@oh-my-pi/pi-ai 13.5.8 → 13.6.1

package/CHANGELOG.md

@@ -2,6 +2,18 @@
 
 ## [Unreleased]
 
+## [13.6.0] - 2026-03-03
+### Added
+
+- Added Anthropic Foundry gateway mode controlled by `CLAUDE_CODE_USE_FOUNDRY`, with support for `FOUNDRY_BASE_URL`, `ANTHROPIC_FOUNDRY_API_KEY`, `ANTHROPIC_CUSTOM_HEADERS`, and optional mTLS material (`CLAUDE_CODE_CLIENT_CERT`, `CLAUDE_CODE_CLIENT_KEY`, `NODE_EXTRA_CA_CERTS`)
+- Added LM Studio provider support with OpenAI-compatible model discovery and OAuth login.
+- Added support for `LM_STUDIO_API_KEY` and `LM_STUDIO_BASE_URL` environment variables for authentication and custom host configuration.
+
+### Changed
+
+- Anthropic key resolution now prefers `ANTHROPIC_FOUNDRY_API_KEY` over `ANTHROPIC_OAUTH_TOKEN` and `ANTHROPIC_API_KEY` when Foundry mode is enabled
+- Anthropic auth base-URL fallback now prefers `FOUNDRY_BASE_URL` when `CLAUDE_CODE_USE_FOUNDRY` is enabled
+
 ## [13.5.8] - 2026-03-02
 ### Fixed
 
@@ -1503,4 +1515,4 @@ _Dedicated to Peter's shoulder ([@steipete](https://twitter.com/steipete))_
 
 ## [0.9.4] - 2025-11-26
 
-Initial release with multi-provider LLM support.
+Initial release with multi-provider LLM support.

package/README.md

@@ -907,7 +907,7 @@ In Node.js environments, you can set environment variables to avoid passing API
 | Provider       | Environment Variable(s)                                                      |
 | -------------- | ---------------------------------------------------------------------------- |
 | OpenAI         | `OPENAI_API_KEY`                                                             |
-| Anthropic      | `ANTHROPIC_API_KEY` or `ANTHROPIC_OAUTH_TOKEN`                               |
+| Anthropic      | `ANTHROPIC_API_KEY` or `ANTHROPIC_OAUTH_TOKEN` (or `ANTHROPIC_FOUNDRY_API_KEY` when `CLAUDE_CODE_USE_FOUNDRY=true`) |
 | Google         | `GEMINI_API_KEY`                                                             |
 | Vertex AI      | `GOOGLE_CLOUD_PROJECT` (or `GCLOUD_PROJECT`) + `GOOGLE_CLOUD_LOCATION` + ADC |
 | Mistral        | `MISTRAL_API_KEY`                                                            |
@@ -936,6 +936,10 @@ In Node.js environments, you can set environment variables to avoid passing API
 For Cloudflare AI Gateway models, use provider base URL format
 `https://gateway.ai.cloudflare.com/v1/<account>/<gateway>/anthropic`.
 
+For Anthropic Foundry routing, set `CLAUDE_CODE_USE_FOUNDRY=true` plus:
+`FOUNDRY_BASE_URL`, `ANTHROPIC_FOUNDRY_API_KEY`, optional `ANTHROPIC_CUSTOM_HEADERS`,
+and optional mTLS material (`CLAUDE_CODE_CLIENT_CERT`, `CLAUDE_CODE_CLIENT_KEY`, `NODE_EXTRA_CA_CERTS`).
+
 Provider endpoint defaults for the current OpenAI-compatible integrations:
 
 - Together: `https://api.together.xyz/v1`

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-ai",
-	"version": "13.5.8",
+	"version": "13.6.1",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"homepage": "https://github.com/can1357/oh-my-pi",
 	"author": "Can Boluk",
@@ -38,10 +38,10 @@
 	},
 	"dependencies": {
 		"@anthropic-ai/sdk": "^0.78",
-		"@aws-sdk/client-bedrock-runtime": "^3.998",
+		"@aws-sdk/client-bedrock-runtime": "^3.1000",
 		"@bufbuild/protobuf": "^2.11",
 		"@google/genai": "^1.43",
-		"@oh-my-pi/pi-utils": "13.5.8",
+		"@oh-my-pi/pi-utils": "13.6.1",
 		"@sinclair/typebox": "^0.34",
 		"@smithy/node-http-handler": "^4.4",
 		"ajv": "^8.18",

package/src/auth-storage.ts

@@ -45,6 +45,7 @@ import { loginHuggingface } from "./utils/oauth/huggingface";
 import { loginKilo } from "./utils/oauth/kilo";
 import { loginKimi } from "./utils/oauth/kimi";
 import { loginLiteLLM } from "./utils/oauth/litellm";
+import { loginLmStudio } from "./utils/oauth/lm-studio";
 import { loginMiniMaxCode, loginMiniMaxCodeCn } from "./utils/oauth/minimax-code";
 import { loginMoonshot } from "./utils/oauth/moonshot";
 import { loginNanoGPT } from "./utils/oauth/nanogpt";
@@ -820,6 +821,11 @@ export class AuthStorage {
 				await saveApiKeyCredential(apiKey);
 				return;
 			}
+			case "lm-studio": {
+				const apiKey = await loginLmStudio(ctrl);
+				await saveApiKeyCredential(apiKey);
+				return;
+			}
 			case "ollama": {
 				const apiKey = await loginOllama(ctrl);
 				if (!apiKey) {

