.client;
 				isOAuth = created.isOAuthToken;
@@ -740,8 +751,10 @@ export const streamSimpleAnthropic: StreamFunction<"anthropic-messages", SimpleS
 		} satisfies AnthropicOptions);
 	}

+	// Undefined means the caller did not request an output cap; let the helper use the model cap.
+	// Do not coerce to 0 here, or the thinking budget would become the entire max_tokens value.
 	const adjusted = adjustMaxTokensForThinking(
-		base.maxTokens || 0,
+		base.maxTokens,
 		model.maxTokens,
 		options.reasoning,
 		options.thinkingBudgets,
@@ -766,6 +779,7 @@ function createClient(
 	useFineGrainedToolStreamingBeta: boolean,
 	optionsHeaders?: Record<string, string>,
 	dynamicHeaders?: Record<string, string>,
+	sessionId?: string,
 ): { client: Anthropic; isOAuthToken: boolean } {
 	// Adaptive thinking models (Opus 4.6, Sonnet 4.6) have interleaved thinking built-in.
 	// The beta header is deprecated on Opus 4.6 and redundant on Sonnet 4.6, so skip it.
@@ -847,8 +861,11 @@ function createClient(
 	}

 	// API key auth
+	const sessionAffinityHeaders: Record<string, string | null> =
+		sessionId && getAnthropicCompat(model).sendSessionAffinityHeaders ? { "x-session-affinity": sessionId } : {};
 	const client = new Anthropic({
 		apiKey,
+		authToken: null,
 		baseURL: model.baseUrl,
 		dangerouslyAllowBrowser: true,
 		defaultHeaders: mergeHeaders(
@@ -857,6 +874,7 @@ function createClient(
 				"anthropic-dangerous-direct-browser-access": "true",
 				...(betaFeatures.length > 0 ? { "anthropic-beta": betaFeatures.join(",") } : {}),
 			},
+			sessionAffinityHeaders,
 			model.headers,
 			optionsHeaders,
 		),
@@ -875,7 +893,7 @@ function buildParams(
 	const params: MessageCreateParamsStreaming = {
 		model: model.id,
 		messages: convertMessages(context.messages, model, isOAuthToken, cacheControl),
-		max_tokens: options?.maxTokens || (model.maxTokens / 3) | 0,
+		max_tokens: options?.maxTokens ?? model.maxTokens,
 		stream: true,
 	};

@@ -912,11 +930,12 @@ function buildParams(
 	}

 	if (context.tools && context.tools.length > 0) {
+		const compat = getAnthropicCompat(model);
 		params.tools = convertTools(
 			context.tools,
 			isOAuthToken,
-			getAnthropicCompat(model).supportsEagerToolInputStreaming,
-			cacheControl,
+			compat.supportsEagerToolInputStreaming,
+			compat.supportsCacheControlOnTools ? cacheControl : undefined,
 		);
 	}

--- a/packages/ai/src/providers/azure-openai-responses.ts
+++ b/packages/ai/src/providers/azure-openai-responses.ts
@@ -1,7 +1,7 @@
 import { AzureOpenAI } from "openai";
 import type { ResponseCreateParamsStreaming } from "openai/resources/responses/responses.js";
-import { getEnvApiKey } from "../env-api-keys.js";
-import { clampThinkingLevel } from "../models.js";
+import { getEnvApiKey } from "../env-api-keys.ts";
+import { clampThinkingLevel } from "../models.ts";
 import type {
 	Api,
 	AssistantMessage,
@@ -10,11 +10,12 @@ import type {
 	SimpleStreamOptions,
 	StreamFunction,
 	StreamOptions,
-} from "../types.js";
-import { AssistantMessageEventStream } from "../utils/event-stream.js";
-import { headersToRecord } from "../utils/headers.js";
-import { convertResponsesMessages, convertResponsesTools, processResponsesStream } from "./openai-responses-shared.js";
-import { buildBaseOptions } from "./simple-options.js";
+} from "../types.ts";
+import { AssistantMessageEventStream } from "../utils/event-stream.ts";
+import { headersToRecord } from "../utils/headers.ts";
+import { clampOpenAIPromptCacheKey } from "./openai-prompt-cache.ts";
+import { convertResponsesMessages, convertResponsesTools, processResponsesStream } from "./openai-responses-shared.ts";
+import { buildBaseOptions } from "./simple-options.ts";

 const DEFAULT_AZURE_API_VERSION = "v1";
 const AZURE_TOOL_CALL_PROVIDERS = new Set(["openai", "openai-codex", "opencode", "azure-openai-responses"]);
@@ -40,6 +41,22 @@ function resolveDeploymentName(model: Model<"azure-openai-responses">, options?:
 	return mappedDeployment || model.id;
 }

+function formatAzureOpenAIError(error: unknown): string {
+	if (error instanceof Error) {
+		const status = (error as Error & { status?: unknown }).status;
+		const statusCode = typeof status === "number" ? status : undefined;
+		if (statusCode !== undefined) {
+			return `Azure OpenAI API error (${statusCode}): ${error.message}`;
+		}
+		return error.message;
+	}
+	try {
+		return JSON.stringify(error);
+	} catch {
+		return String(error);
+	}
+}
+
 // Azure OpenAI Responses-specific options
 export interface AzureOpenAIResponsesOptions extends StreamOptions {
 	reasoningEffort?: "minimal" | "low" | "medium" | "high" | "xhigh";
@@ -119,7 +136,7 @@ export const streamAzureOpenAIResponses: StreamFunction<"azure-openai-responses"
 				delete (block as { partialJson?: string }).partialJson;
 			}
 			output.stopReason = options?.signal?.aborted ? "aborted" : "error";
-			output.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);
+			output.errorMessage = formatAzureOpenAIError(error);
 			stream.push({ type: "error", reason: output.stopReason, error: output });
 			stream.end();
 		}
@@ -245,7 +262,7 @@ function buildParams(
 		model: deploymentName,
 		input: messages,
 		stream: true,
-		prompt_cache_key: options?.sessionId,
+		prompt_cache_key: clampOpenAIPromptCacheKey(options?.sessionId),
 	};

 	if (options?.maxTokens) {
--- a/packages/ai/src/providers/cloudflare.ts
+++ b/packages/ai/src/providers/cloudflare.ts
@@ -1,4 +1,4 @@
-import type { Api, Model } from "../types.js";
+import type { Api, Model } from "../types.ts";

 /** Workers AI direct endpoint. */
 export const CLOUDFLARE_WORKERS_AI_BASE_URL =
--- a/packages/ai/src/providers/faux.ts
+++ b/packages/ai/src/providers/faux.ts
@@ -1,4 +1,4 @@
-import { registerApiProvider, unregisterApiProviders } from "../api-registry.js";
+import { registerApiProvider, unregisterApiProviders } from "../api-registry.ts";
 import type {
 	AssistantMessage,
 	AssistantMessageEventStream,
@@ -14,8 +14,8 @@ import type {
 	ToolCall,
 	ToolResultMessage,
 	Usage,
-} from "../types.js";
-import { createAssistantMessageEventStream } from "../utils/event-stream.js";
+} from "../types.ts";
+import { createAssistantMessageEventStream } from "../utils/event-stream.ts";

 const DEFAULT_API = "faux";
 const DEFAULT_PROVIDER = "faux";
--- a/packages/ai/src/providers/github-copilot-headers.ts
+++ b/packages/ai/src/providers/github-copilot-headers.ts
@@ -1,4 +1,4 @@
-import type { Message } from "../types.js";
+import type { Message } from "../types.ts";

