diff --git a/README.md b/README.md
index 0cb8d6b9..2670a370 100644
--- a/README.md
+++ b/README.md
@@ -65,7 +65,7 @@ CharacterArc（弧光）不是"只会对话的 AI 壳子"，而是一套围绕
   <tr>
     <td colspan="2" valign="top">
       <h4>🌐 多厂商接入</h4>
-      <p>支持所有 OpenAI 兼容接口（DeepSeek、通义千问、智谱、Kimi、SiliconFlow、Ollama 等）和 Anthropic 协议（官方及中转站），只需选协议填地址即可。</p>
+      <p>支持 Codex CLI、OpenAI 兼容接口（DeepSeek、通义千问、智谱、Kimi、SiliconFlow、Ollama 等）和 Anthropic 协议（官方及中转站）。</p>
     </td>
   </tr>
 </table>
@@ -184,7 +184,7 @@ CharacterArc（弧光）不是"只会对话的 AI 壳子"，而是一套围绕
 | 富文本编辑 | TipTap |
 | 持久化 | SQLite（主进程） |
 | 关系图谱 | Cytoscape |
-| AI SDK | Vercel AI SDK (@ai-sdk/openai, @ai-sdk/anthropic) |
+| AI 接入 | Vercel AI SDK（OpenAI / Anthropic）+ 本机 Codex CLI |
 | 文档解析 | mammoth (.docx)、marked (Markdown) |
 
 ## 📋 环境要求
@@ -234,6 +234,7 @@ macOS 产物为 ad-hoc 签名、未公证的 `.dmg` / `.zip`，适合内部安
       <b>📝 文本生成</b>
       <ul>
         <li>支持维护多套接口配置，并在标题栏快速切换</li>
+        <li>Codex CLI：复用本机登录状态，无需 API Key；支持自动查找路径、拉取模型和设置推理强度</li>
         <li>协议类型：OpenAI 兼容协议 / Anthropic 协议</li>
         <li>Base URL（只需填域名或路径前缀，系统自动补全 /v1）</li>
         <li>API Key</li>
diff --git a/electron/main/ai/agent/run-agent.ts b/electron/main/ai/agent/run-agent.ts
index 04f2a805..309b2314 100644
--- a/electron/main/ai/agent/run-agent.ts
+++ b/electron/main/ai/agent/run-agent.ts
@@ -5,6 +5,8 @@ import type { Tool, ToolContext } from './tools/types'
 import { stripReasoningMarkup } from '../reasoning'
 import { isAiStreamIdleTimeoutError } from '../sse'
 import { resolveSamplingOptions } from '../request-options'
+import { isCodexCliProvider } from '@shared/ai-provider-catalog'
+import { runCodexCli } from '../codex-cli'
 
 export type RunAgentParams = {
   settings: AppSettings
@@ -94,6 +96,26 @@ function mergeUsage(a: AiRunUsage | undefined, b: AiRunUsage | undefined): AiRun
 
