coderev-cli 1.0.23 → 1.0.25

package/README.md

@@ -27,9 +27,13 @@
   - [coderev config（配置管理）](#coderev-config配置管理)
   - [coderev init（初始化）](#coderev-init初始化)
   - [coderev serve（GitHub App 自动审查）](#coderev-servegithub-app-自动审查)
+  - [coderev models（模型模板）🆕](#coderev-models模型模板)
+  - [coderev setup（模型配置）🆕](#coderev-setup模型配置)
+  - [coderev rules（规则市场）🆕](#coderev-rules规则市场)
 - [配置详解](#配置详解)
 - [平台集成](#平台集成)
 - [CI/CD 集成](#cicd-集成)
+- [VS Code 扩展 🆕](#vs-code-扩展)
 - [FAQ / 常见问题](#faq--常见问题)
 
 ---
@@ -94,19 +98,25 @@ coderev init
 
 会在当前目录创建 `.coderevrc.json` 配置文件。
 
-### 第 2 步：设置 API Key
+### 第 2 步：设置模型和 API Key
 
-coderev 需要调用 AI API 来审查代码。支持 OpenAI 和 DeepSeek。
+coderev 内置了 11 个热门模型模板，只需选择模板 + 设置 API Key 即可：
 
 ```bash
-# Windows PowerShell
-$env:DEEPSEEK_API_KEY="sk-your-key-here"
+# 查看所有可用模型
+coderev models
 
-# Linux / macOS
-export DEEPSEEK_API_KEY="sk-your-key-here"
+# 选择 DeepSeek（推荐，性价比最高）
+coderev setup --model deepseek
+
+# 设置 API Key
+# Windows PowerShell:
+$env:DEEPSEEK_API_KEY="***"
+# Linux / macOS:
+export DEEPSEEK_API_KEY="***"
 ```
 
-也可以写入配置文件（见下方「配置详解」）。
+支持所有主流 AI 提供商（见 [coderev models](#coderev-models模型模板) 章节）。
 
 ### 第 3 步：运行审查
 
@@ -481,6 +491,125 @@ GITHUB_APP_ID=123456 GITHUB_APP_WEBHOOK_SECRET=mysecret coderev serve
 