 // Copilot expects X-Initiator to indicate whether the request is user-initiated
 // or agent-initiated (e.g. follow-up after assistant/tool messages).
--- a/packages/ai/src/providers/google-shared.ts
+++ b/packages/ai/src/providers/google-shared.ts
@@ -3,9 +3,9 @@
 */

 import { type Content, FinishReason, FunctionCallingConfigMode, type Part } from "@google/genai";
-import type { Context, ImageContent, Model, StopReason, TextContent, Tool } from "../types.js";
-import { sanitizeSurrogates } from "../utils/sanitize-unicode.js";
-import { transformMessages } from "./transform-messages.js";
+import type { Context, ImageContent, Model, StopReason, TextContent, Tool } from "../types.ts";
+import { sanitizeSurrogates } from "../utils/sanitize-unicode.ts";
+import { transformMessages } from "./transform-messages.ts";

 type GoogleApiType = "google-generative-ai" | "google-vertex";

--- a/packages/ai/src/providers/google-vertex.ts
+++ b/packages/ai/src/providers/google-vertex.ts
@@ -7,7 +7,7 @@ import {
 	type ThinkingConfig,
 	ThinkingLevel,
 } from "@google/genai";
-import { calculateCost, clampThinkingLevel } from "../models.js";
+import { calculateCost, clampThinkingLevel } from "../models.ts";
 import type {
 	Api,
 	AssistantMessage,
@@ -21,10 +21,10 @@ import type {
 	ThinkingBudgets,
 	ThinkingContent,
 	ToolCall,
-} from "../types.js";
-import { AssistantMessageEventStream } from "../utils/event-stream.js";
-import { sanitizeSurrogates } from "../utils/sanitize-unicode.js";
-import type { GoogleThinkingLevel } from "./google-shared.js";
+} from "../types.ts";
+import { AssistantMessageEventStream } from "../utils/event-stream.ts";
+import { sanitizeSurrogates } from "../utils/sanitize-unicode.ts";
+import type { GoogleThinkingLevel } from "./google-shared.ts";
 import {
 	convertMessages,
 	convertTools,
@@ -32,8 +32,8 @@ import {
 	mapStopReason,
 	mapToolChoice,
 	retainThoughtSignature,
-} from "./google-shared.js";
-import { buildBaseOptions } from "./simple-options.js";
+} from "./google-shared.ts";
+import { buildBaseOptions } from "./simple-options.ts";

 export interface GoogleVertexOptions extends StreamOptions {
 	toolChoice?: "auto" | "none" | "any";
--- a/packages/ai/src/providers/google.ts
+++ b/packages/ai/src/providers/google.ts
@@ -4,8 +4,8 @@ import {
 	GoogleGenAI,
 	type ThinkingConfig,
 } from "@google/genai";
-import { getEnvApiKey } from "../env-api-keys.js";
-import { calculateCost, clampThinkingLevel } from "../models.js";
+import { getEnvApiKey } from "../env-api-keys.ts";
+import { calculateCost, clampThinkingLevel } from "../models.ts";
 import type {
 	Api,
 	AssistantMessage,
@@ -19,10 +19,10 @@ import type {
 	ThinkingContent,
 	ThinkingLevel,
 	ToolCall,
-} from "../types.js";
-import { AssistantMessageEventStream } from "../utils/event-stream.js";
-import { sanitizeSurrogates } from "../utils/sanitize-unicode.js";
-import type { GoogleThinkingLevel } from "./google-shared.js";
+} from "../types.ts";
+import { AssistantMessageEventStream } from "../utils/event-stream.ts";
+import { sanitizeSurrogates } from "../utils/sanitize-unicode.ts";
+import type { GoogleThinkingLevel } from "./google-shared.ts";
 import {
 	convertMessages,
 	convertTools,
@@ -30,8 +30,8 @@ import {
 	mapStopReason,
 	mapToolChoice,
 	retainThoughtSignature,
-} from "./google-shared.js";
-import { buildBaseOptions } from "./simple-options.js";
+} from "./google-shared.ts";
+import { buildBaseOptions } from "./simple-options.ts";

 export interface GoogleOptions extends StreamOptions {
 	toolChoice?: "auto" | "none" | "any";
--- a/packages/ai/src/providers/images/openrouter.ts
+++ b/packages/ai/src/providers/images/openrouter.ts
@@ -6,7 +6,7 @@ import type {
 	ChatCompletionContentPartText,
 	ChatCompletionCreateParamsNonStreaming,
 } from "openai/resources/chat/completions.js";
-import { getEnvApiKey } from "../../env-api-keys.js";
+import { getEnvApiKey } from "../../env-api-keys.ts";
 import type {
 	AssistantImages,
 	ImageContent,
@@ -15,9 +15,9 @@ import type {
 	ImagesModel,
 	ImagesOptions,
 	TextContent,
-} from "../../types.js";
-import { headersToRecord } from "../../utils/headers.js";
-import { sanitizeSurrogates } from "../../utils/sanitize-unicode.js";
+} from "../../types.ts";
+import { headersToRecord } from "../../utils/headers.ts";
+import { sanitizeSurrogates } from "../../utils/sanitize-unicode.ts";

 interface OpenRouterGeneratedImage {
 	image_url?: string | { url?: string };
--- a/packages/ai/src/providers/images/register-builtins.ts
+++ b/packages/ai/src/providers/images/register-builtins.ts
@@ -1,6 +1,6 @@
-import { registerImagesApiProvider } from "../../images-api-registry.js";
-import type { AssistantImages, ImagesContext, ImagesFunction, ImagesModel, ImagesOptions } from "../../types.js";
-import type { generateImagesOpenRouter as generateImagesOpenRouterFunction } from "./openrouter.js";
+import { registerImagesApiProvider } from "../../images-api-registry.ts";
+import type { AssistantImages, ImagesContext, ImagesFunction, ImagesModel, ImagesOptions } from "../../types.ts";
+import type { generateImagesOpenRouter as generateImagesOpenRouterFunction } from "./openrouter.ts";

 interface OpenRouterImagesProviderModule {
 	generateImagesOpenRouter: typeof generateImagesOpenRouterFunction;
@@ -21,7 +21,7 @@ function createLazyLoadErrorImages(model: ImagesModel<"openrouter-images">, erro
 }

 function loadOpenRouterImagesProviderModule(): Promise<OpenRouterImagesProviderModule> {
-	openRouterImagesProviderModulePromise ||= import("./openrouter.js").then(
+	openRouterImagesProviderModulePromise ||= import("./openrouter.ts").then(
 		(module) => module as OpenRouterImagesProviderModule,
 	);
 	return openRouterImagesProviderModulePromise;
--- a/packages/ai/src/providers/mistral.ts
+++ b/packages/ai/src/providers/mistral.ts
@@ -6,8 +6,8 @@ import type {
 	ContentChunk,
 	FunctionTool,
 } from "@mistralai/mistralai/models/components";
-import { getEnvApiKey } from "../env-api-keys.js";
-import { calculateCost, clampThinkingLevel } from "../models.js";
+import { getEnvApiKey } from "../env-api-keys.ts";
+import { calculateCost, clampThinkingLevel } from "../models.ts";
 import type {
 	AssistantMessage,
 	Context,
@@ -21,13 +21,13 @@ import type {
 	ThinkingContent,
 	Tool,
 	ToolCall,
-} from "../types.js";
-import { AssistantMessageEventStream } from "../utils/event-stream.js";
-import { shortHash } from "../utils/hash.js";
-import { parseStreamingJson } from "../utils/json-parse.js";
-import { sanitizeSurrogates } from "../utils/sanitize-unicode.js";
-import { buildBaseOptions } from "./simple-options.js";
-import { transformMessages } from "./transform-messages.js";
+} from "../types.ts";
+import { AssistantMessageEventStream } from "../utils/event-stream.ts";
+import { shortHash } from "../utils/hash.ts";
+import { parseStreamingJson } from "../utils/json-parse.ts";
+import { sanitizeSurrogates } from "../utils/sanitize-unicode.ts";
+import { buildBaseOptions } from "./simple-options.ts";
+import { transformMessages } from "./transform-messages.ts";

 const MISTRAL_TOOL_CALL_ID_LENGTH = 9;
 const MAX_MISTRAL_ERROR_BODY_CHARS = 4000;
--- a/packages/ai/src/providers/openai-codex-responses.ts
+++ b/packages/ai/src/providers/openai-codex-responses.ts
@@ -6,7 +6,7 @@ import type {
 	ResponseStreamEvent,
 } from "openai/resources/responses/responses.js";