 export async function runAgent(params: RunAgentParams): Promise<RunAgentResult> {
   const maxSteps = params.maxSteps ?? 8
+  if (isCodexCliProvider(params.settings.provider)) {
+    params.handlers.onAgentStatus('正在通过 Codex CLI 思考...', 1, 1)
+    const result = await runCodexCli(
+      params.settings,
+      {
+        system: [
+          params.systemPrompt,
+          '当前通过 Codex CLI 文本模式运行，CharacterArc 的进程内工具不可用。请直接根据已提供的上下文回答，不要尝试调用工具。'
+        ].join('\n\n'),
+        user: params.userPrompt
+      },
+      { signal: params.ctx.signal, handlers: params.handlers }
+    )
+    return {
+      finalText: stripReasoningMarkup(result.text),
+      toolCalls: [],
+      iterations: 1,
+      usage: result.usage
+    }
+  }
   const toolCalls: ToolCallTrace[] = []
   const toolStartTimes = new Map<string, number>()
   let stepCount = 0
diff --git a/electron/main/ai/codex-cli.test.mjs b/electron/main/ai/codex-cli.test.mjs
new file mode 100644
index 00000000..98309192
--- /dev/null
+++ b/electron/main/ai/codex-cli.test.mjs
@@ -0,0 +1,81 @@
+import assert from 'node:assert/strict'
+import test from 'node:test'
+
+import { buildCodexExecArgs, parseCodexJsonLine, resolveCodexCommand } from './codex-cli.ts'
+import { normalizeSettings, validateSettings } from './settings.ts'
+import { getAiProviderCatalogEntry, shouldTryStreamingAgent } from '../../shared/ai-provider-catalog.ts'
+
+const codexSettings = {
+  provider: 'codex-cli',
+  model: '',
+  apiKey: '',
+  baseUrl: '',
+  codexCliPath: '',
+  codexReasoningEffort: 'default',
+  embeddingModel: '',
+  imageModel: '',
+  imageApiKey: '',
+  imageBaseUrl: ''
+}
+
+test('Codex CLI 预设不要求 API 地址或 Key，并使用 default 模型', () => {
+  const preset = getAiProviderCatalogEntry('codex-cli')
+  const normalized = normalizeSettings(codexSettings)
+
+  assert.equal(preset?.transport, 'codex-cli')
+  assert.equal(preset?.supportsEmbedding, false)
+  assert.equal(normalized.model, 'default')
+  assert.equal(normalized.baseUrl, '')
+  assert.doesNotThrow(() => validateSettings(normalized))
+  assert.equal(shouldTryStreamingAgent('global-assistant', 'codex-cli', 'default'), false)
+})
+
+test('Codex exec 使用只读 JSONL 与 stdin，并传递模型和推理强度', () => {
+  const args = buildCodexExecArgs({
+    ...codexSettings,
+    model: 'gpt-5.6-sol',
+    codexReasoningEffort: 'high'
+  })
+
+  assert.deepEqual(args.slice(0, 6), [
+    'exec',
+    '--json',
+    '--skip-git-repo-check',
+    '--sandbox',
+    'read-only',
+    '-c'
+  ])
+  assert.ok(args.includes('features.shell_tool=false'))
+  assert.ok(args.includes('model_reasoning_effort="high"'))
+  assert.deepEqual(args.slice(-3), ['--model', 'gpt-5.6-sol', '-'])
+})
+
+test('Codex CLI 路径拒绝混入 shell 环境变量', () => {
+  assert.throws(
+    () => resolveCodexCommand({
+      ...codexSettings,
+      codexCliPath: 'HTTPS_PROXY=http://127.0.0.1:7890 /opt/homebrew/bin/codex'
+    }),
+    /只能填写可执行文件路径/
+  )
+})
+
+test('Codex JSONL 事件转换为正文、推理和 token 用量', () => {
+  const reasoning = []
+  assert.deepEqual(
+    parseCodexJsonLine(
+      '{"type":"item.completed","item":{"type":"reasoning","text":"分析中"}}',
+      { onTextDelta() {}, onReasoningDelta: (delta) => reasoning.push(delta) }
+    ),
+    {}
+  )
+  assert.deepEqual(reasoning, ['分析中'])
+  assert.deepEqual(
+    parseCodexJsonLine('{"type":"item.completed","item":{"type":"agent_message","text":"完成"}}'),
+    { text: '完成' }
+  )
+  assert.deepEqual(
+    parseCodexJsonLine('{"type":"turn.completed","usage":{"input_tokens":12,"output_tokens":8}}'),
+    { usage: { promptTokens: 12, completionTokens: 8, totalTokens: 20 } }
+  )
+})
diff --git a/electron/main/ai/codex-cli.ts b/electron/main/ai/codex-cli.ts
new file mode 100644
index 00000000..f1860268
--- /dev/null
+++ b/electron/main/ai/codex-cli.ts
@@ -0,0 +1,420 @@
+import { spawn, type ChildProcessWithoutNullStreams } from 'node:child_process'
+import { constants as fsConstants, accessSync, existsSync, statSync } from 'node:fs'
+import { delimiter, dirname, extname, join } from 'node:path'
+import { homedir, tmpdir } from 'node:os'
+import { createInterface } from 'node:readline'
+import type { AiRunUsage, AppSettings, AiStreamHandlers, PromptPair } from './shared-types'
+import { normalizeProxyUrl } from './proxy-fetch'
+
+const CODEX_MODEL_DISCOVERY_TIMEOUT_MS = 15_000
+const CODEX_COMMAND_TIMEOUT_MS = 15_000
+const MAX_CAPTURED_OUTPUT = 5 * 1024 * 1024
+const DEFAULT_CODEX_MODELS = ['default', 'gpt-5.5', 'gpt-5.4-mini']
+
+export type CodexCliModel = {
+  id: string
+  ownedBy: string | null
+}
+
+export type CodexCliResult = {
+  text: string
+  usage?: AiRunUsage
+}
+
+type CodexCommand = {
+  program: string
+  prefixArgs: string[]
+}
+
+type CapturedCommand = {
+  stdout: string
+  stderr: string
+  exitCode: number | null
+}
+
+function isExecutable(path: string): boolean {
+  try {
+    if (!statSync(path).isFile()) return false
+    if (process.platform !== 'win32') accessSync(path, fsConstants.X_OK)
+    return true
+  } catch {
+    return false
+  }
+}
+
+function executableNames(): string[] {
+  return process.platform === 'win32'
+    ? ['codex.cmd', 'codex.exe', 'codex.bat', 'codex.com', 'codex']
+    : ['codex']
+}
+
+function candidateExecutableDirs(): string[] {
+  const dirs = (process.env.PATH ?? '').split(delimiter).filter(Boolean)
+  if (process.platform === 'win32') {
+    if (process.env.APPDATA) dirs.push(join(process.env.APPDATA, 'npm'))
+  } else {
+    if (process.env.HOME) {
+      dirs.push(join(process.env.HOME, '.local', 'bin'))
+      dirs.push(join(process.env.HOME, '.npm-global', 'bin'))
+      dirs.push(join(process.env.HOME, '.volta', 'bin'))
+    }
+    dirs.push('/opt/homebrew/bin', '/usr/local/bin', '/usr/bin')
+    dirs.push('/Applications/ChatGPT.app/Contents/Resources')
+  }
+  return [...new Set(dirs)]
+}
+
+function findCodexInDirectory(directory: string): string | null {
+  for (const name of executableNames()) {
+    const candidate = join(directory, name)
+    if (isExecutable(candidate)) return candidate
+  }
+  return null
+}
+
+function looksLikeEnvironmentAssignment(value: string): boolean {
+  const firstToken = value.split(/\s+/, 1)[0] ?? ''
+  return /^[A-Za-z_][A-Za-z\d_]*=/.test(firstToken)
+}
+
+function expandHomePath(value: string): string {
+  if (value === '~') return homedir()
+  if (value.startsWith('~/') || value.startsWith('~\\')) {
+    return join(homedir(), value.slice(2))
+  }
+  return value
+}
+
+function commandForExecutable(executable: string): CodexCommand {
+  if (process.platform !== 'win32' || !['.cmd', '.bat'].includes(extname(executable).toLowerCase())) {
+    return { program: executable, prefixArgs: [] }
+  }
+
+  // npm 在 Windows 上安装的是 cmd shim。直接调用相邻的 JS 入口，可以避免
+  // shell quoting，也能正确处理包含空格的用户名和安装路径。
+  const codexJs = join(dirname(executable), 'node_modules', '@openai', 'codex', 'bin', 'codex.js')
+  if (!existsSync(codexJs)) {
+    throw new Error('当前 Codex CLI 路径是 .cmd/.bat，但未找到对应的 @openai/codex JS 入口。请填写 codex.exe 或 npm 全局安装目录中的 codex.cmd。')
+  }
+  const bundledNode = join(dirname(executable), 'node.exe')
+  return {
+    program: isExecutable(bundledNode) ? bundledNode : 'node',
+    prefixArgs: [codexJs]
+  }
+}
+
+/** 解析用户配置的 Codex 路径；留空时兼容 Electron GUI 常见的精简 PATH。 */
+export function resolveCodexCommand(settings: AppSettings): CodexCommand {
+  const configured = settings.codexCliPath?.trim() ?? ''
+  if (looksLikeEnvironmentAssignment(configured)) {
+    throw new Error('Codex CLI 路径只能填写可执行文件路径，不能包含环境变量或启动命令。')
+  }
+
+  if (configured) {
+    const expanded = expandHomePath(configured)
+    let executable = expanded
+    try {
+      if (statSync(expanded).isDirectory()) {
+        executable = findCodexInDirectory(expanded) ?? ''
+      }
+    } catch {
+      // 下面统一给出可操作的路径错误。
+    }
+    if (!executable || !isExecutable(executable)) {
+      throw new Error('未找到指定的 Codex CLI。请填写 codex 可执行文件，或包含它的目录。')
+    }
+    return commandForExecutable(executable)
+  }
+
+  for (const directory of candidateExecutableDirs()) {
+    const executable = findCodexInDirectory(directory)
+    if (executable) return commandForExecutable(executable)
+  }
+
+  // 保留裸命令作为最后兜底，让 spawn 返回准确的 ENOENT 错误。
+  return { program: 'codex', prefixArgs: [] }
+}
+
+function buildCodexEnvironment(settings: AppSettings): NodeJS.ProcessEnv {
+  const env = { ...process.env }
+  const proxyUrl = settings.proxyUrl?.trim() ? normalizeProxyUrl(settings.proxyUrl) : ''
+  if (proxyUrl) {
+    env.HTTP_PROXY = proxyUrl
+    env.HTTPS_PROXY = proxyUrl
+    env.http_proxy = proxyUrl
+    env.https_proxy = proxyUrl
+  }
+  return env
+}
+
+function buildCodexPrompt(settings: AppSettings, prompt: PromptPair): string {
+  return [
+    prompt.system.trim(),
+    '',
+    '【运行时模型信息】',
+    `当前请求使用 Codex CLI，模型标识：${settings.model || 'default'}。`,
+    '',
+    '【用户请求】',
+    prompt.user.trim()
+  ].join('\n')
+}
+
+/** 构建只读、无持久会话的 Codex exec 参数。Prompt 始终通过 stdin 传入。 */
+export function buildCodexExecArgs(settings: AppSettings): string[] {
+  const args = [
+    'exec',
+    '--json',
+    '--skip-git-repo-check',
+    '--sandbox',
+    'read-only',
+    '-c',
+    'features.shell_tool=false',
+    '-c',
+    'web_search="disabled"'
+  ]
+  const effort = settings.codexReasoningEffort?.trim().toLowerCase()
+  if (effort && effort !== 'default') {
+    args.push('-c', `model_reasoning_effort=${JSON.stringify(effort)}`)
+  }
+  const model = settings.model?.trim()
+  if (model && model.toLowerCase() !== 'default') {
+    args.push('--model', model)
+  }
+  args.push('-')
+  return args
+}
+
+function makeAbortError(): Error {
+  const error = new Error('任务已取消。')
+  error.name = 'AbortError'
+  return error
+}
+
+function classifySpawnError(error: unknown): Error {
+  const message = error instanceof Error ? error.message : String(error)
+  if (/ENOENT|not found|no such file/i.test(message)) {
+    return new Error('未找到 Codex CLI。请先安装 @openai/codex，或在设置中填写 Codex CLI 路径。')
+  }
+  return new Error(`Codex CLI 启动失败：${message}`)
+}
+
+function classifyRunError(stderr: string, fallback = 'Codex CLI 执行失败。'): Error {
+  const detail = stderr.trim()
+  if (/not authenticated|log\s*in|login|authentication/i.test(detail)) {
+    return new Error('Codex CLI 尚未登录。请先在终端执行 codex login。')
+  }
+  return new Error(detail ? `Codex CLI 执行失败：${detail.slice(0, 1200)}` : fallback)
+}
+
+function appendCaptured(current: string, chunk: Buffer | string): string {
+  if (current.length >= MAX_CAPTURED_OUTPUT) return current
+  return (current + chunk.toString()).slice(0, MAX_CAPTURED_OUTPUT)
+}
+
+export function parseCodexJsonLine(
+  line: string,
+  handlers?: AiStreamHandlers
+): { text?: string; usage?: AiRunUsage; error?: string } {
+  let value: Record<string, unknown>
+  try {
+    value = JSON.parse(line) as Record<string, unknown>
+  } catch {
+    return {}
+  }
+
+  const type = String(value.type ?? '')
+  if (type === 'turn.failed' || type === 'error') {
+    const nested = value.error && typeof value.error === 'object'
+      ? String((value.error as Record<string, unknown>).message ?? '')
+      : String(value.error ?? value.message ?? '')
+    return { error: nested || 'Codex CLI 返回失败事件。' }
+  }
+
+  if (type === 'turn.completed') {
+    const usage = value.usage && typeof value.usage === 'object'
+      ? value.usage as Record<string, unknown>
+      : null
+    if (!usage) return {}
+    const promptTokens = Number(usage.input_tokens ?? usage.prompt_tokens)
+    const completionTokens = Number(usage.output_tokens ?? usage.completion_tokens)
+    const normalized: AiRunUsage = {
+      promptTokens: Number.isFinite(promptTokens) ? promptTokens : undefined,
+      completionTokens: Number.isFinite(completionTokens) ? completionTokens : undefined
+    }
+    if (normalized.promptTokens !== undefined || normalized.completionTokens !== undefined) {
+      normalized.totalTokens = (normalized.promptTokens ?? 0) + (normalized.completionTokens ?? 0)
+      return { usage: normalized }
+    }
+    return {}
+  }
+
+  if (type !== 'item.completed' || !value.item || typeof value.item !== 'object') return {}
+  const item = value.item as Record<string, unknown>
+  const itemType = String(item.type ?? '')
+  const text = typeof item.text === 'string'
+    ? item.text
+    : typeof item.message === 'string'
+      ? item.message
+      : typeof item.content === 'string'
+        ? item.content
+        : ''
+  if (!text) return {}
+  if (itemType === 'reasoning') {
+    handlers?.onReasoningDelta?.(text)
+    return {}
+  }
+  return itemType === 'agent_message' ? { text } : {}
+}
+
+/** 通过 Codex CLI 完成一次生成；JSONL 事件会映射到现有流式回调。 */
+export async function runCodexCli(
+  settings: AppSettings,
+  prompt: PromptPair,
+  options: { signal?: AbortSignal; handlers?: AiStreamHandlers } = {}
+): Promise<CodexCliResult> {
+  options.signal?.throwIfAborted()
+  const command = resolveCodexCommand(settings)
+  const args = [...command.prefixArgs, ...buildCodexExecArgs(settings)]
+  const child = spawn(command.program, args, {
+    cwd: tmpdir(),
+    env: buildCodexEnvironment(settings),
+    stdio: ['pipe', 'pipe', 'pipe'],
+    windowsHide: true
+  }) as ChildProcessWithoutNullStreams
+
+  return await new Promise<CodexCliResult>((resolve, reject) => {
+    let fullText = ''
+    let stderr = ''
+    let usage: AiRunUsage | undefined
+    let eventError = ''
+    let settled = false
+    const lines = createInterface({ input: child.stdout })
+
+    const cleanup = () => {
+      options.signal?.removeEventListener('abort', abort)
+      lines.close()
+    }
+    const finishReject = (error: Error) => {
+      if (settled) return
+      settled = true
+      cleanup()
+      reject(error)
+    }
+    const abort = () => {
+      child.kill()
+      finishReject(makeAbortError())
+    }
+
+    options.signal?.addEventListener('abort', abort, { once: true })
+    child.on('error', (error) => finishReject(classifySpawnError(error)))
+    child.stderr.on('data', (chunk) => { stderr = appendCaptured(stderr, chunk) })
+    lines.on('line', (line) => {
+      const parsed = parseCodexJsonLine(line, options.handlers)
+      if (parsed.text) {
+        fullText += parsed.text
+        options.handlers?.onTextDelta(parsed.text)
+      }
+      if (parsed.usage) usage = parsed.usage
+      if (parsed.error) eventError = parsed.error
+    })
+    child.on('close', (exitCode) => {
+      if (settled) return
+      settled = true
+      cleanup()
+      if (options.signal?.aborted) {
+        reject(makeAbortError())
+      } else if (eventError) {
+        reject(classifyRunError(eventError))
+      } else if (exitCode !== 0) {
+        reject(classifyRunError(stderr))
+      } else if (!fullText.trim()) {
+        reject(new Error('Codex CLI 已结束，但没有返回可见正文。'))
+      } else {
+        resolve({ text: fullText, usage })
+      }
+    })
+
+    child.stdin.on('error', (error) => finishReject(classifyRunError(String(error))))
+    child.stdin.end(buildCodexPrompt(settings, prompt))
+  })
+}
+
+async function captureCodexCommand(
+  settings: AppSettings,
+  args: string[],
+  timeoutMs = CODEX_COMMAND_TIMEOUT_MS
+): Promise<CapturedCommand> {
+  const command = resolveCodexCommand(settings)
+  const child = spawn(command.program, [...command.prefixArgs, ...args], {
+    cwd: tmpdir(),
+    env: buildCodexEnvironment(settings),
+    stdio: ['ignore', 'pipe', 'pipe'],
+    windowsHide: true
+  })
+
+  return await new Promise<CapturedCommand>((resolve, reject) => {
+    let stdout = ''
+    let stderr = ''
+    let settled = false
+    const timer = setTimeout(() => {
+      child.kill()
+      if (!settled) {
+        settled = true
+        reject(new Error('Codex CLI 响应超时。'))
+      }
+    }, timeoutMs)
+    child.on('error', (error) => {
+      clearTimeout(timer)
+      if (!settled) {
+        settled = true
+        reject(classifySpawnError(error))
+      }
+    })
+    child.stdout?.on('data', (chunk) => { stdout = appendCaptured(stdout, chunk) })
+    child.stderr?.on('data', (chunk) => { stderr = appendCaptured(stderr, chunk) })
+    child.on('close', (exitCode) => {
+      clearTimeout(timer)
+      if (!settled) {
+        settled = true
+        resolve({ stdout, stderr, exitCode })
+      }
+    })
+  })
+}
+
+export async function testCodexCliConnection(settings: AppSettings): Promise<void> {
+  const result = await captureCodexCommand(settings, ['login', 'status'])
+  if (result.exitCode !== 0) throw classifyRunError(result.stderr || result.stdout)
+}
+
+function parseCodexModels(stdout: string): CodexCliModel[] | null {
+  const jsonStart = stdout.indexOf('{')
+  if (jsonStart < 0) return null
+  try {
+    const payload = JSON.parse(stdout.slice(jsonStart)) as { models?: Array<Record<string, unknown>> }
+    if (!Array.isArray(payload.models)) return null
+    const ids = payload.models
+      .map((item) => String(item.slug ?? item.id ?? '').trim())
+      .filter(Boolean)
+    if (ids.length === 0) return null
+    return ['default', ...new Set(ids)].map((id) => ({ id, ownedBy: 'codex-cli' }))
+  } catch {
+    return null
+  }
+}
+
+export async function fetchCodexCliModels(settings: AppSettings): Promise<CodexCliModel[]> {
+  const discovered = await captureCodexCommand(settings, ['debug', 'models'], CODEX_MODEL_DISCOVERY_TIMEOUT_MS)
+  const discoveredModels = discovered.exitCode === 0 ? parseCodexModels(discovered.stdout) : null
+  if (discoveredModels) return discoveredModels
+
+  // 较旧版本或未登录环境可能无法刷新目录，继续读取 CLI 内置目录；再失败时
+  // 保留 DBX 同类实现采用的静态入口，用户仍可手动填写任意模型名。
+  const bundled = await captureCodexCommand(
+    settings,
+    ['debug', 'models', '--bundled'],
+    CODEX_MODEL_DISCOVERY_TIMEOUT_MS
+  )
+  return (bundled.exitCode === 0 ? parseCodexModels(bundled.stdout) : null)
+    ?? DEFAULT_CODEX_MODELS.map((id) => ({ id, ownedBy: 'codex-cli' }))
+}
diff --git a/electron/main/ai/generate.ts b/electron/main/ai/generate.ts
index 0137c3c1..27c66a33 100644
--- a/electron/main/ai/generate.ts
+++ b/electron/main/ai/generate.ts
@@ -5,8 +5,9 @@ import { buildSystemPrompt, createModel, providerSupportsNativeStructuredOutput
 import type { AiRunUsage, AppSettings, AiStreamHandlers, PromptPair } from './shared-types'
 import { stripReasoningMarkup } from './reasoning'
 import { isOpenAIReasoningChatModel, resolveProviderOptions, resolveSamplingOptions } from './request-options'
-import { isAnthropicProtocol, isOpenAIChatProtocol } from '@shared/ai-provider-catalog'
+import { isAnthropicProtocol, isCodexCliProvider, isOpenAIChatProtocol } from '@shared/ai-provider-catalog'
 import { AiStreamProtocolError } from './sse'
+import { runCodexCli } from './codex-cli'
 
