@jeffreycao/copilot-api 2.2.10 → 2.2.12

package/README.md

@@ -264,6 +264,42 @@ enabled = false
 >
 > For third-party models that do not support `tool_search`, we recommend disabling features.apps. Otherwise, each prompt may consume an additional 20,000 or more tokens.
 
+### If Codex Is Not Signed In to a GPT Account
+
+```toml
+[model_providers.copilot_api]
+name = "OpenAI"
+base_url = "http://localhost:4141"
+requires_openai_auth = false
+supports_websockets = false
+wire_api = "responses"
+request_max_retries = 3
+stream_max_retries = 3
+stream_idle_timeout_ms = 300000
+
+[model_providers.copilot_api.auth]
+command = "powershell.exe"
+args = [
+    "-NoProfile",
+    "-NonInteractive",
+    "-Command",
+    "[Console]::Out.Write($env:GITHUB_COPILOT_API_KEY)"
+]
+```
+
+macOS, replace the `auth` block with:
+
+```toml
+[model_providers.copilot_api.auth]
+command = "/bin/zsh"
+args = [
+    "-c",
+    "printf '%s' \"$GITHUB_COPILOT_API_KEY\""
+]
+```
+
+Without this configuration, Codex cannot fetch `/v1/models` while not signed in to a GPT account, so custom models are unavailable in the model picker.
+
 When a Codex client (`User-Agent` starts with `codex`) requests the top-level `GET /v1/models`, the gateway merges native Codex models with models available through the Messages adapter. The latter advertise `use_responses_lite: true`: `/v1/responses` uses **Responses → Messages** for Anthropic providers, while OpenAI-compatible providers and Chat-only Copilot models reuse the existing Messages route for **Responses → Messages → Chat Completions**, then translate streaming or JSON results back to Responses.
 
 The merged catalog is what Codex shows in its model picker, including the models exposed by your configured providers:

package/README.zh-CN.md

@@ -280,6 +280,42 @@ enabled = false
 >
 > 对于不支持 `tool_search` 的第三方模型，我们建议禁用 features.apps。否则，每个提示可能会额外消耗 20,000 多个 token。
 
+### Codex 未登录 GPT 账号时
+
+```toml
+[model_providers.copilot_api]
+name = "OpenAI"
+base_url = "http://localhost:4141"
+requires_openai_auth = false
+supports_websockets = false
+wire_api = "responses"
+request_max_retries = 3
+stream_max_retries = 3
+stream_idle_timeout_ms = 300000
+
+[model_providers.copilot_api.auth]
+command = "powershell.exe"
+args = [
+    "-NoProfile",
+    "-NonInteractive",
+    "-Command",
+    "[Console]::Out.Write($env:GITHUB_COPILOT_API_KEY)"
+]
+```
+
+macOS 将 `auth` 段替换为：
+
+```toml
+[model_providers.copilot_api.auth]
+command = "/bin/zsh"
+args = [
+    "-c",
+    "printf '%s' \"$GITHUB_COPILOT_API_KEY\""
+]
+```
+
+未按上述方式配置时，Codex 未登录 GPT 账号拉不到 `/v1/models`，无法选择自定义模型。
+
 Codex 客户端（`User-Agent` 以 `codex` 开头）请求顶层 `GET /v1/models` 时，网关会把原生 Codex 模型与可通过 Messages 适配的模型合并返回。后者会声明 `use_responses_lite: true`：调用 `/v1/responses` 后，Anthropic provider 走 **Responses → Messages**，OpenAI 兼容 provider 以及只支持 Chat 的 Copilot 模型则复用现有 Messages 路由继续走 **Responses → Messages → Chat Completions**，最终统一翻译回 Responses（包括流式事件）。
 
 合并后的模型列表会直接展示在 Codex 的模型选择界面中，包含各 provider 暴露的模型：

package/dist/main.js

@@ -30,7 +30,7 @@ if (isMcpFastPath(process.argv)) {
 	const { auth } = await import("./auth-C-RWNXGU.js");
 	const { debug } = await import("./debug-41w6Whae.js");
 	const { mcp } = await import("./mcp-fpSlKZxK.js");
-	const { start } = await import("./start-tknmnmUA.js");
+	const { start } = await import("./start-CJ6z_8Pn.js");
 	await runMain(defineCommand({
 		meta: {
 			name: "copilot-api",