 ---
 
+### coderev models（模型模板）🆕 v1.0.24
+
+**作用**：列出所有内置的 AI 模型模板及其配置。只需选择一个模板并设置对应的 API Key 即可使用。
+
+```bash
+coderev models
+```
+
+**内置模板清单**：
+
+| 模板 | 提供商 | 默认模型 | API Key 环境变量 | 说明 |
+|------|--------|---------|-----------------|------|
+| ⭐ `deepseek` | DeepSeek | deepseek-chat | `DEEPSEEK_API_KEY` | 国产高性价比，¥1/百万token |
+| 🧠 `deepseek-r1` | DeepSeek | deepseek-reasoner | `DEEPSEEK_API_KEY` | 推理增强，适合复杂漏洞分析 |
+| `openai` | OpenAI | gpt-4o | `OPENAI_API_KEY` | GPT-4o 多模态旗舰 |
+| 🧠 `openai-o3` | OpenAI | o3-mini | `OPENAI_API_KEY` | 推理型，速度快 |
+| `qwen` | 阿里云 | qwen-plus | `DASHSCOPE_API_KEY` | 中文能力强，¥0.8/百万token |
+| ⭐ `qwen-coder` | 阿里云 | qwen-coder-plus | `DASHSCOPE_API_KEY` | 代码专精，¥2/百万token |
+| `claude` | Anthropic | claude-sonnet-4-20250514 | `ANTHROPIC_API_KEY` | 代码理解深度最强 |
+| `gemini` | Google | gemini-2.5-pro | `GEMINI_API_KEY` | 100万token上下文 |
+| `zhipu` | 智谱 AI | glm-4-plus | `ZHIPU_API_KEY` | 国产强推理 |
+| `moonshot` | 月之暗面 | moonshot-v1-8k | `MOONSHOT_API_KEY` | 长文本处理强 |
+| `codestral` | Mistral | codestral-latest | `MISTRAL_API_KEY` | 专注代码生成与审查 |
+
+> ⭐ = 推荐  |  🧠 = 推理增强型
+
+---
+
+### coderev setup（模型配置）🆕 v1.0.24
+
+**作用**：一键配置 AI 模型，支持主模型、从模型（fallback）、以及不同 Agent 使用不同模型。
+
+```bash
+coderev setup [选项]
+```
+
+**参数**：
+
+| 参数 | 说明 | 示例 |
+|------|------|------|
+| `--model <模板名>` | 设置主模型 | `coderev setup --model deepseek` |
+| `--fallback <模板名>` | 设置备用模型（主模型失败时自动切换） | `coderev setup --model deepseek --fallback qwen` |
+| `--agent-security <模板名>` | 安全 Agent 专用模型 | `coderev setup --agent-security deepseek-r1` |
+| `--agent-bugs <模板名>` | 缺陷检测 Agent 专用模型 | `coderev setup --agent-bugs qwen-coder` |
+| `--agent-quality <模板名>` | 质量检查 Agent 专用模型 | `coderev setup --agent-quality gemini` |
+| （无参数） | 查看当前配置 | `coderev setup` |
+
+**使用示例**：
+
+```bash
+# 最简：选择 DeepSeek
+coderev setup --model deepseek
+
+# 主从模型：DeepSeek 挂了自动切到千问
+coderev setup --model deepseek --fallback qwen
+
+# 不同 Agent 不同模型
+coderev setup --model deepseek --agent-security deepseek-r1 --agent-quality qwen
+
+# 查看当前配置
+coderev setup
+```
+
+**主从模型回退**：主模型调用失败（超时/网络/API 错误）时，自动切到从模型，控制台打印 `↩ Falling back to xxx`。
+
+---
+
+### coderev rules（规则市场）🆕 v1.0.23
+
+**作用**：访问 coderev SaaS 规则市场，搜索、安装、发布团队共享的审查规则包。
+
+```bash
+coderev rules <action> [选项]
+```
+
+**子命令**：
+
+| 动作 | 说明 | 示例 |
+|------|------|------|
+| `search [query]` | 搜索规则市场 | `coderev rules search react` |
+| `install <name>` | 安装规则包 | `coderev rules install react-security` |
+| `publish` | 发布本地规则到市场 | `coderev rules publish --name my-rules` |
+| `list` | 查看已安装的规则包 | `coderev rules list` |
+| `uninstall <name>` | 卸载规则包 | `coderev rules uninstall react-security` |
+| `info <name>` | 查看规则包详情 | `coderev rules info react-security` |
+
+**参数**：
+
+| 参数 | 说明 |
+|------|------|
+| `-q, --query <text>` | 搜索查询 |
+| `-n, --name <name>` | 规则包名称 |
+| `--version <ver>` | 发布版本号（默认 1.0.0） |
+| `--desc <text>` | 发布描述 |
+| `--api-url <url>` | 自定义市场 API 地址 |
+
+**使用示例**：
+
+```bash
+# 搜索规则
+coderev rules search react
+
+# 安装规则包（自动合并到 .coderevrc.json）
+coderev rules install react-security
+
+# 查看已安装
+coderev rules list
+
+# 发布当地规则
+coderev rules publish --name team-js-rules --desc "团队JS规范"
+
+# 卸载
+coderev rules uninstall react-security
+```
+
+安装后规则写入 `.coderevrc.json` 的 `rules.custom` 数组，带 `_source` 标记。
+
+---
+
 ## 配置详解
 