package/src/models.json

@@ -3246,7 +3246,8 @@
 				"Editor-Version": "vscode/1.107.0",
 				"Editor-Plugin-Version": "copilot-chat/0.35.0",
 				"Copilot-Integration-Id": "vscode-chat"
-			}
+			},
+			"premiumMultiplier": 0.33
 		},
 		"claude-opus-4.5": {
 			"id": "claude-opus-4.5",
@@ -3298,7 +3299,8 @@
 				"Editor-Version": "vscode/1.107.0",
 				"Editor-Plugin-Version": "copilot-chat/0.35.0",
 				"Copilot-Integration-Id": "vscode-chat"
-			}
+			},
+			"premiumMultiplier": 3
 		},
 		"claude-sonnet-4": {
 			"id": "claude-sonnet-4",
@@ -3562,7 +3564,8 @@
 				"supportsStore": false,
 				"supportsDeveloperRole": false,
 				"supportsReasoningEffort": false
-			}
+			},
+			"premiumMultiplier": 0
 		},
 		"gpt-5": {
 			"id": "gpt-5",
@@ -3772,6 +3775,33 @@
 				"Copilot-Integration-Id": "vscode-chat"
 			}
 		},
+		"gpt-5.3-codex": {
+			"id": "gpt-5.3-codex",
+			"name": "GPT-5.3 Codex",
+			"api": "openai-responses",
+			"provider": "github-copilot",
+			"premiumMultiplier": 1,
+			"baseUrl": "https://api.individual.githubcopilot.com",
+			"reasoning": true,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 0,
+				"output": 0,
+				"cacheRead": 0,
+				"cacheWrite": 0
+			},
+			"contextWindow": 272000,
+			"maxTokens": 128000,
+			"headers": {
+				"User-Agent": "GitHubCopilotChat/0.35.0",
+				"Editor-Version": "vscode/1.107.0",
+				"Editor-Plugin-Version": "copilot-chat/0.35.0",
+				"Copilot-Integration-Id": "vscode-chat"
+			}
+		},
 		"grok-code-fast-1": {
 			"id": "grok-code-fast-1",
 			"name": "Grok Code Fast 1",
@@ -3800,7 +3830,8 @@
 				"supportsStore": false,
 				"supportsDeveloperRole": false,
 				"supportsReasoningEffort": false
-			}
+			},
+			"premiumMultiplier": 0.25
 		}
 	},
 	"mistral": {
@@ -19694,7 +19725,7 @@
 				"cacheWrite": 0
 			},
 			"contextWindow": 163840,
-			"maxTokens": 163840
+			"maxTokens": 65536
 		},
 		"deepseek/deepseek-v3.2-exp": {
 			"id": "deepseek/deepseek-v3.2-exp",
@@ -19715,6 +19746,25 @@
 			"contextWindow": 163840,
 			"maxTokens": 65536
 		},
+		"essentialai/rnj-1-instruct": {
+			"id": "essentialai/rnj-1-instruct",
+			"name": "EssentialAI: Rnj 1 Instruct",
+			"api": "openai-completions",
+			"provider": "openrouter",
+			"baseUrl": "https://openrouter.ai/api/v1",
+			"reasoning": false,
+			"input": [
+				"text"
+			],
+			"cost": {
+				"input": 0.15,
+				"output": 0.15,
+				"cacheRead": 0,
+				"cacheWrite": 0
+			},
+			"contextWindow": 32768,
+			"maxTokens": 8888
+		},
 		"google/gemini-2.0-flash-001": {
 			"id": "google/gemini-2.0-flash-001",
 			"name": "Google: Gemini 2.0 Flash",
@@ -19813,7 +19863,7 @@
 				"cacheWrite": 0.08333333333333334
 			},
 			"contextWindow": 1048576,
-			"maxTokens": 65535
+			"maxTokens": 65536
 		},
 		"google/gemini-2.5-flash-preview-09-2025": {
 			"id": "google/gemini-2.5-flash-preview-09-2025",
@@ -19913,7 +19963,7 @@
 				"cacheWrite": 0.08333333333333334
 			},
 			"contextWindow": 1048576,
-			"maxTokens": 65535
+			"maxTokens": 65536
 		},
 		"google/gemini-3-pro-preview": {
 			"id": "google/gemini-3-pro-preview",
@@ -20089,7 +20139,7 @@
 				"cacheWrite": 0
 			},
 			"contextWindow": 131072,
-			"maxTokens": 32768
+			"maxTokens": 131072
 		},
 		"meta-llama/llama-3-8b-instruct": {
 			"id": "meta-llama/llama-3-8b-instruct",
@@ -20281,7 +20331,7 @@
 				"cacheWrite": 0
 			},
 			"contextWindow": 196608,
-			"maxTokens": 65536
+			"maxTokens": 196608
 		},
 		"minimax/minimax-m2.1": {
 			"id": "minimax/minimax-m2.1",
@@ -22939,12 +22989,12 @@
 			],
 			"cost": {
 				"input": 0.25,
-				"output": 2,
+				"output": 1,
 				"cacheRead": 0,
 				"cacheWrite": 0
 			},
 			"contextWindow": 262144,
-			"maxTokens": 65536
+			"maxTokens": 262144
 		},
 		"qwen/qwen3.5-397b-a17b": {
 			"id": "qwen/qwen3.5-397b-a17b",
@@ -23202,6 +23252,25 @@
 			"contextWindow": 163840,
 			"maxTokens": 65536
 		},
+		"upstage/solar-pro-3": {
+			"id": "upstage/solar-pro-3",
+			"name": "Upstage: Solar Pro 3",
+			"api": "openai-completions",
+			"provider": "openrouter",
+			"baseUrl": "https://openrouter.ai/api/v1",
+			"reasoning": true,
+			"input": [
+				"text"
+			],
+			"cost": {
+				"input": 0.15,
+				"output": 0.6,
+				"cacheRead": 0.015,
+				"cacheWrite": 0
+			},
+			"contextWindow": 128000,
+			"maxTokens": 8888
+		},
 		"upstage/solar-pro-3:free": {
 			"id": "upstage/solar-pro-3:free",
 			"name": "Upstage: Solar Pro 3 (free)",
@@ -23598,13 +23667,13 @@
 				"text"
 			],
 			"cost": {
-				"input": 0.95,
-				"output": 2.5500000000000003,
-				"cacheRead": 0.19999999999999998,
+				"input": 0.7999999999999999,
+				"output": 2.56,
+				"cacheRead": 0.16,
 				"cacheWrite": 0
 			},
-			"contextWindow": 204800,
-			"maxTokens": 131072
+			"contextWindow": 202752,
+			"maxTokens": 8888
 		}
 	},
 	"kilo": {
@@ -29499,6 +29568,25 @@
 			"contextWindow": 222222,
 			"maxTokens": 8888
 		},
+		"upstage/solar-pro-3": {
+			"id": "upstage/solar-pro-3",
+			"name": "Upstage: Solar Pro 3",
+			"api": "openai-completions",
+			"provider": "kilo",
+			"baseUrl": "https://api.kilo.ai/api/gateway",
+			"reasoning": false,
+			"input": [
+				"text"
+			],
+			"cost": {
+				"input": 0,
+				"output": 0,
+				"cacheRead": 0,
+				"cacheWrite": 0
+			},
+			"contextWindow": 222222,
+			"maxTokens": 8888
+		},
 		"writer/palmyra-x5": {
 			"id": "writer/palmyra-x5",
 			"name": "Writer: Palmyra X5",
@@ -36753,6 +36841,26 @@
 			},
 			"contextWindow": 1000000,
 			"maxTokens": 64000
+		},
+		"gemini-3.1-pro-preview": {
+			"id": "gemini-3.1-pro-preview",
+			"name": "Gemini 3.1 Pro Preview",
+			"api": "google-gemini-cli",
+			"provider": "google-gemini-cli",
+			"baseUrl": "https://cloudcode-pa.googleapis.com",
+			"reasoning": true,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 0,
+				"output": 0,
+				"cacheRead": 0,
+				"cacheWrite": 0
+			},
+			"contextWindow": 1048576,
+			"maxTokens": 65536
 		}
 	},
 	"google-vertex": {

package/src/provider-models/descriptors.ts

@@ -17,6 +17,7 @@ import {
 	kiloModelManagerOptions,
 	kimiCodeModelManagerOptions,
 	litellmModelManagerOptions,
+	lmStudioModelManagerOptions,
 	mistralModelManagerOptions,
 	moonshotModelManagerOptions,
 	nanoGptModelManagerOptions,
@@ -209,6 +210,13 @@ export const PROVIDER_DESCRIPTORS: readonly ProviderDescriptor[] = [
 		config => litellmModelManagerOptions(config),
 		catalog("LiteLLM", ["LITELLM_API_KEY"], { allowUnauthenticated: true }),
 	),
+	catalogDescriptor(
+		"lm-studio",
+		"llama-3-8b",
+		config => lmStudioModelManagerOptions(config),
+		catalog("LM Studio", ["LM_STUDIO_API_KEY"], { allowUnauthenticated: true }),
+		{ allowUnauthenticated: true },
+	),
 	catalogDescriptor(
 		"vllm",
 		"gpt-oss-20b",