 function useStreamFallback(settings: AppSettings): boolean {
   return isAnthropicProtocol(settings.provider, settings.model)
@@ -82,6 +83,9 @@ export async function aiGenerateTextWithUsage(
   signal?: AbortSignal,
   options?: AiGenerateOptions
 ): Promise<AiTextGenerationResult> {
+  if (isCodexCliProvider(settings.provider)) {
+    return runCodexCli(settings, prompt, { signal })
+  }
   const system = buildSystemPrompt(settings, prompt.system)
   const canUseNativeStructuredOutput = providerSupportsNativeStructuredOutput(settings)
   const providerOptions = resolveProviderOptions(settings, options)
@@ -195,6 +199,9 @@ export async function aiStreamTextWithUsage(
   signal: AbortSignal,
   maxTokens?: number
 ): Promise<AiTextGenerationResult> {
+  if (isCodexCliProvider(settings.provider)) {
+    return runCodexCli(settings, prompt, { signal, handlers })
+  }
   const providerOptions = resolveProviderOptions(settings, { preferLowReasoning: true })
   const samplingOptions = resolveSamplingOptions(settings)
   let streamError: unknown = null
diff --git a/electron/main/ai/ipc.ts b/electron/main/ai/ipc.ts
index 79cbda09..cf54f2bf 100644
--- a/electron/main/ai/ipc.ts
+++ b/electron/main/ai/ipc.ts
@@ -18,7 +18,7 @@ import type { SpiralBootstrapInput } from './spiral'
 import { formatAiErrorMessage } from './error-message'
 import { testProxyConnection } from './proxy-fetch'
 import { normalizeSettings } from './settings'
-import { shouldTryStreamingAgent } from '@shared/ai-provider-catalog'
+import { isCodexCliProvider, shouldTryStreamingAgent } from '@shared/ai-provider-catalog'
 