-// NEVER convert to top-level runtime imports - breaks browser/Vite builds (web-ui)
+// NEVER convert to top-level runtime imports - breaks browser/Vite builds
 let _os: typeof NodeOs | null = null;

 type DynamicImport = (specifier: string) => Promise<unknown>;
@@ -20,9 +20,9 @@ if (typeof process !== "undefined" && (process.versions?.node || process.version
 	});
 }

-import { getEnvApiKey } from "../env-api-keys.js";
-import { clampThinkingLevel } from "../models.js";
-import { registerSessionResourceCleanup } from "../session-resources.js";
+import { getEnvApiKey } from "../env-api-keys.ts";
+import { clampThinkingLevel } from "../models.ts";
+import { registerSessionResourceCleanup } from "../session-resources.ts";
 import type {
 	Api,
 	AssistantMessage,
@@ -32,16 +32,17 @@ import type {
 	StreamFunction,
 	StreamOptions,
 	Usage,
-} from "../types.js";
+} from "../types.ts";
 import {
 	appendAssistantMessageDiagnostic,
 	createAssistantMessageDiagnostic,
 	formatThrownValue,
-} from "../utils/diagnostics.js";
-import { AssistantMessageEventStream } from "../utils/event-stream.js";
-import { headersToRecord } from "../utils/headers.js";
-import { convertResponsesMessages, convertResponsesTools, processResponsesStream } from "./openai-responses-shared.js";
-import { buildBaseOptions } from "./simple-options.js";
+} from "../utils/diagnostics.ts";
+import { AssistantMessageEventStream } from "../utils/event-stream.ts";
+import { headersToRecord } from "../utils/headers.ts";
+import { clampOpenAIPromptCacheKey } from "./openai-prompt-cache.ts";
+import { convertResponsesMessages, convertResponsesTools, processResponsesStream } from "./openai-responses-shared.ts";
+import { buildBaseOptions } from "./simple-options.ts";

 // ============================================================================
 // Configuration
@@ -254,7 +255,29 @@ export const streamOpenAICodexResponses: StreamFunction<"openai-codex-responses"

 					const errorText = await response.text();
 					if (attempt < MAX_RETRIES && isRetryableError(response.status, errorText)) {
-						const delayMs = BASE_DELAY_MS * 2 ** attempt;
+						let delayMs = BASE_DELAY_MS * 2 ** attempt;
+
+						const retryAfterMs = response.headers.get("retry-after-ms");
+						if (retryAfterMs !== null) {
+							const millis = Number(retryAfterMs);
+							if (Number.isFinite(millis)) {
+								delayMs = Math.max(0, millis);
+							}
+						} else {
+							const retryAfter = response.headers.get("retry-after");
+							if (retryAfter) {
+								const seconds = Number(retryAfter);
+								if (Number.isFinite(seconds)) {
+									delayMs = Math.max(0, seconds * 1000);
+								} else {
+									const date = Date.parse(retryAfter);
+									if (!Number.isNaN(date)) {
+										delayMs = Math.max(0, date - Date.now());
+									}
+								}
+							}
+						}
+
 						await sleep(delayMs, options?.signal);
 						continue;
 					}
@@ -356,7 +379,7 @@ function buildRequestBody(
 		input: messages,
 		text: { verbosity: options?.textVerbosity || "low" },
 		include: ["reasoning.encrypted_content"],
-		prompt_cache_key: options?.sessionId,
+		prompt_cache_key: clampOpenAIPromptCacheKey(options?.sessionId),
 		tool_choice: "auto",
 		parallel_tool_calls: true,
 	};
@@ -711,7 +734,35 @@ type WebSocketConstructor = new (
 	protocols?: string | string[] | { headers?: Record<string, string> },
 ) => WebSocketLike;

-function getWebSocketConstructor(): WebSocketConstructor | null {
+let _cachedWebsocket: WebSocketConstructor | null = null;
+async function getWebSocketConstructor(): Promise<WebSocketConstructor | null> {
+	if (_cachedWebsocket) return _cachedWebsocket;
+
+	// bun doesn't respect http proxy envs, ref: https://github.com/oven-sh/bun/issues/15489
+	// TODO: remove this when bun supports proxy envs in websocket.
+	if (
+		process?.versions?.bun &&
+		(process.env.HTTP_PROXY || process.env.HTTPS_PROXY || process.env.http_proxy || process.env.https_proxy)
+	) {
+		const m = await dynamicImport("proxy-from-env");
+		const getProxyForUrl = (m as { getProxyForUrl: (url: string | object | URL) => string }).getProxyForUrl;
+
+		_cachedWebsocket = class extends WebSocket {
+			constructor(url: string | URL, options?: string | string[] | Record<string, unknown>) {
+				let _opts: Record<string, unknown> = {};
+				if (Array.isArray(options) || typeof options === "string") {
+					_opts = { protocols: options };
+				} else {
+					_opts = { ...options };
+				}
+
+				const proxy = getProxyForUrl(url.toString().replace(/^wss:/, "https:").replace(/^ws:/, "http:"));
+				super(url, { ..._opts, ...(proxy ? { proxy } : {}) } as any);
+			}
+		};
+		return _cachedWebsocket;
+	}
+
 	const ctor = (globalThis as { WebSocket?: unknown }).WebSocket;
 	if (typeof ctor !== "function") return null;
 	return ctor as unknown as WebSocketConstructor;
@@ -760,7 +811,7 @@ function scheduleSessionWebSocketExpiry(sessionId: string, entry: CachedWebSocke
 }

 async function connectWebSocket(url: string, headers: Headers, signal?: AbortSignal): Promise<WebSocketLike> {
-	const WebSocketCtor = getWebSocketConstructor();
+	const WebSocketCtor = await getWebSocketConstructor();
 	if (!WebSocketCtor) {
 		throw new Error("WebSocket transport is not available in this runtime");
 	}
--- a/packages/ai/src/providers/openai-completions.ts
+++ b/packages/ai/src/providers/openai-completions.ts
@@ -10,8 +10,8 @@ import type {
 	ChatCompletionSystemMessageParam,
 	ChatCompletionToolMessageParam,
 } from "openai/resources/chat/completions.js";
-import { getEnvApiKey } from "../env-api-keys.js";
-import { calculateCost, clampThinkingLevel } from "../models.js";
+import { getEnvApiKey } from "../env-api-keys.ts";
+import { calculateCost, clampThinkingLevel } from "../models.ts";
 import type {
 	AssistantMessage,
 	CacheRetention,
@@ -29,15 +29,16 @@ import type {
 	Tool,
 	ToolCall,
 	ToolResultMessage,
-} from "../types.js";
-import { AssistantMessageEventStream } from "../utils/event-stream.js";
-import { headersToRecord } from "../utils/headers.js";
-import { parseStreamingJson } from "../utils/json-parse.js";
-import { sanitizeSurrogates } from "../utils/sanitize-unicode.js";
-import { isCloudflareProvider, resolveCloudflareBaseUrl } from "./cloudflare.js";
-import { buildCopilotDynamicHeaders, hasCopilotVisionInput } from "./github-copilot-headers.js";
-import { buildBaseOptions } from "./simple-options.js";
-import { transformMessages } from "./transform-messages.js";
+} from "../types.ts";
+import { AssistantMessageEventStream } from "../utils/event-stream.ts";
+import { headersToRecord } from "../utils/headers.ts";
+import { parseStreamingJson } from "../utils/json-parse.ts";
+import { sanitizeSurrogates } from "../utils/sanitize-unicode.ts";
+import { isCloudflareProvider, resolveCloudflareBaseUrl } from "./cloudflare.ts";
+import { buildCopilotDynamicHeaders, hasCopilotVisionInput } from "./github-copilot-headers.ts";
+import { clampOpenAIPromptCacheKey } from "./openai-prompt-cache.ts";
+import { buildBaseOptions } from "./simple-options.ts";
+import { transformMessages } from "./transform-messages.ts";