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

@@ -7,6 +7,7 @@ import {
 	type OpenAICompatibleModelMapperContext,
 	type OpenAICompatibleModelRecord,
 } from "../utils/discovery/openai-compatible";
+import { getGitHubCopilotBaseUrl } from "../utils/oauth/github-copilot";
 
 const MODELS_DEV_URL = "https://models.dev/api.json";
 const ANTHROPIC_BASE_URL = "https://api.anthropic.com/v1";
@@ -796,6 +797,37 @@ export function kimiCodeModelManagerOptions(
 	};
 }
 
+// ---------------------------------------------------------------------------
+// 12.5. LM Studio
+// ---------------------------------------------------------------------------
+
+export interface LmStudioModelManagerConfig {
+	apiKey?: string;
+	baseUrl?: string;
+}
+
+export function lmStudioModelManagerOptions(
+	config?: LmStudioModelManagerConfig,
+): ModelManagerOptions<"openai-completions"> {
+	const apiKey = config?.apiKey;
+	const baseUrl = config?.baseUrl ?? Bun.env.LM_STUDIO_BASE_URL ?? "http://127.0.0.1:1234/v1";
+	const references = createBundledReferenceMap<"openai-completions">("lm-studio" as any);
+	return {
+		providerId: "lm-studio",
+		fetchDynamicModels: () =>
+			fetchOpenAICompatibleModels({
+				api: "openai-completions",
+				provider: "lm-studio",
+				baseUrl,
+				apiKey,
+				mapModel: (entry, defaults) => {
+					const reference = references.get(defaults.id);
+					return mapWithBundledReference(entry, defaults, reference);
+				},
+			}),
+	};
+}
+
 // ---------------------------------------------------------------------------
 // 13. Synthetic
 // ---------------------------------------------------------------------------
@@ -1255,7 +1287,11 @@ function extractCopilotLimits(entry: OpenAICompatibleModelRecord): {
 
 export function githubCopilotModelManagerOptions(config?: GithubCopilotModelManagerConfig): ModelManagerOptions<Api> {
 	const apiKey = config?.apiKey;
-	const baseUrl = config?.baseUrl ?? "https://api.individual.githubcopilot.com";
+	const configuredBaseUrl = config?.baseUrl ?? "https://api.individual.githubcopilot.com";
+	const baseUrl =
+		apiKey?.includes("proxy-ep=") && configuredBaseUrl.includes("githubcopilot.com")
+			? getGitHubCopilotBaseUrl(apiKey)
+			: configuredBaseUrl;
 	const references = createBundledReferenceMap<Api>("github-copilot");
 	const globalReferences = createGlobalReferenceMap();
 	return {

package/src/providers/anthropic.ts

@@ -1,4 +1,5 @@
 import * as nodeCrypto from "node:crypto";
+import * as fs from "node:fs";
 import * as tls from "node:tls";
 import Anthropic, { type ClientOptions as AnthropicSdkClientOptions } from "@anthropic-ai/sdk";
 import type {
@@ -6,7 +7,7 @@ import type {
 	MessageCreateParamsStreaming,
 	MessageParam,
 } from "@anthropic-ai/sdk/resources/messages";
-import { abortableSleep } from "@oh-my-pi/pi-utils";
+import { $env, abortableSleep, isEnoent } from "@oh-my-pi/pi-utils";
 import { calculateCost } from "../models";
 import { getEnvApiKey, OUTPUT_FALLBACK_BUFFER } from "../stream";
 import type {
@@ -33,8 +34,8 @@ import { finalizeErrorMessage, type RawHttpRequestDump } from "../utils/http-ins
 import { parseStreamingJson } from "../utils/json-parse";
 import {
 	buildCopilotDynamicHeaders,
-	getCopilotInitiatorOverride,
 	hasCopilotVisionInput,
+	resolveGitHubCopilotBaseUrl,
 } from "./github-copilot-headers";
 import { transformMessages } from "./transform-messages";
 
@@ -381,26 +382,135 @@ export type AnthropicClientOptionsResult = {
 
 const CLAUDE_CODE_TLS_CIPHERS = tls.DEFAULT_CIPHERS;
 