 /**
  * AI IPC 模块的外部依赖注入接口。
@@ -289,58 +289,61 @@ export function registerAiIpcHandlers(injectedDeps: AiIpcDeps): void {
       let streamedContent = ''
       void (async () => {
         try {
-          try {
-            const result = await runStreamingAgentTask(
-              payload,
-              {
-                onTextDelta: (delta) => {
-                  streamedContent += delta
-                  if (!event.sender.isDestroyed()) {
-                    event.sender.send('characterarc:ai-stream-event', { streamId, type: 'chunk', delta, charCount: streamedContent.length })
-                  }
-                },
-                onReasoningDelta: (delta) => {
-                  if (!event.sender.isDestroyed()) {
-                    event.sender.send('characterarc:ai-stream-event', { streamId, type: 'reasoning', delta })
-                  }
-                },
-                onToolUseStart: (toolUseId, toolName, args) => {
-                  if (!event.sender.isDestroyed()) {
-                    event.sender.send('characterarc:ai-stream-event', { streamId, type: 'tool_use_start', toolUseId, toolName, args })
-                  }
-                },
-                onToolResult: (toolUseId, toolName, content, isError, durationMs) => {
-                  if (!event.sender.isDestroyed()) {
-                    event.sender.send('characterarc:ai-stream-event', { streamId, type: 'tool_result', toolUseId, toolName, content, isError, durationMs })
-                  }
-                },
-                onAgentStatus: (message, iteration, maxIterations) => {
-                  if (!event.sender.isDestroyed()) {
-                    event.sender.send('characterarc:ai-stream-event', { streamId, type: 'agent_status', message, iteration, maxIterations })
-                  }
-                },
-                onEditApplied: (chapterId, editType, preview, versionId) => {
-                  if (!event.sender.isDestroyed()) {
-                    event.sender.send('characterarc:ai-stream-event', { streamId, type: 'edit_applied', chapterId, editType, preview, versionId })
+          const normalizedSettings = normalizeSettings(payload.settings)
+          if (!isCodexCliProvider(normalizedSettings.provider)) {
+            try {
+              const result = await runStreamingAgentTask(
+                payload,
+                {
+                  onTextDelta: (delta) => {
+                    streamedContent += delta
+                    if (!event.sender.isDestroyed()) {
+                      event.sender.send('characterarc:ai-stream-event', { streamId, type: 'chunk', delta, charCount: streamedContent.length })
+                    }
+                  },
+                  onReasoningDelta: (delta) => {
+                    if (!event.sender.isDestroyed()) {
+                      event.sender.send('characterarc:ai-stream-event', { streamId, type: 'reasoning', delta })
+                    }
+                  },
+                  onToolUseStart: (toolUseId, toolName, args) => {
+                    if (!event.sender.isDestroyed()) {
+                      event.sender.send('characterarc:ai-stream-event', { streamId, type: 'tool_use_start', toolUseId, toolName, args })
+                    }
+                  },
+                  onToolResult: (toolUseId, toolName, content, isError, durationMs) => {
+                    if (!event.sender.isDestroyed()) {
+                      event.sender.send('characterarc:ai-stream-event', { streamId, type: 'tool_result', toolUseId, toolName, content, isError, durationMs })
+                    }
+                  },
+                  onAgentStatus: (message, iteration, maxIterations) => {
+                    if (!event.sender.isDestroyed()) {
+                      event.sender.send('characterarc:ai-stream-event', { streamId, type: 'agent_status', message, iteration, maxIterations })
+                    }
+                  },
+                  onEditApplied: (chapterId, editType, preview, versionId) => {
+                    if (!event.sender.isDestroyed()) {
+                      event.sender.send('characterarc:ai-stream-event', { streamId, type: 'edit_applied', chapterId, editType, preview, versionId })
+                    }
+                  },
+                  onEditProposed: (chapterId, proposalId, editType, preview, oldContent, newContent) => {
+                    if (!event.sender.isDestroyed()) {
+                      event.sender.send('characterarc:ai-stream-event', { streamId, type: 'edit_proposed', chapterId, proposalId, editType, preview, oldContent, newContent })
+                    }
                   }
                 },
-                onEditProposed: (chapterId, proposalId, editType, preview, oldContent, newContent) => {
-                  if (!event.sender.isDestroyed()) {
-                    event.sender.send('characterarc:ai-stream-event', { streamId, type: 'edit_proposed', chapterId, proposalId, editType, preview, oldContent, newContent })
-                  }
-                }
-              },
-              controller.signal,
-              knowledgeContext
-            )
-            deps!.emitAiRunEvent({ projectId: result.meta.projectId ?? '', meta: { id: randomUUID(), ...result.meta } })
-            if (!event.sender.isDestroyed()) {
-              event.sender.send('characterarc:ai-stream-event', { streamId, type: 'done', content: streamedContent, result: result.result })
+                controller.signal,
+                knowledgeContext
+              )
+              deps!.emitAiRunEvent({ projectId: result.meta.projectId ?? '', meta: { id: randomUUID(), ...result.meta } })
+              if (!event.sender.isDestroyed()) {
+                event.sender.send('characterarc:ai-stream-event', { streamId, type: 'done', content: streamedContent, result: result.result })
+              }
+              return
+            } catch (agentError) {
+              if (!isToolUseNotSupportedError(agentError)) throw agentError
+              streamedContent = ''
             }
-            return
-          } catch (agentError) {
-            if (!isToolUseNotSupportedError(agentError)) throw agentError
-            streamedContent = ''
           }
 