 /**
 * Check if conversation messages contain tool calls or tool results.
@@ -165,6 +166,7 @@ export const streamOpenAICompletions: StreamFunction<"openai-completions", OpenA

 			let textBlock: TextContent | null = null;
 			let thinkingBlock: ThinkingContent | null = null;
+			let hasFinishReason = false;
 			const toolCallBlocksByIndex = new Map<number, StreamingToolCallBlock>();
 			const toolCallBlocksById = new Map<string, StreamingToolCallBlock>();
 			const blocks = output.content as StreamingBlock[];
@@ -288,6 +290,7 @@ export const streamOpenAICompletions: StreamFunction<"openai-completions", OpenA
 					if (finishReasonResult.errorMessage) {
 						output.errorMessage = finishReasonResult.errorMessage;
 					}
+					hasFinishReason = true;
 				}

 				if (choice.delta) {
@@ -324,7 +327,11 @@ export const streamOpenAICompletions: StreamFunction<"openai-completions", OpenA
 					if (foundReasoningField) {
 						const delta = deltaFields[foundReasoningField];
 						if (typeof delta === "string" && delta.length > 0) {
-							const block = ensureThinkingBlock(foundReasoningField);
+							const thinkingSignature =
+								model.provider === "opencode-go" && foundReasoningField === "reasoning"
+									? "reasoning_content"
+									: foundReasoningField;
+							const block = ensureThinkingBlock(thinkingSignature);
 							block.thinking += delta;
 							stream.push({
 								type: "thinking_delta",
@@ -390,6 +397,9 @@ export const streamOpenAICompletions: StreamFunction<"openai-completions", OpenA
 			if (output.stopReason === "error") {
 				throw new Error(output.errorMessage || "Provider returned an error stop reason");
 			}
+			if (!hasFinishReason) {
+				throw new Error("Stream ended without finish_reason");
+			}

 			stream.push({ type: "done", reason: output.stopReason, message: output });
 			stream.end();
@@ -507,7 +517,7 @@ function buildParams(
 		prompt_cache_key:
 			(model.baseUrl.includes("api.openai.com") && cacheRetention !== "none") ||
 			(cacheRetention === "long" && compat.supportsLongCacheRetention)
-				? options?.sessionId
+				? clampOpenAIPromptCacheKey(options?.sessionId)
 				: undefined,
 		prompt_cache_retention: cacheRetention === "long" && compat.supportsLongCacheRetention ? "24h" : undefined,
 	};
@@ -839,7 +849,10 @@ export function convertMessages(
 					}

 					// Use the signature from the first thinking block if available (for llama.cpp server + gpt-oss)
-					const signature = nonEmptyThinkingBlocks[0].thinkingSignature;
+					let signature = nonEmptyThinkingBlocks[0].thinkingSignature;
+					if (model.provider === "opencode-go" && signature === "reasoning") {
+						signature = "reasoning_content";
+					}
 					if (signature && signature.length > 0) {
 						(assistantMsg as any)[signature] = nonEmptyThinkingBlocks.map((block) => block.thinking).join("\n");
 					}
@@ -997,17 +1010,17 @@ function parseChunkUsage(
 	model: Model<"openai-completions">,
 ): AssistantMessage["usage"] {
 	const promptTokens = rawUsage.prompt_tokens || 0;
-	const reportedCachedTokens = rawUsage.prompt_tokens_details?.cached_tokens ?? rawUsage.prompt_cache_hit_tokens ?? 0;
+	const cacheReadTokens = rawUsage.prompt_tokens_details?.cached_tokens ?? rawUsage.prompt_cache_hit_tokens ?? 0;
 	const cacheWriteTokens = rawUsage.prompt_tokens_details?.cache_write_tokens || 0;

-	// Normalize to pi-ai semantics:
-	// - cacheRead: hits from cache created by previous requests only
-	// - cacheWrite: tokens written to cache in this request
-	// Some OpenAI-compatible providers (observed on OpenRouter) report cached_tokens
-	// as (previous hits + current writes). In that case, remove cacheWrite from cacheRead.
-	const cacheReadTokens =
-		cacheWriteTokens > 0 ? Math.max(0, reportedCachedTokens - cacheWriteTokens) : reportedCachedTokens;
-
+	// Follow documented OpenAI/OpenRouter semantics: cached_tokens is cache-read
+	// tokens (hits). OpenAI does not document or emit cache_write_tokens, but
+	// OpenRouter-compatible providers can include it as a separate write count.
+	// OpenRouter's own provider/tests affirm the separate mapping:
+	// https://github.com/OpenRouterTeam/ai-sdk-provider/pull/409
+	// Do not subtract writes from cached_tokens, otherwise spec-compliant
+	// providers are under-reported. DS4 mirrors this contract too:
+	// https://github.com/antirez/ds4/pull/29
 	const input = Math.max(0, promptTokens - cacheReadTokens - cacheWriteTokens);
 	// OpenAI completion_tokens already includes reasoning_tokens.
 	const outputTokens = rawUsage.completion_tokens || 0;
--- a/packages/ai/src/providers/openai-prompt-cache.ts
+++ b/packages/ai/src/providers/openai-prompt-cache.ts
@@ -0,0 +1,8 @@
+export const OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH = 64;
+
+export function clampOpenAIPromptCacheKey(key: string | undefined): string | undefined {
+	if (key === undefined) return undefined;
+	const chars = Array.from(key);
+	if (chars.length <= OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH) return key;
+	return chars.slice(0, OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH).join("");
+}
--- a/packages/ai/src/providers/openai-responses-shared.ts
+++ b/packages/ai/src/providers/openai-responses-shared.ts
@@ -12,7 +12,7 @@ import type {
 	ResponseReasoningItem,
 	ResponseStreamEvent,
 } from "openai/resources/responses/responses.js";
-import { calculateCost } from "../models.js";
+import { calculateCost } from "../models.ts";
 import type {
 	Api,
 	AssistantMessage,
@@ -26,12 +26,12 @@ import type {
 	Tool,
 	ToolCall,
 	Usage,
-} from "../types.js";
-import type { AssistantMessageEventStream } from "../utils/event-stream.js";
-import { shortHash } from "../utils/hash.js";
-import { parseStreamingJson } from "../utils/json-parse.js";
-import { sanitizeSurrogates } from "../utils/sanitize-unicode.js";
-import { transformMessages } from "./transform-messages.js";
+} from "../types.ts";
+import type { AssistantMessageEventStream } from "../utils/event-stream.ts";
+import { shortHash } from "../utils/hash.ts";
+import { parseStreamingJson } from "../utils/json-parse.ts";
+import { sanitizeSurrogates } from "../utils/sanitize-unicode.ts";
+import { transformMessages } from "./transform-messages.ts";