package/dist/{server-DEEWMydw.js → server-C3DnUZoh.js}

@@ -576,9 +576,19 @@ const builtinProviderModelRegistry = new class BuiltinProviderModelRegistry {
 				inputModalities: ["text"],
 				maxOutputTokens: 64e3,
 				pricing: {
-					cachedInput: .02,
-					input: 1,
-					output: 2
+					cachedInput: .1,
+					input: 3,
+					output: 9
+				}
+			},
+			"deepseek-v4-flash-vision-exp": {
+				contextWindow: 1e6,
+				inputModalities: ["text", "image"],
+				maxOutputTokens: 64e3,
+				pricing: {
+					cachedInput: .1,
+					input: 3,
+					output: 9
 				}
 			},
 			"deepseek-v4-pro": {
@@ -586,9 +596,9 @@ const builtinProviderModelRegistry = new class BuiltinProviderModelRegistry {
 				inputModalities: ["text"],
 				maxOutputTokens: 64e3,
 				pricing: {
-					cachedInput: .025,
-					input: 3,
-					output: 6
+					cachedInput: .3,
+					input: 9,
+					output: 27
 				}
 			}
 		},
@@ -661,19 +671,34 @@ const builtinProviderModelRegistry = new class BuiltinProviderModelRegistry {
 				inputModalities: ["text"],
 				maxOutputTokens: 64e3,
 				pricing: {
-					cachedInput: .0028,
-					input: .14,
-					output: .28
+					cachedInput: .007,
+					input: .22,
+					output: .66
 				}
 			},
+			"deepseek-v4-flash-vision-exp": {
+				contextWindow: 1e6,
+				inputModalities: ["text", "image"],
+				maxOutputTokens: 384e3,
+				pricing: {
+					cachedInput: .007,
+					input: .22,
+					output: .66
+				},
+				reasoningEfforts: [
+					"low",
+					"high",
+					"max"
+				]
+			},
 			"deepseek-v4-pro": {
 				contextWindow: 1e6,
 				inputModalities: ["text"],
 				maxOutputTokens: 64e3,
 				pricing: {
-					cachedInput: .0145,
-					input: 1.74,
-					output: 3.48
+					cachedInput: .022,
+					input: .66,
+					output: 1.98
 				}
 			},
 			"kimi-k2.7-code": {
@@ -781,6 +806,21 @@ const builtinProviderModelRegistry = new class BuiltinProviderModelRegistry {
 					maxInputTokens: 512e3,
 					output: 2.4
 				}] }
+			},
+			"ox-alpha-free": {
+				contextWindow: 1e6,
+				inputModalities: ["text", "image"],
+				maxOutputTokens: 131072,
+				pricing: {
+					cachedInput: 0,
+					input: 0,
+					output: 0
+				},
+				reasoningEfforts: [
+					"low",
+					"high",
+					"max"
+				]
 			}
 		},
 		kimi: {
@@ -8440,6 +8480,14 @@ const MESSAGES_COMPACTION_PROMPT = [
 	"",
 	compactTextOnlyGuard
 ].join("\n");
+const MESSAGES_TOOL_CALL_TIPS = [
+	"# Tool Call Tips",
+	"- Do NOT call `exec_command` directly; that tool does not exist. Use `functions__exec` to run commands instead.",
+	"- The functions__exec tool accepts parameters only as {\"input\":\"...\"}; put the complete executable code inside input, including properly constructed tools.exec_command(...) calls and text(...) output handling.",
+	"- Construct all tools.exec_command(...) arguments strictly according to its tool definition, and use OS/shell-compatible commands for Windows, Linux, and macOS.",
+	"- Always assign the awaited tools.exec_command(...) call to a variable, then forward its output with text(result.output) and inspect result.exit_code; unforwarded output is silently dropped and makes results look empty.",
+	"- Read files with the OS-native command (Get-Content/Test-Path on Windows PowerShell, cat/ls on POSIX), quote paths containing spaces, and verify the forwarded output is non-empty before concluding a file was read."
+].join("\n");
 const COMPACTION_REPLAY_PROMPT = "The previous conversation was compacted. Continue from this handoff summary:\n\n";
 const MAX_DEVELOPER_PROMPTS = 5;
 const TOOL_NAME_PATTERN = /^[A-Za-z0-9_-]{1,64}$/u;
@@ -8460,7 +8508,7 @@ var ResponsesMessagesTranslationError = class extends Error {
 function translateResponsesToMessages(payload, options) {
 	const registry = createToolRegistry(payload);
 	const normalized = normalizeResponsesInput(payload.input);
-	const { messages, system } = translateInputToAnthropic(normalized.input, registry, payload.instructions, payload.input);
+	const { messages, system } = translateInputToAnthropic(normalized.input, registry, payload.instructions, payload.input, options.toolCallTips ?? false);
 	if (normalized.compaction) messages.push({
 		role: "user",
 		content: MESSAGES_COMPACTION_PROMPT
@@ -8649,17 +8697,11 @@ function registerMessagesTool(registration, registry) {
 	const originalKey = createOriginalToolKey(registration);
 	const existing = registry.byOriginal.get(originalKey);
 	if (existing) return existing;
-	const alias = createToolAlias(registration.namespace ? `${registration.namespace.replaceAll(".", "_")}__${registration.name}` : registration.name, originalKey, registry);
-	const descriptor = {
-		alias,
-		kind: registration.kind,
-		name: registration.name,