           const result = await streamAiTask(
diff --git a/electron/main/ai/provider.ts b/electron/main/ai/provider.ts
index 049d94ed..091328a6 100644
--- a/electron/main/ai/provider.ts
+++ b/electron/main/ai/provider.ts
@@ -1,6 +1,6 @@
 import type { LanguageModel } from 'ai'
 import type { AppSettings } from './shared-types'
-import { isAnthropicProtocol, resolveAiProviderProtocol } from '@shared/ai-provider-catalog'
+import { isAnthropicProtocol, isCodexCliProvider, resolveAiProviderProtocol } from '@shared/ai-provider-catalog'
 import { createProxyFetch } from './proxy-fetch'
 import { createProviderTransportFetch } from './sse'
 import { createProtocolModel } from './protocol-adapter'
@@ -15,6 +15,7 @@ function isOfficialOpenAIProvider(settings: AppSettings): boolean {
 }
 
 export function providerSupportsNativeStructuredOutput(settings: AppSettings): boolean {
+  if (isCodexCliProvider(settings.provider)) return false
   // Anthropic's SDK object streaming can produce an empty text stream or fail
   // object parsing for otherwise recoverable JSON tasks. Keep Claude JSON tasks
   // on the text path and let task normalizers/repair prompts handle the JSON.
@@ -26,6 +27,9 @@ export function createModel(
   settings: AppSettings,
   options?: { requestFetch?: typeof fetch }
 ): LanguageModel {
+  if (isCodexCliProvider(settings.provider)) {
+    throw new Error('Codex CLI 由本机子进程传输层调用，不能创建 HTTP 模型。')
+  }
   const requestFetch = options?.requestFetch ?? createProxyFetch(settings.proxyUrl)
   const customFetch = createProviderTransportFetch(requestFetch)
   return createProtocolModel({
diff --git a/electron/main/ai/runtime/orchestrator.ts b/electron/main/ai/runtime/orchestrator.ts
index d8439849..9ca9c4b1 100644
--- a/electron/main/ai/runtime/orchestrator.ts
+++ b/electron/main/ai/runtime/orchestrator.ts
@@ -33,6 +33,7 @@ import { formatAiErrorMessage } from '../error-message'
 import {
   isOpenCodeProvider,
   isOpenAIChatProtocol,
+  isCodexCliProvider,
   resolveAiProviderProtocol
 } from '@shared/ai-provider-catalog'
 import { createHash, randomUUID } from 'node:crypto'
@@ -43,6 +44,7 @@ import {
   finishChapterProcessing,
   type ChapterProcessingStageStatus
 } from './chapter-processing-store'
+import { testCodexCliConnection } from '../codex-cli'
 
 const postGenerationTasks = new BackgroundTaskCoordinator()
 
@@ -88,7 +90,10 @@ export async function runAiTask(
   // 白名单内的任务直接尝试走 agent loop，不预判 provider 能力。
   // 如果模型不支持 tool_use，运行时会抛错，在 catch 中降级或提示用户。
   const settingsForRouting = normalizeSettings(task.settings)
+  // Codex CLI 自己运行在独立进程中，无法直接执行 CharacterArc 的进程内 tools；
+  // 此模式仍使用相同的完整任务 Prompt，并走单次/流式文本链路。
   const usesAgentRoute = AGENT_TASK_WHITELIST.has(task.task)
+    && !isCodexCliProvider(settingsForRouting.provider)
   if (usesAgentRoute) {
     await enrichTaskContextForGeneration(task, settingsForRouting)
     try {
@@ -420,6 +425,14 @@ export async function testAiConnection(rawSettings: AppSettings): Promise<{
 }> {
   const settings = normalizeSettings(rawSettings)
   validateSettings(settings)
+  if (isCodexCliProvider(settings.provider)) {
+    await testCodexCliConnection(settings)
+    return {
+      provider: settings.provider,
+      model: settings.model,
+      protocol: 'codex-cli'
+    }
+  }
   const probePrompt = {
     system: 'You are a connectivity probe. Reply briefly to confirm the request succeeded.',
     user: 'Reply with CONNECTED.'
diff --git a/electron/main/ai/settings.ts b/electron/main/ai/settings.ts
index fe151965..184e49ed 100644
--- a/electron/main/ai/settings.ts
+++ b/electron/main/ai/settings.ts
@@ -3,7 +3,8 @@ import {
   getAiProviderCatalogEntry,
   normalizeAiProtocolPreference,
   normalizeAiBaseUrl,
-  normalizeAiProviderName
+  normalizeAiProviderName,
+  isCodexCliProvider
 } from '@shared/ai-provider-catalog'
 