 ### 配置加载顺序
@@ -815,6 +944,55 @@ git diff origin/main...HEAD | coderev review --ci --min-confidence 70
 
 ---
 
+有关 GitHub Actions、GitLab CI、Jenkins 的 CI/CD 集成说明及模板，见 [docs/ci-cd.md](docs/ci-cd.md)。
+
+一键生成对应 CI 模板：
+
+```bash
+# GitHub Actions 模板
+coderev init --github-action
+# → 生成 .github/workflows/coderev.yml
+
+# GitLab CI 模板
+coderev init --gitlab-ci
+# → 生成 .gitlab-ci.yml
+```
+
+---
+
+## VS Code 扩展 🆕 v1.0.22
+
+coderev 自带 VS Code 扩展，安装后在编码时实时审查代码。
+
+### 安装
+
+```bash
+# 从本地安装（源码已包含在 vscode/ 目录中）
+cd vscode && npm install && npm run package
+# 在 VS Code 中：Ctrl+Shift+P → Extensions: Install from VSIX...
+# 选择 vscode/coderev-*.vsix
+```
+
+> 暂未上架 VS Code Marketplace，通过本地 VSIX 安装。
+
+### 功能
+
+| 功能 | 说明 |
+|------|------|
+| 工作区审查 | 右键项目目录 → `coderev: Review Workspace` |
+| 单文件审查 | 打开文件 → `Ctrl+Shift+P` → `coderev: Review File` |
+| 自动修复 | 审查发现问题后 → 一键调用 AI 修复 |
+| 保存即审查 | 开启 `coderev.lintOnSave` → 每次保存文件自动审查 |
+| Problems 面板集成 | 问题自动输出到 VS Code Problems 面板，点击跳转 |
+
+### 扩展设置
+
+- `coderev.enabled` — 是否启用自动检查（默认 true）
+- `coderev.lintOnSave` — 保存时审查（默认 false）
+- `coderev.minConfidence` — 最低置信度阈值（默认 60）
+
+---
+
 ## FAQ / 常见问题
 
 ### Q：为什么审查结果为空？

package/package.json

@@ -1,10 +1,10 @@
 {
   "name": "coderev-cli",
-  "version": "1.0.23",
+  "version": "1.0.25",
   "description": "Multi-agent AI code review for git -- parallel agents with confidence scoring",
   "main": "src/index.js",
   "bin": {
-    "coderev": "src/cli.js"
+    "coderev": "src/bin.js"
   },
   "scripts": {
     "start": "node src/cli.js",

package/src/bin.js

@@ -0,0 +1,3 @@
+// coderev entry point (no shebang — Windows + BOM safe)
+// npm bin wrapper calls this via node, so shebang is unnecessary
+require('./cli');

package/src/cli.js

@@ -848,6 +848,124 @@ program
     }
   });
 
