@ia-ccun/code-agent-cli 0.0.57-beta.1 → 0.0.57-beta.2

package/config/agent/extensions/aicode-llm-provider.ts

@@ -0,0 +1,149 @@
+/**
+ * @author: xujianjiang
+ * AICode 内网 LLM Provider — 动态模型列表
+ *
+ * 内网网关每次上下线模型，用户不该手改 models.json。
+ * 这里用 pi.registerProvider + refreshModels 钩子，在会话启动和打开 /model 时
+ * 自动从 GET {baseUrl}/models 拉取真实模型列表，用户零操作。
+ *
+ * ponytail:
+ *   - 内网不返回 maxTokens，统一用 DEFAULT_MAX_TOKENS；个别模型要改走
+ *     models.json 的 providers.aicode.modelOverrides（对动态模型同样生效）
+ *   - 大小写重复的 id（如 Qwen2.5-VL-7B-Instruct / qwen2.5-vl-7b-instruct）
+ *     按 lowercase 去重，优先保留全小写的那个
+ *   - embedding / rerank 不是 chat 模型，pi 用不了，直接过滤
+ *   - 静态兜底列表放在扩展里，网关不可达时至少还有一个可选模型
+ */
+
+import { type ExtensionAPI, type ProviderModelConfig } from "@earendil-works/pi-coding-agent";
+import { type RefreshModelsContext } from "@earendil-works/pi-ai";
+
+const PROVIDER_ID = "aicode";
+const DEFAULT_BASE_URL = "https://ai-llm.nucc.com/v1";
+const DEFAULT_MAX_TOKENS = 8192;
+
+const BASE_URL = process.env.AICODE_BASE_URL ?? DEFAULT_BASE_URL;
+
+/** chat 模型的配置形状（ProviderModelConfig 联合类型里的 chat 分支） */
+type ChatModelConfig = Extract<ProviderModelConfig, { type?: "chat" }>;
+
+/** 内网 /v1/models 的 ModelItem（只列我们用到的字段） */
+interface InternalModelItem {
+	id: string;
+	type?: "llm" | "vision" | "embedding" | "rerank";
+	max_model_len?: number;
+	supports_image_input?: boolean;
+	supports_reasoning?: boolean;
+	reasoning_parameter_format?: string;
+}
+
+interface InternalModelList {
+	data?: InternalModelItem[];
+}
+
+/** 内网 reasoning_parameter_format → pi compat.thinkingFormat */
+const THINKING_FORMAT: Record<string, "qwen-chat-template" | "qwen"> = {
+	"qwen-chat-template": "qwen-chat-template",
+	qwen: "qwen",
+};
+
+/** 静态兜底列表：网关不可达时用 */
+const FALLBACK_MODELS: ChatModelConfig[] = [
+	{
+		id: "deepseek-v4-flash",
+		name: "deepseek-v4-flash",
+		reasoning: true,
+		input: ["text"],
+		contextWindow: 262144,
+		maxTokens: DEFAULT_MAX_TOKENS,
+		cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+		compat: { thinkingFormat: "qwen-chat-template" },
+	},
+];
+
+/**
+ * 把内网模型列表转成 pi 的模型配置。
+ * 纯函数，便于单测。
+ */
+export function toProviderModels(
+	items: InternalModelItem[],
+): ChatModelConfig[] {
+	const byId = new Map<string, InternalModelItem>();
+
+	for (const item of items) {
+		if (!item.id) continue;
+		// embedding / rerank 不是 chat 模型
+		if (item.type && item.type !== "llm" && item.type !== "vision") continue;
+
+		const key = item.id.toLowerCase();
+		const existing = byId.get(key);
+		// 大小写重复时优先保留全小写的 id（与内网其它模型 id 命名一致）
+		if (!existing || (item.id === key && existing.id !== key)) {
+			byId.set(key, item);
+		}
+	}
+
+	return [...byId.values()].map((item) => {
+		const reasoning = item.supports_reasoning === true;
+		const model: ChatModelConfig = {
+			id: item.id,
+			name: item.id,
+			reasoning,
+			input: item.supports_image_input ? ["text", "image"] : ["text"],
+			contextWindow: item.max_model_len ?? 128000,
+			maxTokens: DEFAULT_MAX_TOKENS,
+			cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+		};
+
+		if (reasoning) {
+			model.compat = {
+				thinkingFormat:
+					THINKING_FORMAT[item.reasoning_parameter_format ?? ""] ??
+					"qwen-chat-template",
+			};
+		}
+
+		return model;
+	});
+}
+
+/** 从内网网关拉取模型列表 */
+async function fetchInternalModels(
+	apiKey: string | undefined,
+	signal: AbortSignal,
+): Promise<ChatModelConfig[]> {
+	const headers: Record<string, string> = { accept: "application/json" };
+	if (apiKey) headers.authorization = `Bearer ${apiKey}`;
+
+	const res = await fetch(`${BASE_URL}/models`, { headers, signal });
+	if (!res.ok) {
+		throw new Error(`aicode: GET /v1/models failed with ${res.status}`);
+	}
+
+	const body = (await res.json()) as InternalModelList;
+	return toProviderModels(body.data ?? []);
+}
+
+export default function (pi: ExtensionAPI) {
+	pi.registerProvider(PROVIDER_ID, {
+		name: "AICode (ai-llm)",
+		baseUrl: BASE_URL,
+		api: "openai-completions",
+		// 环境变量没设时回落 /login 的 API key 输入框，凭据存 auth.json
+		apiKey: "$AICODE_API_KEY",
+		headers: { "User-Agent": "aicode-cli" },
+		models: FALLBACK_MODELS,
+		async refreshModels(
+			ctx: RefreshModelsContext,
+		): Promise<ProviderModelConfig[]> {
+			// 离线阶段（allowNetwork=false）不发请求，沿用静态兜底列表
+			if (!ctx.allowNetwork) return FALLBACK_MODELS;
+
+			const credential = ctx.credential;
+			const apiKey = credential?.type === "api_key" ? credential.key : undefined;
+			const models = await fetchInternalModels(apiKey, ctx.signal);
+			// 空列表会清空 /model 里的模型，宁可用兜底也不清空
+			return models.length > 0 ? models : FALLBACK_MODELS;
+		},
+	});
+}

package/config/agent/models.json

@@ -1,29 +1,3 @@
 {
-  "providers": {
-    "aicode": {
-      "name": "aicode llm",
-      "baseUrl": "https://ai-llm.cc.com/v1",
-      "api": "openai-completions",
-      "apiKey": "你自己的APIKEY sk-xxxxxxxxxxxxxxxxxx",
-      "headers": {
-        "User-Agent": "${VERSION}"
-      },
-      "models": [
-        {
-          "id": "qwen3.6-35b-a3b",
-          "name": "qwen3.6",
-          "contextWindow": 128000,
-          "maxTokens": 100000,
-          "reasoning": false,
-          "input": ["text"],
-          "cost": {
-            "input": 5,
-            "output": 5,
-            "cacheRead": 0,
-            "cacheWrite": 0
-          }
-        }
-      ]
-    }
-  }
+  "providers": {}
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ia-ccun/code-agent-cli",
-  "version": "0.0.57-beta.1",
+  "version": "0.0.57-beta.2",
   "type": "module",
   "main": "./dist/index.js",
   "types": "./dist/index.d.ts",

package/scripts/regression/results/.last_pid

@@ -1 +1 @@
-80819
+5531

package/scripts/regression/results/.last_workdir

@@ -1 +1 @@
-/Users/jianjiang/.aicode-regression-80797-7450000
+/Users/jianjiang/.aicode-regression-5502-6601000