 /**
@@ -25,6 +26,13 @@ function normalizeOptionalNumber(value: unknown, min: number, max: number): numb
   return Math.min(max, Math.max(min, value))
 }
 
+function normalizeCodexReasoningEffort(value: unknown): AppSettings['codexReasoningEffort'] {
+  return typeof value === 'string'
+    && ['default', 'minimal', 'low', 'medium', 'high', 'xhigh', 'max', 'ultra'].includes(value)
+    ? value as AppSettings['codexReasoningEffort']
+    : 'default'
+}
+
 /**
  * 规范化用户设置：trim、转小写、缺失字段回退到供应商默认值。
  *
@@ -42,6 +50,8 @@ export function normalizeSettings(settings: AppSettings): AppSettings {
     apiKey: settings.apiKey?.trim() || '',
     baseUrl,
     apiProtocol: normalizeAiProtocolPreference(settings.apiProtocol),
+    codexCliPath: settings.codexCliPath?.trim() || '',
+    codexReasoningEffort: normalizeCodexReasoningEffort(settings.codexReasoningEffort),
     proxyUrl: settings.proxyUrl?.trim() || '',
     temperature: normalizeOptionalNumber(settings.temperature, 0, 2),
     topP: normalizeOptionalNumber(settings.topP, 0, 1),
@@ -74,7 +84,7 @@ export function isLocalBaseUrl(baseUrl: string): boolean {
  * @returns 是否需要填写 API Key
  */
 export function requiresApiKey(settings: AppSettings): boolean {
-  if (settings.provider === 'ollama') {
+  if (settings.provider === 'ollama' || isCodexCliProvider(settings.provider)) {
     return false
   }
   return !isLocalBaseUrl(settings.baseUrl)
@@ -90,7 +100,7 @@ export function validateSettings(settings: AppSettings): void {
   if (!settings.model.trim()) {
     throw new Error('请先填写模型名称。')
   }
-  if (!settings.baseUrl.trim()) {
+  if (!isCodexCliProvider(settings.provider) && !settings.baseUrl.trim()) {
     throw new Error('请先填写 Base URL。')
   }
   if (requiresApiKey(settings) && !settings.apiKey.trim()) {
diff --git a/electron/main/ai/shared-types.ts b/electron/main/ai/shared-types.ts
index a8bb0ea0..85d4392a 100644
--- a/electron/main/ai/shared-types.ts
+++ b/electron/main/ai/shared-types.ts
@@ -30,6 +30,10 @@ export type AppSettings = {
   baseUrl: string
   /** API 线协议；auto 时按供应商和模型目录解析 */
   apiProtocol?: 'auto' | 'openai-responses' | 'openai-chat' | 'anthropic'
+  /** Codex CLI 可执行文件或其所在目录；留空时从 PATH 和常见目录查找 */
+  codexCliPath?: string
+  /** Codex CLI 的 model_reasoning_effort；default 时沿用 Codex 配置 */
+  codexReasoningEffort?: 'default' | 'minimal' | 'low' | 'medium' | 'high' | 'xhigh' | 'max' | 'ultra'
   /** 可选：AI 请求使用的 HTTP/HTTPS 代理地址 */
   proxyUrl?: string
   /** 可选：模型采样温度，留空时使用服务端默认值 */
diff --git a/electron/main/ai/transport/models.ts b/electron/main/ai/transport/models.ts
index 57d3c24d..b22f726f 100644
--- a/electron/main/ai/transport/models.ts
+++ b/electron/main/ai/transport/models.ts
@@ -3,9 +3,11 @@ import { normalizeSettings } from '../settings'
 import { createProxyFetch } from '../proxy-fetch'
 import {
   isAnthropicProtocol,
+  isCodexCliProvider,
   isOpenCodeProvider,
   isSupportedProviderModel
 } from '@shared/ai-provider-catalog'
+import { fetchCodexCliModels } from '../codex-cli'
 
 /** 从模型列表接口获取到的模型信息 */
 export interface FetchedModel {
@@ -137,6 +139,9 @@ async function fetchModelsAnthropic(baseUrl: string, apiKey: string, requestFetc
  */
 export async function fetchModels(settings: AppSettings): Promise<FetchedModel[]> {
   const normalized = normalizeSettings(settings)
+  if (isCodexCliProvider(normalized.provider)) {
+    return fetchCodexCliModels(normalized)
+  }
   if (!normalized.baseUrl.trim()) throw new Error('请先填写 Base URL。')
   if (!normalized.apiKey.trim() && normalized.provider !== 'ollama') {
     throw new Error('需要 API Key 才能获取模型列表。')
diff --git a/electron/main/workspace-types.test.mjs b/electron/main/workspace-types.test.mjs
index 603f9680..e94d6f07 100644
--- a/electron/main/workspace-types.test.mjs
+++ b/electron/main/workspace-types.test.mjs
@@ -110,6 +110,57 @@ test('旧 AI 设置和配置档案缺少协议时迁移为 auto', () => {
   assert.equal(merged.appSettings.aiProfiles[0].apiProtocol, 'auto')
 })
 