+// ── Models ────────────────────────────────────────────────────────
+program
+  .command('models')
+  .description('List available AI model templates')
+  .option('--json', 'Output as JSON')
+  .action((options) => {
+    const { listTemplates } = require('./models');
+    const templates = listTemplates();
+
+    if (options.json) {
+      console.log(JSON.stringify(templates, null, 2));
+      return;
+    }
+
+    const tierIcons = { recommended: '⭐', reasoning: '🧠', standard: '  ' };
+    console.log(chalk.bold('\n📋 Available Model Templates\n'));
+    console.log(chalk.gray('  Usage: coderev setup --model <name>        (set primary model)'));
+    console.log(chalk.gray('         coderev setup --model <name> --fallback <name>  (set primary + fallback)'));
+    console.log(chalk.gray('         coderev setup --agent-security <name> --agent-quality <name>'));
+    console.log('');
+
+    for (const t of templates) {
+      const icon = tierIcons[t.tier] || '  ';
+      const providerTag = chalk.gray(`[${t.provider}]`);
+      console.log(`  ${icon} ${chalk.green(t.name)}  ${providerTag}`);
+      console.log(`     ${t.desc}`);
+      console.log(chalk.gray(`     API Key: ${t.apiKeyEnv}`));
+      console.log('');
+    }
+  });
+
+// ── Config ────────────────────────────────────────────────────────
+program
+  .command('setup')
+  .description('Manage coderev model configuration')
+  .option('--model <name>', 'Set primary model template')
+  .option('--fallback <name>', 'Set fallback model template')
+  .option('--agent-security <name>', 'Model for security agent')
+  .option('--agent-bugs <name>', 'Model for bug detection agent')
+  .option('--agent-quality <name>', 'Model for quality agent')
+  .action((options) => {
+    const fs = require('fs');
+    const { resolveTemplate } = require('./models');
+    const configPath = path.join(process.cwd(), '.coderevrc.json');
+
+    let config = {};
+    if (fs.existsSync(configPath)) {
+      config = JSON.parse(fs.readFileSync(configPath, 'utf-8'));
+    }
+
+    if (!config.ai) config.ai = {};
+
+    // Set primary model
+    if (options.model) {
+      try {
+        const resolved = resolveTemplate(options.model);
+        Object.assign(config.ai, resolved);
+        console.log(chalk.green(`✔ Primary model set to "${options.model}" (${resolved.model})`));
+      } catch (err) {
+        console.error(chalk.red(`✖ ${err.message}`));
+        process.exit(1);
+      }
+    }
+
+    // Set fallback model
+    if (options.fallback) {
+      try {
+        const resolved = resolveTemplate(options.fallback);
+        config.ai.fallback = {
+          enabled: true,
+          provider: resolved.provider,
+          baseURL: resolved.baseURL,
+          model: resolved.model,
+          temperature: resolved.temperature,
+          maxTokens: resolved.maxTokens,
+        };
+        console.log(chalk.green(`✔ Fallback model set to "${options.fallback}" (${resolved.model})`));
+      } catch (err) {
+        console.error(chalk.red(`✖ ${err.message}`));
+        process.exit(1);
+      }
+    }
+
+    // Set per-agent models
+    if (!config.ai.agents) config.ai.agents = {};
+    const agentOpts = [
+      ['security', options.agentSecurity],
+      ['bugs', options.agentBugs],
+      ['quality', options.agentQuality],
+    ];
+    for (const [agent, templateName] of agentOpts) {
+      if (templateName) {
+        try {
+          const resolved = resolveTemplate(templateName);
+          config.ai.agents[agent] = {
+            provider: resolved.provider,
+            baseURL: resolved.baseURL,
+            model: resolved.model,
+            temperature: resolved.temperature,
+          };
+          console.log(chalk.green(`✔ ${agent} agent model set to "${templateName}" (${resolved.model})`));
+        } catch (err) {
+          console.error(chalk.red(`✖ ${err.message}`));
+          process.exit(1);
+        }
+      }
+    }
+
+    fs.writeFileSync(configPath, JSON.stringify(config, null, 2));
+    console.log(chalk.blue(`   Config saved to ${configPath}`));
+
+    if (!options.model && !options.fallback && !options.agentSecurity && !options.agentBugs && !options.agentQuality) {
+      // No options: show current config
+      console.log(chalk.bold('\n📋 Current coderev config:\n'));
+      console.log(JSON.stringify(config, null, 2));
+    }
+  });
+
 // ── Rules Marketplace ─────────────────────────────────────────────
 program
   .command('rules <action>')