+type FoundryTlsOptions = {
+	ca?: string | string[];
+	cert?: string;
+	key?: string;
+};
+
+function isFoundryEnabled(): boolean {
+	const value = $env.CLAUDE_CODE_USE_FOUNDRY;
+	if (!value) return false;
+	const normalized = value.trim().toLowerCase();
+	return normalized === "1" || normalized === "true" || normalized === "yes" || normalized === "on";
+}
+
+function normalizeBaseUrl(baseUrl: string | undefined): string | undefined {
+	const trimmed = baseUrl?.trim();
+	return trimmed ? trimmed.replace(/\/+$/, "") : undefined;
+}
+
+function resolveAnthropicBaseUrl(model: Model<"anthropic-messages">, apiKey?: string): string | undefined {
+	if (model.provider === "github-copilot") {
+		return normalizeBaseUrl(resolveGitHubCopilotBaseUrl(model.baseUrl, apiKey) ?? model.baseUrl);
+	}
+	if (model.provider === "anthropic" && isFoundryEnabled()) {
+		const foundryBaseUrl = normalizeBaseUrl($env.FOUNDRY_BASE_URL);
+		if (foundryBaseUrl) {
+			return foundryBaseUrl;
+		}
+	}
+	if (model.provider === "anthropic") {
+		return normalizeBaseUrl(model.baseUrl) ?? "https://api.anthropic.com";
+	}
+	return normalizeBaseUrl(model.baseUrl);
+}
+
+function parseAnthropicCustomHeaders(rawHeaders: string | undefined): Record<string, string> | undefined {
+	const source = rawHeaders?.trim();
+	if (!source) return undefined;
+
+	const parsed: Record<string, string> = {};
+	for (const token of source.split(/\r?\n|,/)) {
+		const entry = token.trim();
+		if (!entry) continue;
+		const separatorIndex = entry.indexOf(":");
+		if (separatorIndex <= 0) continue;
+		const key = entry.slice(0, separatorIndex).trim();
+		const value = entry.slice(separatorIndex + 1).trim();
+		if (!key || !value) continue;
+		parsed[key] = value;
+	}
+
+	return Object.keys(parsed).length > 0 ? parsed : undefined;
+}
+
+function resolveAnthropicCustomHeaders(model: Model<"anthropic-messages">): Record<string, string> | undefined {
+	if (model.provider !== "anthropic") return undefined;
+	if (!isFoundryEnabled()) return undefined;
+	return parseAnthropicCustomHeaders($env.ANTHROPIC_CUSTOM_HEADERS);
+}
+
+function looksLikeFilePath(value: string): boolean {
+	return value.includes("/") || value.includes("\\") || /\.(pem|crt|cer|key)$/i.test(value);
+}
+
+function resolvePemValue(value: string | undefined, name: string): string | undefined {
+	const trimmed = value?.trim();
+	if (!trimmed) return undefined;
+
+	const inline = trimmed.replace(/\\n/g, "\n");
+	if (inline.includes("-----BEGIN")) {
+		return inline;
+	}
+
+	if (looksLikeFilePath(trimmed)) {
+		try {
+			return fs.readFileSync(trimmed, "utf8");
+		} catch (error) {
+			if (isEnoent(error)) {
+				throw new Error(`${name} path does not exist: ${trimmed}`);
+			}
+			throw error;
+		}
+	}
+
+	return inline;
+}
+
+function resolveFoundryTlsOptions(model: Model<"anthropic-messages">): FoundryTlsOptions | undefined {
+	if (model.provider !== "anthropic") return undefined;
+	if (!isFoundryEnabled()) return undefined;
+
+	const ca = resolvePemValue($env.NODE_EXTRA_CA_CERTS, "NODE_EXTRA_CA_CERTS");
+	const cert = resolvePemValue($env.CLAUDE_CODE_CLIENT_CERT, "CLAUDE_CODE_CLIENT_CERT");
+	const key = resolvePemValue($env.CLAUDE_CODE_CLIENT_KEY, "CLAUDE_CODE_CLIENT_KEY");
+
+	if ((cert && !key) || (!cert && key)) {
+		throw new Error("Both CLAUDE_CODE_CLIENT_CERT and CLAUDE_CODE_CLIENT_KEY must be set for mTLS.");
+	}
+
+	const options: FoundryTlsOptions = {};
+	if (ca) options.ca = [...tls.rootCertificates, ca];
+	if (cert) options.cert = cert;
+	if (key) options.key = key;
+	return Object.keys(options).length > 0 ? options : undefined;
+}
+
 function buildClaudeCodeTlsFetchOptions(
 	model: Model<"anthropic-messages">,
+	baseUrl: string | undefined,
 ): AnthropicSdkClientOptions["fetchOptions"] | undefined {
 	if (model.provider !== "anthropic") return undefined;
-	if (!model.baseUrl) return undefined;
+	if (!baseUrl) return undefined;
 
 	let serverName: string;
 	try {
-		serverName = new URL(model.baseUrl).hostname;
+		serverName = new URL(baseUrl).hostname;
 	} catch {
 		return undefined;
 	}
 
 	if (!serverName) return undefined;
 
+	const foundryTlsOptions = resolveFoundryTlsOptions(model);
+
 	return {
 		tls: {
 			rejectUnauthorized: true,
 			serverName,
 			...(CLAUDE_CODE_TLS_CIPHERS ? { ciphers: CLAUDE_CODE_TLS_CIPHERS } : {}),
+			...(foundryTlsOptions ?? {}),
 		},
 	};
 }
@@ -450,6 +560,15 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 		const startTime = Date.now();
 		let firstTokenTime: number | undefined;
 
+		const copilotDynamicHeaders =
+			model.provider === "github-copilot"
+				? buildCopilotDynamicHeaders({
+						messages: context.messages,
+						hasImages: hasCopilotVisionInput(context.messages),
+						premiumMultiplier: model.premiumMultiplier,
+						headers: { ...(model.headers ?? {}), ...(options?.headers ?? {}) },
+					})
+				: undefined;
 		const output: AssistantMessage = {
 			role: "assistant",
 			content: [],
@@ -471,20 +590,7 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 
 		try {
 			const apiKey = options?.apiKey ?? getEnvApiKey(model.provider) ?? "";
-
-			let copilotDynamicHeaders: Record<string, string> | undefined;
-			if (model.provider === "github-copilot") {
-				const hasImages = hasCopilotVisionInput(context.messages);
-				const initiatorOverride = getCopilotInitiatorOverride({
-					...(model.headers ?? {}),
-					...(options?.headers ?? {}),
-				});
-				copilotDynamicHeaders = buildCopilotDynamicHeaders({
-					messages: context.messages,
-					hasImages,
-					initiatorOverride,
-				});
-			}
+			const baseUrl = resolveAnthropicBaseUrl(model, apiKey) ?? "https://api.anthropic.com";
 
 			const { client, isOAuthToken } = createClient(model, {
 				model,
@@ -493,17 +599,17 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 				stream: true,
 				interleavedThinking: options?.interleavedThinking ?? true,
 				headers: options?.headers,
-				dynamicHeaders: copilotDynamicHeaders,
+				dynamicHeaders: copilotDynamicHeaders?.headers,
 				isOAuth: options?.isOAuth,
 			});
-			const params = buildParams(model, context, isOAuthToken, options);
+			const params = buildParams(model, baseUrl, context, isOAuthToken, options);
 			options?.onPayload?.(params);
 			rawRequestDump = {
 				provider: model.provider,
 				api: output.api,
 				model: model.id,
 				method: "POST",
-				url: `${model.baseUrl ?? "https://api.anthropic.com"}/v1/messages`,
+				url: `${baseUrl}/v1/messages`,
 				body: params,
 			};
 
@@ -517,6 +623,9 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 			let started = false;
 			do {
 				const anthropicStream = client.messages.stream({ ...params, stream: true }, { signal: options?.signal });
+				if (copilotDynamicHeaders && output.usage.premiumRequests === undefined) {
+					output.usage.premiumRequests = copilotDynamicHeaders.premiumRequests;
+				}
 
 				try {
 					for await (const event of anthropicStream) {
@@ -830,8 +939,9 @@ export function buildAnthropicClientOptions(args: AnthropicClientOptionsArgs): A
 		isOAuth,
 	} = args;
 	const oauthToken = isOAuth ?? isAnthropicOAuthToken(apiKey);
-
-	const tlsFetchOptions = buildClaudeCodeTlsFetchOptions(model);
+	const baseUrl = resolveAnthropicBaseUrl(model, apiKey);
+	const foundryCustomHeaders = resolveAnthropicCustomHeaders(model);
+	const tlsFetchOptions = buildClaudeCodeTlsFetchOptions(model, baseUrl);
 	if (model.provider === "github-copilot") {
 		const betaFeatures = [...extraBetas];
 		if (interleavedThinking) {
@@ -853,7 +963,7 @@ export function buildAnthropicClientOptions(args: AnthropicClientOptionsArgs): A
 			isOAuthToken: false,
 			apiKey: null,
 			authToken: apiKey,
-			baseURL: model.baseUrl,
+			baseURL: baseUrl,
 			maxRetries: 5,
 			dangerouslyAllowBrowser: true,
 			defaultHeaders,
@@ -868,18 +978,18 @@ export function buildAnthropicClientOptions(args: AnthropicClientOptionsArgs): A
 