+test('Codex CLI 路径和推理强度随配置档案持久化', () => {
+  const normalized = normalizeAppSettings({
+    provider: 'codex-cli',
+    model: 'default',
+    codexCliPath: ' /opt/homebrew/bin/codex ',
+    codexReasoningEffort: 'high',
+    activeAiProfileId: 'profile-codex',
+    aiProfiles: [{
+      id: 'profile-codex',
+      name: 'Codex CLI',
+      provider: 'codex-cli',
+      baseUrl: '',
+      apiKey: '',
+      model: 'default',
+      codexCliPath: ' /opt/homebrew/bin/codex ',
+      codexReasoningEffort: 'xhigh'
+    }]
+  })
+
+  assert.equal(normalized.codexCliPath, '/opt/homebrew/bin/codex')
+  assert.equal(normalized.codexReasoningEffort, 'xhigh')
+  assert.equal(normalized.aiProfiles[0].codexCliPath, '/opt/homebrew/bin/codex')
+  assert.equal(normalized.aiProfiles[0].codexReasoningEffort, 'xhigh')
+})
+
+test('活动配置缺失时自动激活首个档案并同步运行模型', () => {
+  const normalized = normalizeAppSettings({
+    provider: 'openai-compatible',
+    model: '',
+    apiKey: '',
+    baseUrl: '',
+    activeAiProfileId: '',
+    aiProfiles: [{
+      id: 'profile-codex',
+      name: '我的Codex',
+      provider: 'codex-cli',
+      baseUrl: '',
+      apiKey: '',
+      model: 'gpt-5.6-luna',
+      codexCliPath: '/opt/homebrew/bin/codex',
+      codexReasoningEffort: 'high'
+    }]
+  })
+
+  assert.equal(normalized.activeAiProfileId, 'profile-codex')
+  assert.equal(normalized.provider, 'codex-cli')
+  assert.equal(normalized.model, 'gpt-5.6-luna')
+  assert.equal(normalized.codexCliPath, '/opt/homebrew/bin/codex')
+  assert.equal(normalized.codexReasoningEffort, 'high')
+})
+
 test('工作台菜单顺序按应用设置保存并清理重复值', () => {
   const normalized = normalizeWorkspacePayload({
     theme: 'ocean',
diff --git a/electron/main/workspace-types.ts b/electron/main/workspace-types.ts
index aef2f048..df5df766 100644
--- a/electron/main/workspace-types.ts
+++ b/electron/main/workspace-types.ts
@@ -318,6 +318,8 @@ export type WorkspacePayload = {
     apiKey: string
     baseUrl: string
     apiProtocol?: 'auto' | 'openai-responses' | 'openai-chat' | 'anthropic'
+    codexCliPath?: string
+    codexReasoningEffort?: 'default' | 'minimal' | 'low' | 'medium' | 'high' | 'xhigh' | 'max' | 'ultra'
     proxyUrl: string
     temperature?: number
     topP?: number
@@ -331,6 +333,8 @@ export type WorkspacePayload = {
       apiKey: string
       model: string
       apiProtocol?: 'auto' | 'openai-responses' | 'openai-chat' | 'anthropic'
+      codexCliPath?: string
+      codexReasoningEffort?: 'default' | 'minimal' | 'low' | 'medium' | 'high' | 'xhigh' | 'max' | 'ultra'
       temperature?: number
       topP?: number
       presencePenalty?: number
@@ -477,6 +481,15 @@ function normalizeApiProtocol(
     : 'auto'
 }
 
+function normalizeCodexReasoningEffort(
+  value: unknown
+): 'default' | 'minimal' | 'low' | 'medium' | 'high' | 'xhigh' | 'max' | 'ultra' {
+  return typeof value === 'string'
+    && ['default', 'minimal', 'low', 'medium', 'high', 'xhigh', 'max', 'ultra'].includes(value)
+    ? value as 'default' | 'minimal' | 'low' | 'medium' | 'high' | 'xhigh' | 'max' | 'ultra'
+    : 'default'
+}
+
 export function normalizeAppSettings(
   settings?: Partial<WorkspacePayload['appSettings']> | null
 ): WorkspacePayload['appSettings'] {
@@ -501,48 +514,63 @@ export function normalizeAppSettings(
       ? Math.min(2, Math.max(-2, settings.frequencyPenalty))
       : undefined
 
+  const aiProfiles = Array.isArray(settings?.aiProfiles)
+    ? settings.aiProfiles
+        .filter((item): item is NonNullable<typeof settings.aiProfiles>[number] => !!item && typeof item === 'object')
+        .map((item) => ({
+          id: String(item.id ?? '').trim(),
+          name: String(item.name ?? '').trim(),
+          provider: String(item.provider ?? '').trim(),
+          baseUrl: String(item.baseUrl ?? '').trim(),
+          apiKey: String(item.apiKey ?? '').trim(),
+          model: String(item.model ?? '').trim(),
+          apiProtocol: normalizeApiProtocol(item.apiProtocol),
+          codexCliPath: typeof item.codexCliPath === 'string' ? item.codexCliPath.trim() : '',
+          codexReasoningEffort: normalizeCodexReasoningEffort(item.codexReasoningEffort),
+          temperature:
+            typeof item.temperature === 'number' && Number.isFinite(item.temperature)
+              ? Math.min(2, Math.max(0, item.temperature))
+              : undefined,
+          topP:
+            typeof item.topP === 'number' && Number.isFinite(item.topP)
+              ? Math.min(1, Math.max(0, item.topP))
+              : undefined,
+          presencePenalty:
+            typeof item.presencePenalty === 'number' && Number.isFinite(item.presencePenalty)
+              ? Math.min(2, Math.max(-2, item.presencePenalty))
+              : undefined,
+          frequencyPenalty:
+            typeof item.frequencyPenalty === 'number' && Number.isFinite(item.frequencyPenalty)
+              ? Math.min(2, Math.max(-2, item.frequencyPenalty))
+              : undefined
+        }))
+        .filter((item) => item.id)
+    : []
+  const requestedActiveProfileId = typeof settings?.activeAiProfileId === 'string'
+    ? settings.activeAiProfileId.trim()
+    : ''
+  const activeAiProfileId = aiProfiles.some((item) => item.id === requestedActiveProfileId)
+    ? requestedActiveProfileId
+    : aiProfiles[0]?.id ?? ''
+  const activeProfile = aiProfiles.find((item) => item.id === activeAiProfileId)
+
   return {
-    provider: settings?.provider || 'openai-compatible',
-    model: settings?.model || '',
-    apiKey: settings?.apiKey || '',
-    baseUrl: settings?.baseUrl || '',
-    apiProtocol: normalizeApiProtocol(settings?.apiProtocol),
+    provider: activeProfile?.provider || settings?.provider || 'openai-compatible',
+    model: activeProfile?.model ?? settings?.model ?? '',
+    apiKey: activeProfile?.apiKey ?? settings?.apiKey ?? '',
+    baseUrl: activeProfile?.baseUrl ?? settings?.baseUrl ?? '',
+    apiProtocol: activeProfile?.apiProtocol ?? normalizeApiProtocol(settings?.apiProtocol),
+    codexCliPath: activeProfile?.codexCliPath
+      ?? (typeof settings?.codexCliPath === 'string' ? settings.codexCliPath.trim() : ''),
+    codexReasoningEffort: activeProfile?.codexReasoningEffort
+      ?? normalizeCodexReasoningEffort(settings?.codexReasoningEffort),
     proxyUrl: settings?.proxyUrl || '',
-    temperature,
-    topP,
-    presencePenalty,
-    frequencyPenalty,
-    aiProfiles: Array.isArray(settings?.aiProfiles)
-      ? settings.aiProfiles
-          .filter((item): item is NonNullable<typeof settings.aiProfiles>[number] => !!item && typeof item === 'object')
-          .map((item) => ({
-            id: String(item.id ?? '').trim(),
-            name: String(item.name ?? '').trim(),
-            provider: String(item.provider ?? '').trim(),
-            baseUrl: String(item.baseUrl ?? '').trim(),
-            apiKey: String(item.apiKey ?? '').trim(),
-            model: String(item.model ?? '').trim(),
-            apiProtocol: normalizeApiProtocol(item.apiProtocol),
-            temperature:
-              typeof item.temperature === 'number' && Number.isFinite(item.temperature)
-                ? Math.min(2, Math.max(0, item.temperature))
-                : undefined,
-            topP:
-              typeof item.topP === 'number' && Number.isFinite(item.topP)
-                ? Math.min(1, Math.max(0, item.topP))
-                : undefined,
-            presencePenalty:
-              typeof item.presencePenalty === 'number' && Number.isFinite(item.presencePenalty)
-                ? Math.min(2, Math.max(-2, item.presencePenalty))
-                : undefined,
-            frequencyPenalty:
-              typeof item.frequencyPenalty === 'number' && Number.isFinite(item.frequencyPenalty)
-                ? Math.min(2, Math.max(-2, item.frequencyPenalty))
-                : undefined
-          }))
-          .filter((item) => item.id)
-      : [],
-    activeAiProfileId: typeof settings?.activeAiProfileId === 'string' ? settings.activeAiProfileId : '',
+    temperature: activeProfile?.temperature ?? temperature,
+    topP: activeProfile?.topP ?? topP,
+    presencePenalty: activeProfile?.presencePenalty ?? presencePenalty,
+    frequencyPenalty: activeProfile?.frequencyPenalty ?? frequencyPenalty,
+    aiProfiles,
+    activeAiProfileId,
     imageProvider: settings?.imageProvider || '',
     imageModel: settings?.imageModel || '',
     imageApiKey: settings?.imageApiKey || '',
diff --git a/electron/shared/ai-provider-catalog.ts b/electron/shared/ai-provider-catalog.ts
index 6026d234..dc210bcc 100644
--- a/electron/shared/ai-provider-catalog.ts
+++ b/electron/shared/ai-provider-catalog.ts
@@ -4,6 +4,7 @@ export type AiProtocolPreference = 'auto' | AiProviderProtocol
 export interface AiProviderCatalogEntry {
   label: string
   value: string
+  transport?: 'http' | 'codex-cli'
   protocol: AiProviderProtocol
   baseUrl: string
   model: string
@@ -13,6 +14,7 @@ export interface AiProviderCatalogEntry {
 }
 