 // =============================================================================
 // Utilities
--- a/packages/ai/src/providers/openai-responses.ts
+++ b/packages/ai/src/providers/openai-responses.ts
@@ -1,7 +1,7 @@
 import OpenAI from "openai";
 import type { ResponseCreateParamsStreaming } from "openai/resources/responses/responses.js";
-import { getEnvApiKey } from "../env-api-keys.js";
-import { clampThinkingLevel } from "../models.js";
+import { getEnvApiKey } from "../env-api-keys.ts";
+import { clampThinkingLevel } from "../models.ts";
 import type {
 	Api,
 	AssistantMessage,
@@ -13,13 +13,14 @@ import type {
 	StreamFunction,
 	StreamOptions,
 	Usage,
-} from "../types.js";
-import { AssistantMessageEventStream } from "../utils/event-stream.js";
-import { headersToRecord } from "../utils/headers.js";
-import { isCloudflareProvider, resolveCloudflareBaseUrl } from "./cloudflare.js";
-import { buildCopilotDynamicHeaders, hasCopilotVisionInput } from "./github-copilot-headers.js";
-import { convertResponsesMessages, convertResponsesTools, processResponsesStream } from "./openai-responses-shared.js";
-import { buildBaseOptions } from "./simple-options.js";
+} from "../types.ts";
+import { AssistantMessageEventStream } from "../utils/event-stream.ts";
+import { headersToRecord } from "../utils/headers.ts";
+import { isCloudflareProvider, resolveCloudflareBaseUrl } from "./cloudflare.ts";
+import { buildCopilotDynamicHeaders, hasCopilotVisionInput } from "./github-copilot-headers.ts";
+import { clampOpenAIPromptCacheKey } from "./openai-prompt-cache.ts";
+import { convertResponsesMessages, convertResponsesTools, processResponsesStream } from "./openai-responses-shared.ts";
+import { buildBaseOptions } from "./simple-options.ts";

 const OPENAI_TOOL_CALL_PROVIDERS = new Set(["openai", "openai-codex", "opencode"]);

@@ -51,6 +52,22 @@ function getPromptCacheRetention(
 	return cacheRetention === "long" && compat.supportsLongCacheRetention ? "24h" : undefined;
 }

+function formatOpenAIResponsesError(error: unknown): string {
+	if (error instanceof Error) {
+		const status = (error as Error & { status?: unknown }).status;
+		const statusCode = typeof status === "number" ? status : undefined;
+		if (statusCode !== undefined) {
+			return `OpenAI API error (${statusCode}): ${error.message}`;
+		}
+		return error.message;
+	}
+	try {
+		return JSON.stringify(error);
+	} catch {
+		return String(error);
+	}
+}
+
 // OpenAI Responses-specific options
 export interface OpenAIResponsesOptions extends StreamOptions {
 	reasoningEffort?: "minimal" | "low" | "medium" | "high" | "xhigh";
@@ -130,7 +147,7 @@ export const streamOpenAIResponses: StreamFunction<"openai-responses", OpenAIRes
 				delete (block as { partialJson?: string }).partialJson;
 			}
 			output.stopReason = options?.signal?.aborted ? "aborted" : "error";
-			output.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);
+			output.errorMessage = formatOpenAIResponsesError(error);
 			stream.push({ type: "error", reason: output.stopReason, error: output });
 			stream.end();
 		}
@@ -224,7 +241,7 @@ function buildParams(model: Model<"openai-responses">, context: Context, options
 		model: model.id,
 		input: messages,
 		stream: true,
-		prompt_cache_key: cacheRetention === "none" ? undefined : options?.sessionId,
+		prompt_cache_key: cacheRetention === "none" ? undefined : clampOpenAIPromptCacheKey(options?.sessionId),
 		prompt_cache_retention: getPromptCacheRetention(compat, cacheRetention),
 		store: false,
 	};
--- a/packages/ai/src/providers/register-builtins.ts
+++ b/packages/ai/src/providers/register-builtins.ts
@@ -1,4 +1,4 @@
-import { clearApiProviders, registerApiProvider } from "../api-registry.js";
+import { clearApiProviders, registerApiProvider } from "../api-registry.ts";
 import type {
 	Api,
 	AssistantMessage,
@@ -8,17 +8,17 @@ import type {
 	SimpleStreamOptions,
 	StreamFunction,
 	StreamOptions,
-} from "../types.js";
-import { AssistantMessageEventStream } from "../utils/event-stream.js";
-import type { BedrockOptions } from "./amazon-bedrock.js";
-import type { AnthropicOptions } from "./anthropic.js";
-import type { AzureOpenAIResponsesOptions } from "./azure-openai-responses.js";
-import type { GoogleOptions } from "./google.js";
-import type { GoogleVertexOptions } from "./google-vertex.js";
-import type { MistralOptions } from "./mistral.js";
-import type { OpenAICodexResponsesOptions } from "./openai-codex-responses.js";
-import type { OpenAICompletionsOptions } from "./openai-completions.js";
-import type { OpenAIResponsesOptions } from "./openai-responses.js";
+} from "../types.ts";
+import { AssistantMessageEventStream } from "../utils/event-stream.ts";
+import type { BedrockOptions } from "./amazon-bedrock.ts";
+import type { AnthropicOptions } from "./anthropic.ts";
+import type { AzureOpenAIResponsesOptions } from "./azure-openai-responses.ts";
+import type { GoogleOptions } from "./google.ts";
+import type { GoogleVertexOptions } from "./google-vertex.ts";
+import type { MistralOptions } from "./mistral.ts";
+import type { OpenAICodexResponsesOptions } from "./openai-codex-responses.ts";
+import type { OpenAICompletionsOptions } from "./openai-completions.ts";
+import type { OpenAIResponsesOptions } from "./openai-responses.ts";

 interface LazyProviderModule<
 	TApi extends Api,
@@ -86,7 +86,10 @@ interface BedrockProviderModule {
 	) => AsyncIterable<AssistantMessageEvent>;
 }

-const importNodeOnlyProvider = (specifier: string): Promise<unknown> => import(specifier);
+const importNodeOnlyProvider = (specifier: string): Promise<unknown> => {
+	const runtimeSpecifier = import.meta.url.endsWith(".js") ? specifier.replace(/\.ts$/, ".js") : specifier;
+	return import(runtimeSpecifier);
+};