 	const defaultHeaders = buildAnthropicHeaders({
 		apiKey,
-		baseUrl: model.baseUrl,
+		baseUrl,
 		isOAuth: oauthToken,
 		extraBetas: betaFeatures,
 		stream,
-		modelHeaders: mergeHeaders(model.headers, headers, dynamicHeaders),
+		modelHeaders: mergeHeaders(model.headers, foundryCustomHeaders, headers, dynamicHeaders),
 	});
 
 	return {
 		isOAuthToken: oauthToken,
 		apiKey: oauthToken ? null : apiKey,
 		authToken: oauthToken ? apiKey : undefined,
-		baseURL: model.baseUrl,
+		baseURL: baseUrl,
 		maxRetries: 5,
 		dangerouslyAllowBrowser: true,
 		defaultHeaders,
@@ -1152,11 +1262,12 @@ function enforceCacheControlLimit(params: MessageCreateParamsStreaming, maxBreak
 }
 function buildParams(
 	model: Model<"anthropic-messages">,
+	baseUrl: string,
 	context: Context,
 	isOAuthToken: boolean,
 	options?: AnthropicOptions,
 ): MessageCreateParamsStreaming {
-	const { cacheControl } = getCacheControl(model.baseUrl, options?.cacheRetention);
+	const { cacheControl } = getCacheControl(baseUrl, options?.cacheRetention);
 	const params: AnthropicSamplingParams = {
 		model: model.id,
 		messages: convertAnthropicMessages(context.messages, model, isOAuthToken),

package/src/providers/github-copilot-headers.ts

@@ -1,10 +1,25 @@
 import type { Message } from "../types";
-
+import { getGitHubCopilotBaseUrl } from "../utils/oauth/github-copilot";
 /**
  * Infer whether the current request to Copilot is user-initiated or agent-initiated.
  * Accepts `unknown[]` because providers may pass pre-converted message shapes.
  */
-export function inferCopilotInitiator(messages: unknown[]): "user" | "agent" {
+export type CopilotInitiator = "user" | "agent";
+export type CopilotPremiumRequests = number;
+export type CopilotDynamicHeaders = {
+	headers: Record<string, string>;
+	initiator: CopilotInitiator;
+	premiumRequests: CopilotPremiumRequests;
+};
+export function resolveGitHubCopilotBaseUrl(
+	baseUrl: string | undefined,
+	apiKey: string | undefined,
+): string | undefined {
+	if (!apiKey?.includes("proxy-ep=")) return baseUrl;
+	if (baseUrl && !baseUrl.includes("githubcopilot.com")) return baseUrl;
+	return getGitHubCopilotBaseUrl(apiKey);
+}
+export function inferCopilotInitiator(messages: unknown[]): CopilotInitiator {
 	if (messages.length === 0) return "user";
 
 	const last = messages[messages.length - 1] as Record<string, unknown>;
@@ -50,10 +65,10 @@ export function hasCopilotVisionInput(messages: Message[]): boolean {
  * Resolve an explicitly configured Copilot initiator header, if present.
  * Handles case-insensitive X-Initiator keys and returns the last valid value.
  */
-export function getCopilotInitiatorOverride(headers: Record<string, string> | undefined): "user" | "agent" | undefined {
+export function getCopilotInitiatorOverride(headers: Record<string, string> | undefined): CopilotInitiator | undefined {
 	if (!headers) return undefined;
 
-	let override: "user" | "agent" | undefined;
+	let override: CopilotInitiator | undefined;
 	for (const [key, value] of Object.entries(headers)) {
 		if (key.toLowerCase() !== "x-initiator") continue;
 		const normalized = value.trim().toLowerCase();
@@ -64,6 +79,30 @@ export function getCopilotInitiatorOverride(headers: Record<string, string> | un
 
 	return override;
 }
+
+export type CopilotPlanTier = "free" | "paid";
+
+function normalizeCopilotPlanTier(planTier: string | undefined): CopilotPlanTier {
+	if (planTier === "paid") return "paid";
+	return "free";
+}
+export function getCopilotPremiumMultiplier(premiumMultiplier: number | undefined, planTier?: string): number {
+	const normalizedMultiplier = premiumMultiplier ?? 1;
+	if (normalizeCopilotPlanTier(planTier) === "free" && normalizedMultiplier === 0) {
+		return 1;
+	}
+	return normalizedMultiplier;
+}
+
+export function getCopilotPremiumRequests(params: {
+	initiator: CopilotInitiator;
+	premiumMultiplier?: number;
+	planTier?: string;
+}): CopilotPremiumRequests {
+	if (params.initiator === "agent") return 0;
+	return getCopilotPremiumMultiplier(params.premiumMultiplier, params.planTier);
+}
+
 /**
  * Build dynamic Copilot headers that vary per-request.
  * Static headers (User-Agent, Editor-Version, etc.) come from model.headers.
@@ -71,10 +110,15 @@ export function getCopilotInitiatorOverride(headers: Record<string, string> | un
 export function buildCopilotDynamicHeaders(params: {
 	messages: unknown[];
 	hasImages: boolean;
-	initiatorOverride?: "user" | "agent";
-}): Record<string, string> {
+	premiumMultiplier?: number;
+	headers?: Record<string, string>;
+	initiatorOverride?: CopilotInitiator;
+	planTier?: string;
+}): CopilotDynamicHeaders {
+	const initiator =
+		params.initiatorOverride ?? getCopilotInitiatorOverride(params.headers) ?? inferCopilotInitiator(params.messages);
 	const headers: Record<string, string> = {
-		"X-Initiator": params.initiatorOverride ?? inferCopilotInitiator(params.messages),
+		"X-Initiator": initiator,
 		"Openai-Intent": "conversation-edits",
 	};
 
@@ -82,5 +126,13 @@ export function buildCopilotDynamicHeaders(params: {
 		headers["Copilot-Vision-Request"] = "true";
 	}
 
-	return headers;
+	return {
+		headers,
+		initiator,
+		premiumRequests: getCopilotPremiumRequests({
+			initiator,
+			premiumMultiplier: params.premiumMultiplier,
+			planTier: params.planTier,
+		}),
+	};
 }

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

@@ -24,7 +24,7 @@ const GEMINI_TIER_MAP: Array<{ tier: string; models: string[] }> = [
 	},
 	{
 		tier: "Pro",
-		models: ["gemini-2.5-pro", "gemini-3-pro-preview", "gemini-3-pro", "gemini-1.5-pro"],
+		models: ["gemini-2.5-pro", "gemini-3-pro-preview", "gemini-3.1-pro-preview", "gemini-3-pro", "gemini-1.5-pro"],
 	},
 ];
 

package/src/providers/openai-completions.ts

@@ -35,8 +35,8 @@ import { adaptSchemaForStrict, NO_STRICT } from "../utils/schema";
 import { mapToOpenAICompletionsToolChoice } from "../utils/tool-choice";
 import {
 	buildCopilotDynamicHeaders,
-	getCopilotInitiatorOverride,
 	hasCopilotVisionInput,
+	resolveGitHubCopilotBaseUrl,
 } from "./github-copilot-headers";
 import { transformMessages } from "./transform-messages";
 