package/src/models.js

@@ -0,0 +1,160 @@
+/**
+ * Model Templates — built-in hot model presets for coderev.
+ *
+ * Usage:
+ *   coderev config --model deepseek          # switch to DeepSeek template
+ *   coderev config --model openai --fallback qwen  # primary + fallback
+ *   coderev config --agent-security deepseek --agent-quality qwen
+ *   coderev models                           # list all templates
+ *
+ * Each template provides provider + baseURL + default model,
+ * user only needs to set the corresponding API key env var.
+ */
+
+const BUILTIN_TEMPLATES = {
+  deepseek: {
+    provider: 'deepseek',
+    baseURL: 'https://api.deepseek.com',
+    model: 'deepseek-chat',
+    apiKeyEnv: 'DEEPSEEK_API_KEY',
+    desc: 'DeepSeek V3 — 国产高性价比，¥1/百万token',
+    tier: 'recommended',
+  },
+  'deepseek-r1': {
+    provider: 'deepseek',
+    baseURL: 'https://api.deepseek.com',
+    model: 'deepseek-reasoner',
+    apiKeyEnv: 'DEEPSEEK_API_KEY',
+    desc: 'DeepSeek R1 — 推理增强，适合复杂漏洞分析',
+    tier: 'reasoning',
+  },
+  openai: {
+    provider: 'openai',
+    baseURL: 'https://api.openai.com/v1',
+    model: 'gpt-4o',
+    apiKeyEnv: 'OPENAI_API_KEY',
+    desc: 'GPT-4o — OpenAI 多模态旗舰',
+    tier: 'standard',
+  },
+  'openai-o3': {
+    provider: 'openai',
+    baseURL: 'https://api.openai.com/v1',
+    model: 'o3-mini',
+    apiKeyEnv: 'OPENAI_API_KEY',
+    desc: 'OpenAI o3-mini — 推理型，速度快',
+    tier: 'reasoning',
+  },
+  qwen: {
+    provider: 'openai',
+    baseURL: 'https://dashscope.aliyuncs.com/compatible-mode/v1',
+    model: 'qwen-plus',
+    apiKeyEnv: 'DASHSCOPE_API_KEY',
+    desc: '通义千问 Qwen-Plus — 中文能力强，¥0.8/百万token',
+    tier: 'standard',
+  },
+  'qwen-coder': {
+    provider: 'openai',
+    baseURL: 'https://dashscope.aliyuncs.com/compatible-mode/v1',
+    model: 'qwen-coder-plus',
+    apiKeyEnv: 'DASHSCOPE_API_KEY',
+    desc: '通义千问 Coder — 代码专精，¥2/百万token',
+    tier: 'recommended',
+  },
+  claude: {
+    provider: 'openai',
+    baseURL: 'https://api.anthropic.com/v1',
+    model: 'claude-sonnet-4-20250514',
+    apiKeyEnv: 'ANTHROPIC_API_KEY',
+    desc: 'Claude Sonnet 4 — 代码理解深度最强',
+    tier: 'standard',
+  },
+  gemini: {
+    provider: 'openai',
+    baseURL: 'https://generativelanguage.googleapis.com/v1beta/openai',
+    model: 'gemini-2.5-pro',
+    apiKeyEnv: 'GEMINI_API_KEY',
+    desc: 'Gemini 2.5 Pro — Google，100万token上下文',
+    tier: 'standard',
+  },
+  zhipu: {
+    provider: 'openai',
+    baseURL: 'https://open.bigmodel.cn/api/paas/v4',
+    model: 'glm-4-plus',
+    apiKeyEnv: 'ZHIPU_API_KEY',
+    desc: '智谱 GLM-4-Plus — 国产强推理',
+    tier: 'standard',
+  },
+  moonshot: {
+    provider: 'openai',
+    baseURL: 'https://api.moonshot.cn/v1',
+    model: 'moonshot-v1-8k',
+    apiKeyEnv: 'MOONSHOT_API_KEY',
+    desc: '月之暗面 Kimi — 长文本处理强',
+    tier: 'standard',
+  },
+  codestral: {
+    provider: 'openai',
+    baseURL: 'https://api.mistral.ai/v1',
+    model: 'codestral-latest',
+    apiKeyEnv: 'MISTRAL_API_KEY',
+    desc: 'Mistral Codestral — 专注代码生成与审查',
+    tier: 'standard',
+  },
+};
+
+/**
+ * Resolve a model template by name, with optional user overrides.
+ * Returns a full ai config block ready for use.
+ *
+ * @param {string} templateName - Template name (e.g. 'deepseek', 'qwen')
+ * @param {object} [overrides] - User overrides (model, baseURL, provider, etc.)
+ * @returns {object} Resolved ai config
+ */
+function resolveTemplate(templateName, overrides = {}) {
+  const template = BUILTIN_TEMPLATES[templateName];
+  if (!template) {
+    throw new Error(
+      `Unknown model template "${templateName}". ` +
+      `Available: ${Object.keys(BUILTIN_TEMPLATES).join(', ')}`
+    );
+  }
+
+  return {
+    provider: overrides.provider || template.provider,
+    baseURL: overrides.baseURL || template.baseURL,
+    model: overrides.model || template.model,
+    apiKeyEnv: overrides.apiKeyEnv || template.apiKeyEnv,
+    temperature: overrides.temperature ?? 0.3,
+    maxTokens: overrides.maxTokens || 4096,
+    _template: templateName,
+  };
+}
+
+/**
+ * List all built-in templates with key info.
+ * @returns {Array<{name: string, provider: string, model: string, apiKeyEnv: string, desc: string, tier: string}>}
+ */
+function listTemplates() {
+  return Object.entries(BUILTIN_TEMPLATES).map(([name, t]) => ({
+    name,
+    provider: t.provider,
+    model: t.model,
+    apiKeyEnv: t.apiKeyEnv,
+    desc: t.desc,
+    tier: t.tier,
+  }));
+}
+
+/**
+ * Get a single template by name. Returns null if not found.
+ */
+function getTemplate(name) {
+  return BUILTIN_TEMPLATES[name] || null;
+}
+
+module.exports = {
+  BUILTIN_TEMPLATES,
+  resolveTemplate,
+  listTemplates,
+  getTemplate,
+};