 let anthropicProviderModulePromise:
 	| Promise<LazyProviderModule<"anthropic-messages", AnthropicOptions, SimpleStreamOptions>>
@@ -203,7 +206,7 @@ function createLazySimpleStream<
 function loadAnthropicProviderModule(): Promise<
 	LazyProviderModule<"anthropic-messages", AnthropicOptions, SimpleStreamOptions>
 > {
-	anthropicProviderModulePromise ||= import("./anthropic.js").then((module) => {
+	anthropicProviderModulePromise ||= import("./anthropic.ts").then((module) => {
 		const provider = module as AnthropicProviderModule;
 		return {
 			stream: provider.streamAnthropic,
@@ -216,7 +219,7 @@ function loadAnthropicProviderModule(): Promise<
 function loadAzureOpenAIResponsesProviderModule(): Promise<
 	LazyProviderModule<"azure-openai-responses", AzureOpenAIResponsesOptions, SimpleStreamOptions>
 > {
-	azureOpenAIResponsesProviderModulePromise ||= import("./azure-openai-responses.js").then((module) => {
+	azureOpenAIResponsesProviderModulePromise ||= import("./azure-openai-responses.ts").then((module) => {
 		const provider = module as AzureOpenAIResponsesProviderModule;
 		return {
 			stream: provider.streamAzureOpenAIResponses,
@@ -229,7 +232,7 @@ function loadAzureOpenAIResponsesProviderModule(): Promise<
 function loadGoogleProviderModule(): Promise<
 	LazyProviderModule<"google-generative-ai", GoogleOptions, SimpleStreamOptions>
 > {
-	googleProviderModulePromise ||= import("./google.js").then((module) => {
+	googleProviderModulePromise ||= import("./google.ts").then((module) => {
 		const provider = module as GoogleProviderModule;
 		return {
 			stream: provider.streamGoogle,
@@ -242,7 +245,7 @@ function loadGoogleProviderModule(): Promise<
 function loadGoogleVertexProviderModule(): Promise<
 	LazyProviderModule<"google-vertex", GoogleVertexOptions, SimpleStreamOptions>
 > {
-	googleVertexProviderModulePromise ||= import("./google-vertex.js").then((module) => {
+	googleVertexProviderModulePromise ||= import("./google-vertex.ts").then((module) => {
 		const provider = module as GoogleVertexProviderModule;
 		return {
 			stream: provider.streamGoogleVertex,
@@ -255,7 +258,7 @@ function loadGoogleVertexProviderModule(): Promise<
 function loadMistralProviderModule(): Promise<
 	LazyProviderModule<"mistral-conversations", MistralOptions, SimpleStreamOptions>
 > {
-	mistralProviderModulePromise ||= import("./mistral.js").then((module) => {
+	mistralProviderModulePromise ||= import("./mistral.ts").then((module) => {
 		const provider = module as MistralProviderModule;
 		return {
 			stream: provider.streamMistral,
@@ -268,7 +271,7 @@ function loadMistralProviderModule(): Promise<
 function loadOpenAICodexResponsesProviderModule(): Promise<
 	LazyProviderModule<"openai-codex-responses", OpenAICodexResponsesOptions, SimpleStreamOptions>
 > {
-	openAICodexResponsesProviderModulePromise ||= import("./openai-codex-responses.js").then((module) => {
+	openAICodexResponsesProviderModulePromise ||= import("./openai-codex-responses.ts").then((module) => {
 		const provider = module as OpenAICodexResponsesProviderModule;
 		return {
 			stream: provider.streamOpenAICodexResponses,
@@ -281,7 +284,7 @@ function loadOpenAICodexResponsesProviderModule(): Promise<
 function loadOpenAICompletionsProviderModule(): Promise<
 	LazyProviderModule<"openai-completions", OpenAICompletionsOptions, SimpleStreamOptions>
 > {
-	openAICompletionsProviderModulePromise ||= import("./openai-completions.js").then((module) => {
+	openAICompletionsProviderModulePromise ||= import("./openai-completions.ts").then((module) => {
 		const provider = module as OpenAICompletionsProviderModule;
 		return {
 			stream: provider.streamOpenAICompletions,
@@ -294,7 +297,7 @@ function loadOpenAICompletionsProviderModule(): Promise<
 function loadOpenAIResponsesProviderModule(): Promise<
 	LazyProviderModule<"openai-responses", OpenAIResponsesOptions, SimpleStreamOptions>
 > {
-	openAIResponsesProviderModulePromise ||= import("./openai-responses.js").then((module) => {
+	openAIResponsesProviderModulePromise ||= import("./openai-responses.ts").then((module) => {
 		const provider = module as OpenAIResponsesProviderModule;
 		return {
 			stream: provider.streamOpenAIResponses,
@@ -310,7 +313,7 @@ function loadBedrockProviderModule(): Promise<
 	if (bedrockProviderModuleOverride) {
 		return Promise.resolve(bedrockProviderModuleOverride);
 	}
-	bedrockProviderModulePromise ||= importNodeOnlyProvider("./amazon-bedrock.js").then((module) => {
+	bedrockProviderModulePromise ||= importNodeOnlyProvider("./amazon-bedrock.ts").then((module) => {
 		const provider = module as BedrockProviderModule;
 		return {
 			stream: provider.streamBedrock,
--- a/packages/ai/src/providers/simple-options.ts
+++ b/packages/ai/src/providers/simple-options.ts
@@ -1,9 +1,9 @@
-import type { Api, Model, SimpleStreamOptions, StreamOptions, ThinkingBudgets, ThinkingLevel } from "../types.js";
+import type { Api, Model, SimpleStreamOptions, StreamOptions, ThinkingBudgets, ThinkingLevel } from "../types.ts";

-export function buildBaseOptions(model: Model<Api>, options?: SimpleStreamOptions, apiKey?: string): StreamOptions {
+export function buildBaseOptions(_model: Model<Api>, options?: SimpleStreamOptions, apiKey?: string): StreamOptions {
 	return {
 		temperature: options?.temperature,
-		maxTokens: options?.maxTokens ?? (model.maxTokens > 0 ? Math.min(model.maxTokens, 32000) : undefined),
+		maxTokens: options?.maxTokens,
 		signal: options?.signal,
 		apiKey: apiKey || options?.apiKey,
 		transport: options?.transport,
@@ -24,7 +24,8 @@ export function clampReasoning(effort: ThinkingLevel | undefined): Exclude<Think
 }

 export function adjustMaxTokensForThinking(
-	baseMaxTokens: number,
+	// Undefined means no explicit caller cap. Use the model cap and fit thinking inside it.
+	baseMaxTokens: number | undefined,
 	modelMaxTokens: number,
 	reasoningLevel: ThinkingLevel,
 	customBudgets?: ThinkingBudgets,
@@ -40,7 +41,8 @@ export function adjustMaxTokensForThinking(
 	const minOutputTokens = 1024;
 	const level = clampReasoning(reasoningLevel)!;
 	let thinkingBudget = budgets[level]!;
-	const maxTokens = Math.min(baseMaxTokens + thinkingBudget, modelMaxTokens);
+	const maxTokens =
+		baseMaxTokens === undefined ? modelMaxTokens : Math.min(baseMaxTokens + thinkingBudget, modelMaxTokens);

 	if (maxTokens <= thinkingBudget) {
 		thinkingBudget = Math.max(0, maxTokens - minOutputTokens);
--- a/packages/ai/src/providers/transform-messages.ts
+++ b/packages/ai/src/providers/transform-messages.ts
@@ -7,7 +7,7 @@ import type {
 	TextContent,
 	ToolCall,
 	ToolResultMessage,
-} from "../types.js";
+} from "../types.ts";

 const NON_VISION_USER_IMAGE_PLACEHOLDER = "(image omitted: model does not support images)";
 const NON_VISION_TOOL_IMAGE_PLACEHOLDER = "(tool image omitted: model does not support images)";
--- a/packages/ai/src/stream.ts
+++ b/packages/ai/src/stream.ts
@@ -1,6 +1,6 @@
-import "./providers/register-builtins.js";
+import "./providers/register-builtins.ts";

-import { getApiProvider } from "./api-registry.js";
+import { getApiProvider } from "./api-registry.ts";
 import type {
 	Api,
 	AssistantMessage,
@@ -10,9 +10,9 @@ import type {
 	ProviderStreamOptions,
 	SimpleStreamOptions,
 	StreamOptions,
-} from "./types.js";
+} from "./types.ts";

-export { getEnvApiKey } from "./env-api-keys.js";
+export { getEnvApiKey } from "./env-api-keys.ts";

 function resolveApiProvider(api: Api) {
 	const provider = getApiProvider(api);
--- a/packages/ai/src/types.ts
+++ b/packages/ai/src/types.ts
@@ -1,7 +1,7 @@
-import type { AssistantMessageDiagnostic } from "./utils/diagnostics.js";
-import type { AssistantMessageEventStream } from "./utils/event-stream.js";
+import type { AssistantMessageDiagnostic } from "./utils/diagnostics.ts";
+import type { AssistantMessageEventStream } from "./utils/event-stream.ts";