@@ -188,7 +188,12 @@ export const streamOpenAICompletions: StreamFunction<"openai-completions"> = (
 
 		try {
 			const apiKey = options?.apiKey || getEnvApiKey(model.provider) || "";
-			const client = await createClient(model, context, apiKey, options?.headers);
+			const { client, copilotPremiumRequests, baseUrl } = await createClient(
+				model,
+				context,
+				apiKey,
+				options?.headers,
+			);
 			const params = buildParams(model, context, options);
 			options?.onPayload?.(params);
 			rawRequestDump = {
@@ -196,10 +201,11 @@ export const streamOpenAICompletions: StreamFunction<"openai-completions"> = (
 				api: output.api,
 				model: model.id,
 				method: "POST",
-				url: `${model.baseUrl ?? "https://api.openai.com/v1"}/chat/completions`,
+				url: `${baseUrl ?? "https://api.openai.com/v1"}/chat/completions`,
 				body: params,
 			};
 			const openaiStream = await client.chat.completions.create(params, { signal: options?.signal });
+			if (copilotPremiumRequests !== undefined) output.usage.premiumRequests = copilotPremiumRequests;
 			stream.push({ type: "start", partial: output });
 
 			let currentBlock: TextContent | ThinkingContent | (ToolCall & { partialArgs?: string }) | null = null;
@@ -340,6 +346,7 @@ export const streamOpenAICompletions: StreamFunction<"openai-completions"> = (
 						// Compute totalTokens ourselves since we add reasoning_tokens to output
 						// and some providers (e.g., Groq) don't include them in total_tokens
 						totalTokens: input + outputTokens + cachedTokens,
+						...(copilotPremiumRequests !== undefined ? { premiumRequests: copilotPremiumRequests } : {}),
 						cost: {
 							input: 0,
 							output: 0,
@@ -510,23 +517,32 @@ async function createClient(
 	if (model.provider === "kimi-code") {
 		headers = { ...(await getKimiCommonHeaders()), ...headers };
 	}
+	let copilotPremiumRequests: number | undefined;
+
+	let baseUrl = model.baseUrl;
 	if (model.provider === "github-copilot") {
 		const hasImages = hasCopilotVisionInput(context.messages);
-		const copilotHeaders = buildCopilotDynamicHeaders({
+		const copilot = buildCopilotDynamicHeaders({
 			messages: context.messages,
 			hasImages,
-			initiatorOverride: getCopilotInitiatorOverride(headers),
+			premiumMultiplier: model.premiumMultiplier,
+			headers,
 		});
-		Object.assign(headers, copilotHeaders);
+		Object.assign(headers, copilot.headers);
+		copilotPremiumRequests = copilot.premiumRequests;
+		baseUrl = resolveGitHubCopilotBaseUrl(model.baseUrl, apiKey) ?? model.baseUrl;
 	}
-
-	return new OpenAI({
-		apiKey,
-		baseURL: model.baseUrl,
-		dangerouslyAllowBrowser: true,
-		maxRetries: 5,
-		defaultHeaders: headers,
-	});
+	return {
+		client: new OpenAI({
+			apiKey,
+			baseURL: baseUrl,
+			dangerouslyAllowBrowser: true,
+			maxRetries: 5,
+			defaultHeaders: headers,
+		}),
+		copilotPremiumRequests,
+		baseUrl,
+	};
 }
 
 function buildParams(model: Model<"openai-completions">, context: Context, options?: OpenAICompletionsOptions) {

package/src/providers/openai-responses.ts

@@ -36,8 +36,8 @@ import { adaptSchemaForStrict, NO_STRICT } from "../utils/schema";
 import { mapToOpenAIResponsesToolChoice } from "../utils/tool-choice";
 import {
 	buildCopilotDynamicHeaders,
-	getCopilotInitiatorOverride,
 	hasCopilotVisionInput,
+	resolveGitHubCopilotBaseUrl,
 } from "./github-copilot-headers";
 import { transformMessages } from "./transform-messages";
 
@@ -113,7 +113,7 @@ export const streamOpenAIResponses: StreamFunction<"openai-responses"> = (
 		try {
 			// Create OpenAI client
 			const apiKey = options?.apiKey || getEnvApiKey(model.provider) || "";
-			const client = createClient(model, context, apiKey, options?.headers);
+			const { client, copilotPremiumRequests, baseUrl } = createClient(model, context, apiKey, options?.headers);
 			const params = buildParams(model, context, options);
 			options?.onPayload?.(params);
 			rawRequestDump = {
@@ -121,13 +121,14 @@ export const streamOpenAIResponses: StreamFunction<"openai-responses"> = (
 				api: output.api,
 				model: model.id,
 				method: "POST",
-				url: `${model.baseUrl ?? "https://api.openai.com/v1"}/responses`,
+				url: `${baseUrl ?? "https://api.openai.com/v1"}/responses`,
 				body: params,
 			};
 			const openaiStream = await client.responses.create(
 				params,
 				options?.signal ? { signal: options.signal } : undefined,
 			);
+			if (copilotPremiumRequests !== undefined) output.usage.premiumRequests = copilotPremiumRequests;
 			stream.push({ type: "start", partial: output });
 
 			let currentItem: ResponseReasoningItem | ResponseOutputMessage | ResponseFunctionToolCall | null = null;
@@ -332,6 +333,7 @@ export const streamOpenAIResponses: StreamFunction<"openai-responses"> = (
 							cacheRead: cachedTokens,
 							cacheWrite: 0,
 							totalTokens: response.usage.total_tokens || 0,
+							...(copilotPremiumRequests !== undefined ? { premiumRequests: copilotPremiumRequests } : {}),
 							cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
 						};
 					}
@@ -392,23 +394,32 @@ function createClient(
 	}
 
 	const headers = { ...(model.headers ?? {}), ...(extraHeaders ?? {}) };
+	let copilotPremiumRequests: number | undefined;
+
+	let baseUrl = model.baseUrl;
 	if (model.provider === "github-copilot") {
 		const hasImages = hasCopilotVisionInput(context.messages);
-		const copilotHeaders = buildCopilotDynamicHeaders({
+		const copilot = buildCopilotDynamicHeaders({
 			messages: context.messages,
 			hasImages,
-			initiatorOverride: getCopilotInitiatorOverride(headers),
+			premiumMultiplier: model.premiumMultiplier,
+			headers,
 		});
-		Object.assign(headers, copilotHeaders);
+		Object.assign(headers, copilot.headers);
+		copilotPremiumRequests = copilot.premiumRequests;
+		baseUrl = resolveGitHubCopilotBaseUrl(model.baseUrl, apiKey) ?? model.baseUrl;
 	}
-
-	return new OpenAI({
-		apiKey,
-		baseURL: model.baseUrl,
-		dangerouslyAllowBrowser: true,
-		maxRetries: 5,
-		defaultHeaders: headers,
-	});
+	return {
+		client: new OpenAI({
+			apiKey,
+			baseURL: baseUrl,
+			dangerouslyAllowBrowser: true,
+			maxRetries: 5,
+			defaultHeaders: headers,
+		}),
+		copilotPremiumRequests,
+		baseUrl,
+	};
 }
 
 function buildParams(model: Model<"openai-responses">, context: Context, options?: OpenAIResponsesOptions) {

package/src/stream.ts

@@ -53,6 +53,13 @@ function hasVertexAdcCredentials(): boolean {
 
 type KeyResolver = string | (() => string | undefined);
 