package/src/models.test.js

@@ -0,0 +1,80 @@
+const { describe, it } = require('node:test');
+const assert = require('assert');
+const { BUILTIN_TEMPLATES, resolveTemplate, listTemplates, getTemplate } = require('./models');
+
+describe('models.js', () => {
+  it('should have all 11 built-in templates', () => {
+    const templates = listTemplates();
+    assert.strictEqual(templates.length, 11);
+  });
+
+  it('should resolve deepseek template correctly', () => {
+    const result = resolveTemplate('deepseek');
+    assert.strictEqual(result.provider, 'deepseek');
+    assert.strictEqual(result.model, 'deepseek-chat');
+    assert.strictEqual(result.baseURL, 'https://api.deepseek.com');
+    assert.strictEqual(result.apiKeyEnv, 'DEEPSEEK_API_KEY');
+    assert.strictEqual(result._template, 'deepseek');
+  });
+
+  it('should resolve qwen template with openai-compatible provider', () => {
+    const result = resolveTemplate('qwen');
+    assert.strictEqual(result.provider, 'openai');
+    assert.strictEqual(result.model, 'qwen-plus');
+    assert.ok(result.baseURL.includes('dashscope'));
+    assert.strictEqual(result.apiKeyEnv, 'DASHSCOPE_API_KEY');
+  });
+
+  it('should apply user overrides on template', () => {
+    const result = resolveTemplate('openai', {
+      model: 'gpt-4o-mini',
+      temperature: 0.1,
+    });
+    assert.strictEqual(result.provider, 'openai');
+    assert.strictEqual(result.model, 'gpt-4o-mini');
+    assert.strictEqual(result.temperature, 0.1);
+    assert.strictEqual(result._template, 'openai');
+  });
+
+  it('should throw on unknown template', () => {
+    assert.throws(() => resolveTemplate('nonexistent'), {
+      message: /Unknown model template/,
+    });
+  });
+
+  it('getTemplate should return null for unknown', () => {
+    assert.strictEqual(getTemplate('nonexistent'), null);
+  });
+
+  it('getTemplate should return object for known', () => {
+    const t = getTemplate('deepseek');
+    assert.ok(t);
+    assert.strictEqual(t.model, 'deepseek-chat');
+  });
+
+  it('listTemplates should return array with required fields', () => {
+    const templates = listTemplates();
+    for (const t of templates) {
+      assert.ok(t.name, `template should have name: ${JSON.stringify(t)}`);
+      assert.ok(t.provider, `template should have provider: ${JSON.stringify(t)}`);
+      assert.ok(t.model, `template should have model: ${JSON.stringify(t)}`);
+      assert.ok(t.apiKeyEnv, `template should have apiKeyEnv: ${JSON.stringify(t)}`);
+      assert.ok(t.desc, `template should have desc: ${JSON.stringify(t)}`);
+    }
+  });
+
+  it('should have recommended and reasoning tiers', () => {
+    const templates = listTemplates();
+    const recommended = templates.filter(t => t.tier === 'recommended');
+    const reasoning = templates.filter(t => t.tier === 'reasoning');
+    assert.ok(recommended.length >= 2, 'should have at least 2 recommended templates');
+    assert.ok(reasoning.length >= 2, 'should have at least 2 reasoning templates');
+  });
+
+  it('deepseek-r1 should be a reasoning model', () => {
+    const t = getTemplate('deepseek-r1');
+    assert.ok(t);
+    assert.strictEqual(t.tier, 'reasoning');
+    assert.strictEqual(t.model, 'deepseek-reasoner');
+  });
+});

package/src/reviewer.js

@@ -186,7 +186,8 @@ async function runParallelAgents(apiKey, config, prompts) {
 
   const tasks = prompts.map(async (p) => {
     try {
-      const text = await callAI(apiKey, p.messages, config);
+      // Pass agent name so callAI can select per-agent model config
+      const text = await callAI(apiKey, p.messages, config, p.role.name);
       const parsed = parseReviewResponse(text);
       return { name: p.role.name, success: true, ...parsed };
     } catch (err) {
@@ -440,32 +441,84 @@ Important:
 }
 