-export type { AssistantMessageEventStream } from "./utils/event-stream.js";
+export type { AssistantMessageEventStream } from "./utils/event-stream.ts";

 export type KnownApi =
 	| "openai-completions"
@@ -419,6 +419,22 @@ export interface AnthropicMessagesCompat {
 	supportsEagerToolInputStreaming?: boolean;
 	/** Whether the provider supports Anthropic long cache retention (`cache_control.ttl: "1h"`). Default: true. */
 	supportsLongCacheRetention?: boolean;
+	/**
+	 * Whether to send the `x-session-affinity` header from `options.sessionId`
+	 * when caching is enabled. Required for providers like Fireworks that use
+	 * session affinity for prompt cache routing (requests to the same replica
+	 * maximize cache hits).
+	 * Default: false.
+	 */
+	sendSessionAffinityHeaders?: boolean;
+	/**
+	 * Whether the provider supports Anthropic-style `cache_control` markers on
+	 * tool definitions. When false, `cache_control` is omitted from tool params.
+	 * Some Anthropic-compatible providers (e.g., Fireworks) do not support this
+	 * field on tools and may reject or ignore it.
+	 * Default: true.
+	 */
+	supportsCacheControlOnTools?: boolean;
 }

 /**
--- a/packages/ai/src/utils/event-stream.ts
+++ b/packages/ai/src/utils/event-stream.ts
@@ -1,4 +1,4 @@
-import type { AssistantMessage, AssistantMessageEvent } from "../types.js";
+import type { AssistantMessage, AssistantMessageEvent } from "../types.ts";

 // Generic event stream class for async iteration
 export class EventStream<T, R = T> implements AsyncIterable<T> {
@@ -7,11 +7,12 @@ export class EventStream<T, R = T> implements AsyncIterable<T> {
 	private done = false;
 	private finalResultPromise: Promise<R>;
 	private resolveFinalResult!: (result: R) => void;
+	private isComplete: (event: T) => boolean;
+	private extractResult: (event: T) => R;

-	constructor(
-		private isComplete: (event: T) => boolean,
-		private extractResult: (event: T) => R,
-	) {
+	constructor(isComplete: (event: T) => boolean, extractResult: (event: T) => R) {
+		this.isComplete = isComplete;
+		this.extractResult = extractResult;
 		this.finalResultPromise = new Promise((resolve) => {
 			this.resolveFinalResult = resolve;
 		});
--- a/packages/ai/src/utils/node-http-proxy.ts
+++ b/packages/ai/src/utils/node-http-proxy.ts
@@ -0,0 +1,123 @@
+import type { Agent as HttpAgent } from "node:http";
+import type { Agent as HttpsAgent } from "node:https";
+import { HttpProxyAgent } from "http-proxy-agent";
+import { HttpsProxyAgent } from "https-proxy-agent";
+
+const DEFAULT_PROXY_PORTS: Record<string, number> = {
+	ftp: 21,
+	gopher: 70,
+	http: 80,
+	https: 443,
+	ws: 80,
+	wss: 443,
+};
+
+export interface NodeHttpProxyAgents {
+	httpAgent: HttpAgent;
+	httpsAgent: HttpsAgent;
+}
+
+export const UNSUPPORTED_PROXY_PROTOCOL_MESSAGE =
+	"Unsupported proxy protocol. SOCKS and PAC proxy URLs are not supported; use an HTTP or HTTPS proxy URL.";
+
+function getProxyEnv(key: string): string {
+	return process.env[key.toLowerCase()] || process.env[key.toUpperCase()] || "";
+}
+
+function parseProxyTargetUrl(targetUrl: string | URL): URL | undefined {
+	if (targetUrl instanceof URL) {
+		return targetUrl;
+	}
+
+	try {
+		return new URL(targetUrl);
+	} catch {
+		return undefined;
+	}
+}
+
+function shouldProxyHostname(hostname: string, port: number): boolean {
+	const noProxy = getProxyEnv("no_proxy").toLowerCase();
+	if (!noProxy) {
+		return true;
+	}
+	if (noProxy === "*") {
+		return false;
+	}
+
+	return noProxy.split(/[,\s]/).every((proxy) => {
+		if (!proxy) {
+			return true;
+		}
+
+		const parsedProxy = proxy.match(/^(.+):(\d+)$/);
+		let proxyHostname = parsedProxy ? parsedProxy[1] : proxy;
+		const proxyPort = parsedProxy ? Number.parseInt(parsedProxy[2]!, 10) : 0;
+		if (proxyPort && proxyPort !== port) {
+			return true;
+		}
+
+		if (!/^[.*]/.test(proxyHostname)) {
+			return hostname !== proxyHostname;
+		}
+
+		if (proxyHostname.startsWith("*")) {
+			proxyHostname = proxyHostname.slice(1);
+		}
+		return !hostname.endsWith(proxyHostname);
+	});
+}
+
+function getProxyForUrl(targetUrl: string | URL): string {
+	const parsedUrl = parseProxyTargetUrl(targetUrl);
+	if (!parsedUrl?.protocol || !parsedUrl.host) {
+		return "";
+	}
+
+	const protocol = parsedUrl.protocol.split(":", 1)[0]!;
+	const hostname = parsedUrl.host.replace(/:\d*$/, "");
+	const port = Number.parseInt(parsedUrl.port, 10) || DEFAULT_PROXY_PORTS[protocol] || 0;
+	if (!shouldProxyHostname(hostname, port)) {
+		return "";
+	}
+
+	let proxy = getProxyEnv(`${protocol}_proxy`) || getProxyEnv("all_proxy");
+	if (proxy && !proxy.includes("://")) {
+		proxy = `${protocol}://${proxy}`;
+	}
+	return proxy;
+}
+
+export function resolveHttpProxyUrlForTarget(targetUrl: string | URL): URL | undefined {
+	const proxy = getProxyForUrl(targetUrl);
+	if (!proxy) {
+		return undefined;
+	}
+
+	let proxyUrl: URL;
+	try {
+		proxyUrl = new URL(proxy);
+	} catch (error) {
+		throw new Error(
+			`Invalid proxy URL ${JSON.stringify(proxy)}: ${error instanceof Error ? error.message : String(error)}`,
+		);
+	}
+
+	if (proxyUrl.protocol !== "http:" && proxyUrl.protocol !== "https:") {
+		throw new Error(`${UNSUPPORTED_PROXY_PROTOCOL_MESSAGE} Got ${proxyUrl.protocol}`);
+	}
+
+	return proxyUrl;
+}
+
+export function createHttpProxyAgentsForTarget(targetUrl: string | URL): NodeHttpProxyAgents | undefined {
+	const proxyUrl = resolveHttpProxyUrlForTarget(targetUrl);
+	if (!proxyUrl) {
+		return undefined;
+	}
+
+	return {
+		httpAgent: new HttpProxyAgent(proxyUrl),
+		httpsAgent: new HttpsProxyAgent(proxyUrl) as unknown as HttpsAgent,
+	};
+}
--- a/packages/ai/src/utils/oauth/anthropic.ts
+++ b/packages/ai/src/utils/oauth/anthropic.ts
@@ -6,9 +6,9 @@
 */

 import type { Server } from "node:http";
-import { oauthErrorHtml, oauthSuccessHtml } from "./oauth-page.js";
-import { generatePKCE } from "./pkce.js";
-import type { OAuthCredentials, OAuthLoginCallbacks, OAuthPrompt, OAuthProviderInterface } from "./types.js";
+import { oauthErrorHtml, oauthSuccessHtml } from "./oauth-page.ts";
+import { generatePKCE } from "./pkce.ts";
+import type { OAuthCredentials, OAuthLoginCallbacks, OAuthPrompt, OAuthProviderInterface } from "./types.ts";

 type CallbackServerInfo = {
 	server: Server;
--- a/packages/ai/src/utils/oauth/github-copilot.ts
+++ b/packages/ai/src/utils/oauth/github-copilot.ts
@@ -2,9 +2,9 @@
 * GitHub Copilot OAuth flow
 */