+function isFoundryEnabled(): boolean {
+	const value = $env.CLAUDE_CODE_USE_FOUNDRY;
+	if (!value) return false;
+	const normalized = value.trim().toLowerCase();
+	return normalized === "1" || normalized === "true" || normalized === "yes" || normalized === "on";
+}
+
 const serviceProviderMap: Record<string, KeyResolver> = {
 	openai: "OPENAI_API_KEY",
 	google: "GEMINI_API_KEY",
@@ -77,8 +84,11 @@ const serviceProviderMap: Record<string, KeyResolver> = {
 	kagi: "KAGI_API_KEY",
 	// GitHub Copilot uses GitHub personal access token
 	"github-copilot": () => $pickenv("COPILOT_GITHUB_TOKEN", "GH_TOKEN", "GITHUB_TOKEN"),
-	// ANTHROPIC_OAUTH_TOKEN takes precedence over ANTHROPIC_API_KEY
-	anthropic: () => $pickenv("ANTHROPIC_OAUTH_TOKEN", "ANTHROPIC_API_KEY"),
+	// Foundry mode optionally switches Anthropic auth to enterprise gateway credentials.
+	anthropic: () =>
+		isFoundryEnabled()
+			? $pickenv("ANTHROPIC_FOUNDRY_API_KEY", "ANTHROPIC_OAUTH_TOKEN", "ANTHROPIC_API_KEY")
+			: $pickenv("ANTHROPIC_OAUTH_TOKEN", "ANTHROPIC_API_KEY"),
 	"gitlab-duo": "GITLAB_TOKEN",
 	// Vertex AI uses Application Default Credentials, not API keys.
 	// Auth is configured via `gcloud auth application-default login`.
@@ -117,6 +127,7 @@ const serviceProviderMap: Record<string, KeyResolver> = {
 	moonshot: "MOONSHOT_API_KEY",
 	nvidia: "NVIDIA_API_KEY",
 	nanogpt: "NANO_GPT_API_KEY",
+	"lm-studio": "LM_STUDIO_API_KEY",
 	ollama: "OLLAMA_API_KEY",
 	qianfan: "QIANFAN_API_KEY",
 	"qwen-portal": () => $pickenv("QWEN_OAUTH_TOKEN", "QWEN_PORTAL_API_KEY"),
@@ -563,7 +574,7 @@ function mapOptionsForApi<TApi extends Api>(
 			const googleModel = model as Model<"google-generative-ai">;
 			const effort = clampReasoning(options.reasoning)!;
 
-			// Gemini 3 models use thinkingLevel exclusively instead of thinkingBudget.
+			// Gemini 3+ models use thinkingLevel exclusively instead of thinkingBudget.
 			// https://ai.google.dev/gemini-api/docs/thinking#set-budget
 			if (isGemini3ProModel(googleModel) || isGemini3FlashModel(googleModel)) {
 				return {
@@ -597,8 +608,8 @@ function mapOptionsForApi<TApi extends Api>(
 
 			const effort = clampReasoning(options.reasoning)!;
 
-			// Gemini 3 models use thinkingLevel instead of thinkingBudget
-			if (model.id.includes("3-pro") || model.id.includes("3-flash")) {
+			// Gemini 3+ models use thinkingLevel instead of thinkingBudget
+			if (isGemini3ProModelId(model.id) || isGemini3FlashModelId(model.id)) {
 				return {
 					...base,
 					thinking: {
@@ -688,14 +699,22 @@ function mapOptionsForApi<TApi extends Api>(
 
 type ClampedThinkingLevel = Exclude<ThinkingLevel, "xhigh">;
 
+function isGemini3ProModelId(modelId: string): boolean {
+	return /3(?:\.\d+)?-pro/.test(modelId);
+}
+
+function isGemini3FlashModelId(modelId: string): boolean {
+	return /3(?:\.\d+)?-flash/.test(modelId);
+}
+
 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");
+	// Covers gemini-3-pro, gemini-3-pro-preview, gemini-3.1-pro-preview, and future 3.x variants
+	return isGemini3ProModelId(model.id);
 }
 
 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");
+	// Covers gemini-3-flash, gemini-3-flash-preview, gemini-3.1-flash, and future 3.x variants
+	return isGemini3FlashModelId(model.id);
 }
 
 function getGemini3ThinkingLevel(
@@ -727,7 +746,7 @@ function getGemini3ThinkingLevel(
 }
 
 function getGeminiCliThinkingLevel(effort: ClampedThinkingLevel, modelId: string): GoogleThinkingLevel {
-	if (modelId.includes("3-pro")) {
+	if (isGemini3ProModelId(modelId)) {
 		// Gemini 3 Pro only supports LOW/HIGH (for now)
 		switch (effort) {
 			case "minimal":

package/src/types.ts

@@ -104,7 +104,8 @@ export type KnownProvider =
 	| "together"
 	| "venice"
 	| "vllm"
-	| "xiaomi";
+	| "xiaomi"
+	| "lm-studio";
 export type Provider = KnownProvider | string;
 
 export type ThinkingLevel = "minimal" | "low" | "medium" | "high" | "xhigh";
@@ -243,6 +244,7 @@ export interface Usage {
 	cacheRead: number;
 	cacheWrite: number;
 	totalTokens: number;
+	premiumRequests?: number;
 	cost: {
 		input: number;
 		output: number;
@@ -438,6 +440,8 @@ export interface Model<TApi extends Api = any> {
 		cacheRead: number; // $/million tokens
 		cacheWrite: number; // $/million tokens
 	};
+	/** Premium Copilot requests charged per user-initiated request (defaults to 1). */
+	premiumMultiplier?: number;
 	contextWindow: number;
 	maxTokens: number;
 	headers?: Record<string, string>;

package/src/utils/anthropic-auth.ts

@@ -4,7 +4,7 @@
  * 3-tier auth resolution:
  *   1. ANTHROPIC_SEARCH_API_KEY / ANTHROPIC_SEARCH_BASE_URL env vars
  *   2. OAuth credentials in ~/.omp/agent/agent.db (with expiry check)
- *   3. ANTHROPIC_API_KEY / ANTHROPIC_BASE_URL fallback
+ *   3. Generic Anthropic fallback (Foundry-aware key/base URL resolution)
  */
 import { $env, getAgentDbPath } from "@oh-my-pi/pi-utils";
 import { type AuthCredential, AuthCredentialStore } from "../auth-storage";
@@ -29,6 +29,26 @@ export interface AnthropicOAuthCredential {
 
 const DEFAULT_BASE_URL = "https://api.anthropic.com";
 