 /**
- * Call the AI provider.
+ * Call the AI provider with support for:
+ * - Per-agent model selection (ai.agents[agentType])
+ * - Primary/fallback model switching (ai.fallback)
+ *
+ * @param {string} apiKey - API key
+ * @param {Array} messages - Chat messages
+ * @param {object} config - Full config object
+ * @param {string} [agentType] - Agent type for per-agent model selection ('security'|'bugs'|'quality')
+ * @returns {Promise<string>} AI response text
  */
-async function callAI(apiKey, messages, config) {
+async function callAI(apiKey, messages, config, agentType) {
   const aiConfig = config.ai || {};
-  const provider = aiConfig.provider || 'openai';
 
-  const OpenAI = require('openai');
+  // Resolve model config for this agent
+  let modelConfig = { ...aiConfig };
+  if (agentType && aiConfig.agents && aiConfig.agents[agentType]) {
+    const agentCfg = aiConfig.agents[agentType];
+    modelConfig = { ...aiConfig, ...agentCfg };
+  }
 
-  // Determine base URL from provider or config
-  let baseURL = aiConfig.baseURL;
+  const provider = modelConfig.provider || 'openai';
+
+  // Determine base URL
+  let baseURL = modelConfig.baseURL;
   if (!baseURL) {
     if (provider === 'deepseek') baseURL = 'https://api.deepseek.com';
-    // openai and all others default to undefined (official endpoint)
   }
 
-  const defaultModels = {
-    openai: 'gpt-4o',
-    deepseek: 'deepseek-chat',
-  };
+  const defaultModels = { openai: 'gpt-4o', deepseek: 'deepseek-chat' };
+  const model = modelConfig.model || defaultModels[provider] || provider;
+
+  // Try primary model first
+  try {
+    return await _doCallAI(apiKey, messages, {
+      model,
+      baseURL,
+      temperature: modelConfig.temperature ?? 0.3,
+      maxTokens: modelConfig.maxTokens || 4096,
+    });
+  } catch (err) {
+    const fallback = aiConfig.fallback;
+    const fallbackEnabled = fallback && fallback.enabled !== false;
+
+    if (!fallbackEnabled || !fallback.model) {
+      throw err;
+    }
 
+    // Log fallback attempt (visible in --debug mode)
+    const chalk = require('chalk');
+    console.error(chalk.yellow(`⚠️  Primary model failed: ${err.message}`));
+    console.error(chalk.blue(`↩  Falling back to ${fallback.model}...`));
+
+    const fbProvider = fallback.provider || provider;
+    let fbBaseURL = fallback.baseURL;
+    if (!fbBaseURL) {
+      if (fbProvider === 'deepseek') fbBaseURL = 'https://api.deepseek.com';
+    }
+
+    return await _doCallAI(apiKey, messages, {
+      model: fallback.model,
+      baseURL: fbBaseURL,
+      temperature: fallback.temperature ?? 0.3,
+      maxTokens: fallback.maxTokens || 4096,
+    });
+  }
+}
+
+/**
+ * Internal: make a single AI call.
+ */
+async function _doCallAI(apiKey, messages, { model, baseURL, temperature, maxTokens }) {
+  const OpenAI = require('openai');
   const client = new OpenAI({ apiKey, baseURL: baseURL || undefined });
 
   const response = await client.chat.completions.create({
-    model: aiConfig.model || defaultModels[provider] || provider,
-    temperature: aiConfig.temperature ?? 0.3,
-    max_tokens: aiConfig.maxTokens || 4096,
+    model,
+    temperature,
+    max_tokens: maxTokens,
     messages,
   });