-import { getModels } from "../../models.js";
-import type { Api, Model } from "../../types.js";
-import type { OAuthCredentials, OAuthLoginCallbacks, OAuthProviderInterface } from "./types.js";
+import { getModels } from "../../models.ts";
+import type { Api, Model } from "../../types.ts";
+import type { OAuthCredentials, OAuthLoginCallbacks, OAuthProviderInterface } from "./types.ts";

 type CopilotCredentials = OAuthCredentials & {
 	enterpriseUrl?: string;
--- a/packages/ai/src/utils/oauth/index.ts
+++ b/packages/ai/src/utils/oauth/index.ts
@@ -8,7 +8,7 @@
 */

 // Anthropic
-export { anthropicOAuthProvider, loginAnthropic, refreshAnthropicToken } from "./anthropic.js";
+export { anthropicOAuthProvider, loginAnthropic, refreshAnthropicToken } from "./anthropic.ts";
 // GitHub Copilot
 export {
 	getGitHubCopilotBaseUrl,
@@ -16,20 +16,20 @@ export {
 	loginGitHubCopilot,
 	normalizeDomain,
 	refreshGitHubCopilotToken,
-} from "./github-copilot.js";
+} from "./github-copilot.ts";
 // OpenAI Codex (ChatGPT OAuth)
-export { loginOpenAICodex, openaiCodexOAuthProvider, refreshOpenAICodexToken } from "./openai-codex.js";
+export { loginOpenAICodex, openaiCodexOAuthProvider, refreshOpenAICodexToken } from "./openai-codex.ts";

-export * from "./types.js";
+export * from "./types.ts";

 // ============================================================================
 // Provider Registry
 // ============================================================================

-import { anthropicOAuthProvider } from "./anthropic.js";
-import { githubCopilotOAuthProvider } from "./github-copilot.js";
-import { openaiCodexOAuthProvider } from "./openai-codex.js";
-import type { OAuthCredentials, OAuthProviderId, OAuthProviderInfo, OAuthProviderInterface } from "./types.js";
+import { anthropicOAuthProvider } from "./anthropic.ts";
+import { githubCopilotOAuthProvider } from "./github-copilot.ts";
+import { openaiCodexOAuthProvider } from "./openai-codex.ts";
+import type { OAuthCredentials, OAuthProviderId, OAuthProviderInfo, OAuthProviderInterface } from "./types.ts";

 const BUILT_IN_OAUTH_PROVIDERS: OAuthProviderInterface[] = [
 	anthropicOAuthProvider,
--- a/packages/ai/src/utils/oauth/openai-codex.ts
+++ b/packages/ai/src/utils/oauth/openai-codex.ts
@@ -5,7 +5,7 @@
 * It is only intended for CLI use, not browser environments.
 */

-// NEVER convert to top-level imports - breaks browser/Vite builds (web-ui)
+// NEVER convert to top-level imports - breaks browser/Vite builds
 let _randomBytes: typeof import("node:crypto").randomBytes | null = null;
 let _http: typeof import("node:http") | null = null;
 if (typeof process !== "undefined" && (process.versions?.node || process.versions?.bun)) {
@@ -17,9 +17,9 @@ if (typeof process !== "undefined" && (process.versions?.node || process.version
 	});
 }

-import { oauthErrorHtml, oauthSuccessHtml } from "./oauth-page.js";
-import { generatePKCE } from "./pkce.js";
-import type { OAuthCredentials, OAuthLoginCallbacks, OAuthPrompt, OAuthProviderInterface } from "./types.js";
+import { oauthErrorHtml, oauthSuccessHtml } from "./oauth-page.ts";
+import { generatePKCE } from "./pkce.ts";
+import type { OAuthCredentials, OAuthLoginCallbacks, OAuthPrompt, OAuthProviderInterface } from "./types.ts";

 const CALLBACK_HOST = process.env.PI_OAUTH_CALLBACK_HOST || "127.0.0.1";
 const CLIENT_ID = "app_EMoamEEZ73f0CkXaXp7hrann";
--- a/packages/ai/src/utils/oauth/types.ts
+++ b/packages/ai/src/utils/oauth/types.ts
@@ -1,4 +1,4 @@
-import type { Api, Model } from "../../types.js";
+import type { Api, Model } from "../../types.ts";

 export type OAuthCredentials = {
 	refresh: string;
--- a/packages/ai/src/utils/overflow.ts
+++ b/packages/ai/src/utils/overflow.ts
@@ -1,4 +1,4 @@
-import type { AssistantMessage } from "../types.js";
+import type { AssistantMessage } from "../types.ts";

 /**
 * Regex patterns to detect context overflow errors from different providers.
@@ -11,6 +11,7 @@ import type { AssistantMessage } from "../types.js";
 * - Anthropic: "prompt is too long: 213462 tokens > 200000 maximum"
 * - Anthropic: "413 {\"error\":{\"type\":\"request_too_large\",\"message\":\"Request exceeds the maximum size\"}}"
 * - OpenAI: "Your input exceeds the context window of this model"
+ * - OpenAI/LiteLLM: "Requested token count exceeds the model's maximum context length of 131072 tokens"
 * - Google: "The input token count (1196265) exceeds the maximum number of tokens allowed (1048575)"
 * - xAI: "This model's maximum prompt length is 131072 but the request contains 537812 tokens"
 * - Groq: "Please reduce the length of the messages or completion"
@@ -34,6 +35,7 @@ const OVERFLOW_PATTERNS = [
 	/request_too_large/i, // Anthropic request byte-size overflow (HTTP 413)
 	/input is too long for requested model/i, // Amazon Bedrock
 	/exceeds the context window/i, // OpenAI (Completions & Responses API)
+	/exceeds (?:the )?(?:model'?s )?maximum context length of [\d,]+ tokens?/i, // OpenAI-compatible proxies (LiteLLM)
 	/input token count.*exceeds the maximum/i, // Google (Gemini)
 	/maximum prompt length is \d+/i, // xAI (Grok)
 	/reduce the length of the messages/i, // Groq
@@ -81,7 +83,7 @@ const NON_OVERFLOW_PATTERNS = [
 *
 * **Reliable detection (returns error with detectable message):**
 * - Anthropic: "prompt is too long: X tokens > Y maximum" or "request_too_large"
- * - OpenAI (Completions & Responses): "exceeds the context window"
+ * - OpenAI (Completions & Responses): "exceeds the context window" or "exceeds the model's maximum context length of X tokens"
 * - Google Gemini: "input token count exceeds the maximum"
 * - xAI (Grok): "maximum prompt length is X but request contains Y"
 * - Groq: "reduce the length of the messages"
--- a/packages/ai/src/utils/validation.ts
+++ b/packages/ai/src/utils/validation.ts
@@ -1,7 +1,7 @@
 import { Compile } from "typebox/compile";
 import type { TLocalizedValidationError } from "typebox/error";
 import { Value } from "typebox/value";
-import type { Tool, ToolCall } from "../types.js";
+import type { Tool, ToolCall } from "../types.ts";

 const validatorCache = new WeakMap<object, ReturnType<typeof Compile>>();
 const TYPEBOX_KIND = Symbol.for("TypeBox.Kind");