+function isFoundryEnabled(): boolean {
+	const value = $env.CLAUDE_CODE_USE_FOUNDRY;
+	if (!value) return false;
+	const normalized = value.trim().toLowerCase();
+	return normalized === "1" || normalized === "true" || normalized === "yes" || normalized === "on";
+}
+
+function normalizeBaseUrl(baseUrl: string | undefined): string | undefined {
+	const trimmed = baseUrl?.trim();
+	return trimmed ? trimmed.replace(/\/+$/, "") : undefined;
+}
+function resolveAnthropicBaseUrlFromEnv(): string | undefined {
+	if (isFoundryEnabled()) {
+		const foundryBaseUrl = normalizeBaseUrl($env.FOUNDRY_BASE_URL);
+		if (foundryBaseUrl) return foundryBaseUrl;
+	}
+	const anthropicBaseUrl = normalizeBaseUrl($env.ANTHROPIC_BASE_URL);
+	return anthropicBaseUrl || undefined;
+}
+
 /**
  * Checks if a token is an OAuth token by looking for sk-ant-oat prefix.
  * @param apiKey - The API key to check
@@ -81,10 +101,11 @@ async function readAnthropicOAuthCredentials(store?: AuthCredentialStore): Promi
 }
 
 /**
- * Finds Anthropic auth config using 3-tier priority:
+ * Finds Anthropic auth config using priority:
  *   1. ANTHROPIC_SEARCH_API_KEY / ANTHROPIC_SEARCH_BASE_URL
- *   2. OAuth in agent.db (with 5-minute expiry buffer)
- *   3. ANTHROPIC_API_KEY / ANTHROPIC_BASE_URL fallback
+ *   2. ANTHROPIC_FOUNDRY_API_KEY override when Foundry mode is enabled
+ *   3. OAuth in agent.db (with 5-minute expiry buffer)
+ *   4. ANTHROPIC_API_KEY / ANTHROPIC_BASE_URL fallback
  * @param store - Optional credential store (creates one from default db path if not provided)
  * @returns The first valid auth configuration found, or null if none available
  */
@@ -100,7 +121,16 @@ export async function findAnthropicAuth(store?: AuthCredentialStore): Promise<An
 		};
 	}
 
-	// 2. OAuth credentials in agent.db (with 5-minute expiry buffer)
+	// 2. Foundry explicit env override
+	const foundryApiKey = isFoundryEnabled() ? $env.ANTHROPIC_FOUNDRY_API_KEY?.trim() : undefined;
+	if (foundryApiKey) {
+		return {
+			apiKey: foundryApiKey,
+			baseUrl: resolveAnthropicBaseUrlFromEnv() ?? DEFAULT_BASE_URL,
+			isOAuth: isOAuthToken(foundryApiKey),
+		};
+	}
+	// 3. OAuth credentials in agent.db (with 5-minute expiry buffer)
 	const expiryBuffer = 5 * 60 * 1000; // 5 minutes
 	const now = Date.now();
 	const credentials = await readAnthropicOAuthCredentials(store);
@@ -115,9 +145,9 @@ export async function findAnthropicAuth(store?: AuthCredentialStore): Promise<An
 		}
 	}
 
-	// 3. Generic ANTHROPIC_API_KEY fallback
+	// 4. Generic ANTHROPIC_API_KEY fallback
 	const apiKey = getEnvApiKey("anthropic");
-	const baseUrl = $env.ANTHROPIC_BASE_URL;
+	const baseUrl = resolveAnthropicBaseUrlFromEnv();
 	if (apiKey) {
 		return {
 			apiKey,

package/src/utils/oauth/index.ts

@@ -72,6 +72,8 @@ export { loginKilo } from "./kilo";
 export { loginKimi, refreshKimiToken } from "./kimi";
 // LiteLLM (API key)
 export { loginLiteLLM } from "./litellm";
+// LM Studio (optional API key)
+export { DEFAULT_LOCAL_TOKEN, loginLmStudio } from "./lm-studio";
 // MiniMax Coding Plan (API key)
 export { loginMiniMaxCode, loginMiniMaxCodeCn } from "./minimax-code";
 // Moonshot (API key)
@@ -163,6 +165,11 @@ const builtInOAuthProviders: OAuthProviderInfo[] = [
 		name: "LiteLLM",
 		available: true,
 	},
+	{
+		id: "lm-studio",
+		name: "LM Studio (Local OpenAI-compatible)",
+		available: true,
+	},
 	{
 		id: "ollama",
 		name: "Ollama (Local OpenAI-compatible)",
@@ -338,6 +345,7 @@ export async function refreshOAuthToken(
 		case "synthetic":
 		case "together":
 		case "litellm":
+		case "lm-studio":
 		case "ollama":
 		case "xiaomi":
 		case "zai":

package/src/utils/oauth/lm-studio.ts

@@ -0,0 +1,40 @@
+/**
+ * LM Studio login flow.
+ *
+ * LM Studio provides an OpenAI-compatible API at a local base URL.
+ * It usually runs unauthenticated but can be configured to require a bearer token.
+ *
+ * This flow stores an API-key-style credential used by `/login` and auth storage.
+ */
+
+import type { OAuthController, OAuthProvider } from "./types";
+
+const PROVIDER_ID: OAuthProvider = "lm-studio";
+const _AUTH_URL = "https://lmstudio.ai/docs/api";
+const _DEFAULT_LOCAL_BASE_URL = "http://127.0.0.1:1234/v1";
+export const DEFAULT_LOCAL_TOKEN = "lm-studio-local";
+
+/**
+ * Login to LM Studio.
+ *
+ * Opens LM Studio API docs, prompts for an optional token,
+ * and returns a stored key value.
+ */
+export async function loginLmStudio(options: OAuthController): Promise<string> {
+	if (!options.onPrompt) {
+		throw new Error(`${PROVIDER_ID} login requires onPrompt callback`);
+	}
+
+	const apiKey = await options.onPrompt({
+		message: "Optional: Paste LM Studio API key (to customize endpoint URL, set LM_STUDIO_BASE_URL env var)",
+		placeholder: DEFAULT_LOCAL_TOKEN,
+		allowEmpty: true,
+	});
+
+	if (options.signal?.aborted) {
+		throw new Error("Login cancelled");
+	}
+
+	const trimmed = apiKey.trim();
+	return trimmed || DEFAULT_LOCAL_TOKEN;
+}

package/src/utils/oauth/types.ts

@@ -21,6 +21,7 @@ export type OAuthProvider =
 	| "kimi-code"
 	| "kilo"
 	| "litellm"
+	| "lm-studio"
 	| "minimax-code"
 	| "minimax-code-cn"
 	| "moonshot"