-		...registration.namespace ? { namespace: registration.namespace } : {}
-	};
-	registry.byAlias.set(alias, descriptor);
+	const descriptor = createMessagesToolDescriptor(registration, registry);
+	registry.byAlias.set(descriptor.alias, descriptor);
 	registry.byOriginal.set(originalKey, descriptor);
 	registry.tools.push({
-		name: alias,
+		name: descriptor.alias,
 		...registration.description ? { description: registration.description } : {},
 		input_schema: registration.kind === "custom" ? CUSTOM_TOOL_INPUT_SCHEMA : registration.parameters ?? {
 			type: "object",
@@ -8669,7 +8711,16 @@ function registerMessagesTool(registration, registry) {
 	});
 	return descriptor;
 }
-function translateInputToAnthropic(input, registry, instructions, originalInput) {
+function createMessagesToolDescriptor(registration, registry) {
+	const originalKey = createOriginalToolKey(registration);
+	return {
+		alias: createToolAlias(registration.namespace ? `${registration.namespace.replaceAll(".", "_")}__${registration.name}` : registration.name, originalKey, registry),
+		kind: registration.kind,
+		name: registration.name,
+		...registration.namespace ? { namespace: registration.namespace } : {}
+	};
+}
+function translateInputToAnthropic(input, registry, instructions, originalInput, toolCallTips) {
 	const messages = [];
 	const system = [];
 	if (instructions?.trim()) system.push({
@@ -8677,20 +8728,29 @@ function translateInputToAnthropic(input, registry, instructions, originalInput)
 		text: instructions
 	});
 	appendDeveloperPrompts(system, originalInput);
-	if (typeof input === "string") {
-		messages.push({
-			role: "user",
-			content: input
-		});
-		return {
-			messages,
-			system
-		};
-	}
-	if (!Array.isArray(input)) return {
+	if (typeof input === "string") messages.push({
+		role: "user",
+		content: input
+	});
+	else if (Array.isArray(input)) translateInputItems(input, messages, system, registry);
+	if (toolCallTips) appendToolCallTips(system);
+	return {
 		messages,
 		system
 	};
+}
+function appendToolCallTips(system) {
+	const lastSystemBlock = system.at(-1);
+	if (!lastSystemBlock) {
+		system.push({
+			type: "text",
+			text: MESSAGES_TOOL_CALL_TIPS
+		});
+		return;
+	}
+	lastSystemBlock.text = `${lastSystemBlock.text}\n\n${MESSAGES_TOOL_CALL_TIPS}`;
+}
+function translateInputItems(input, messages, system, registry) {
 	let userMessageSeen = false;
 	for (const item of input) {
 		const type = getItemType(item);
@@ -8726,10 +8786,6 @@ function translateInputToAnthropic(input, registry, instructions, originalInput)
 			default: throw new ResponsesMessagesTranslationError(`Responses Messages fallback does not support input item type '${type}'`);
 		}
 	}
-	return {
-		messages,
-		system
-	};
 }
 function translateInputMessage(item, messages, system, userMessageSeen) {
 	if (!isRecord(item)) throw new ResponsesMessagesTranslationError("Invalid Responses message item");
@@ -8794,11 +8850,12 @@ function translateInputReasoning(item, messages) {
 }
 function translateInputToolCall(item, kind, messages, registry) {
 	if (!isRecord(item)) throw new ResponsesMessagesTranslationError("Invalid Responses tool call item");
-	const descriptor = registerMessagesTool({
+	const toolIdentity = {
 		kind,
 		name: requireStringField(item, "name", `${kind}_tool_call`),
 		namespace: getOptionalStringField(item, "namespace") ?? void 0
-	}, registry);
+	};
+	const descriptor = registry.byOriginal.get(createOriginalToolKey(toolIdentity)) ?? createMessagesToolDescriptor(toolIdentity, registry);
 	const input = kind === "custom" ? { input: getStringField(item, "input") ?? "" } : parseFunctionArguments(getStringField(item, "arguments") ?? "{}", `${kind}_tool_call.arguments`);
 	appendAssistantBlock(messages, {
 		type: "tool_use",
@@ -9938,13 +9995,15 @@ async function handleResponsesViaMessages(c, options) {
 			model: options.publicModel
 		}, {
 			model: options.targetModel,
-			publicModel: options.publicModel
+			publicModel: options.publicModel,
+			toolCallTips: isCodexUserAgent(c.req.header("user-agent"))
 		});
 		const context = translation;
 		debugJson(logger$3, "Translated Messages request:", {
 			payload: translation.messagesPayload,
 			publicModel: options.publicModel,
-			targetModel: options.targetModel
+			targetModel: options.targetModel,
+			userAgent: c.req.header("user-agent") ?? ""
 		});
 		const messagesResponse = await responsesMessagesDependencies.handleCompletionPayload(c, translation.messagesPayload, {
 			compactType: translation.compaction ? 1 : void 0,
@@ -10514,4 +10573,4 @@ server.route("/:provider/images", providerImageRoutes);
 //#endregion
 export { server };
 
-//# sourceMappingURL=server-DEEWMydw.js.map
+//# sourceMappingURL=server-C3DnUZoh.js.map