 export const AI_PROVIDER_CATALOG: readonly AiProviderCatalogEntry[] = [
+  { label: 'Codex CLI（本机）', value: 'codex-cli', transport: 'codex-cli', protocol: 'openai-responses', baseUrl: '', model: 'default', customBaseUrl: false, supportsEmbedding: false, hint: '复用本机 Codex CLI 的登录状态，无需 API Key。请先在终端执行 codex login。' },
   { label: 'OpenAI', value: 'openai', protocol: 'openai-responses', baseUrl: 'https://api.openai.com/v1', model: '', customBaseUrl: false, supportsEmbedding: true, hint: '官方接口，填写 API Key 后即可使用，也可拉取账号可用模型。' },
   { label: 'Anthropic', value: 'anthropic', protocol: 'anthropic', baseUrl: 'https://api.anthropic.com/v1', model: '', customBaseUrl: false, supportsEmbedding: false, hint: 'Claude 官方接口，使用 Anthropic Messages 协议。' },
   { label: 'DeepSeek', value: 'deepseek', protocol: 'openai-chat', baseUrl: 'https://api.deepseek.com/v1', model: '', customBaseUrl: false, supportsEmbedding: false, hint: 'DeepSeek 官方接口，支持 deepseek-chat 和 deepseek-reasoner。' },
@@ -38,6 +40,10 @@ export function getAiProviderCatalogEntry(provider: string): AiProviderCatalogEn
   return AI_PROVIDER_CATALOG.find((item) => item.value === normalized)
 }
 
+export function isCodexCliProvider(provider: string): boolean {
+  return provider.trim().toLowerCase() === 'codex-cli'
+}
+
 const KNOWN_ENDPOINT_SUFFIXES = [
   '/chat/completions',
   '/messages',
@@ -149,6 +155,7 @@ export function shouldTryStreamingAgent(
   model: string,
   protocol: AiProtocolPreference = 'auto'
 ): boolean {
+  if (isCodexCliProvider(provider)) return false
   if (task === 'global-assistant') return true
   if (task !== 'chapter-first-draft') return false
 
diff --git a/electron/shared/ipc-types.ts b/electron/shared/ipc-types.ts
index abc4bc6b..342f0bf3 100644
--- a/electron/shared/ipc-types.ts
+++ b/electron/shared/ipc-types.ts
@@ -16,12 +16,14 @@ export interface AppSettingsPayload {
   apiKey: string
   baseUrl: string
   apiProtocol?: 'auto' | 'openai-responses' | 'openai-chat' | 'anthropic'
+  codexCliPath?: string
+  codexReasoningEffort?: 'default' | 'minimal' | 'low' | 'medium' | 'high' | 'xhigh' | 'max' | 'ultra'
   proxyUrl: string
   temperature?: number
   topP?: number
   presencePenalty?: number
   frequencyPenalty?: number
-  aiProfiles: Array<{ id: string; name: string; provider: string; baseUrl: string; apiKey: string; model: string; apiProtocol?: 'auto' | 'openai-responses' | 'openai-chat' | 'anthropic'; temperature?: number; topP?: number; presencePenalty?: number; frequencyPenalty?: number }>
+  aiProfiles: Array<{ id: string; name: string; provider: string; baseUrl: string; apiKey: string; model: string; apiProtocol?: 'auto' | 'openai-responses' | 'openai-chat' | 'anthropic'; codexCliPath?: string; codexReasoningEffort?: 'default' | 'minimal' | 'low' | 'medium' | 'high' | 'xhigh' | 'max' | 'ultra'; temperature?: number; topP?: number; presencePenalty?: number; frequencyPenalty?: number }>
   activeAiProfileId: string
   imageProvider: string
   imageModel: string
diff --git a/package.json b/package.json
index 46737405..5779da50 100644
--- a/package.json
+++ b/package.json
@@ -8,7 +8,7 @@
   "license": "MIT",
   "scripts": {
     "dev": "electron-vite dev",
-    "test": "set ELECTRON_RUN_AS_NODE=1&& electron --test electron/shared/continuation-import.test.mjs electron/main/workspace-types.test.mjs renderer/src/features/workspace/outlineReorder.test.mjs renderer/src/features/workspace/volumeCollapseState.test.mjs renderer/src/features/workspace/workbenchMenu.test.mjs renderer/src/features/chapters/chapterFormatting.test.mjs renderer/src/features/ai/taskRegistry.test.mjs renderer/src/features/ai/outlineReferences.test.mjs renderer/src/features/tutorials/tutorials.test.mjs renderer/src/features/knowledge/knowledgeCenter.test.mjs electron/main/story-state-store.test.mjs electron/main/knowledge-document-schema.test.mjs electron/main/ai/settings.test.mjs electron/main/ai/generate.test.mjs electron/main/ai/provider.test.mjs electron/main/ai/protocol-adapter.test.mjs electron/main/ai/proxy-fetch.test.mjs electron/main/ai/knowledge-retrieval.test.mjs electron/main/ai/state-backfill-store.test.mjs electron/main/ai/state-backfill-task-controller.test.mjs electron/main/ai/runtime/background-task-coordinator.test.mjs electron/main/ai/runtime/chapter-processing-store.test.mjs electron/main/ai/tasks/chapter-session-note.test.mjs electron/main/ai/tasks/outline-context.test.mjs electron/main/ai/tasks/story-state-generation.test.mjs electron/main/ai/tasks/worldview-type.test.mjs electron/main/ai/tasks/ranking-analysis.test.mjs electron/main/ai/tasks/qimao-ranking.test.mjs electron/main/ai/tasks/zongheng-ranking.test.mjs electron/main/ai/runtime-v2/conversation-manager.test.mjs electron/main/ai/runtime-v2/staged-changes-store.test.mjs electron/main/ai/runtime-v2/commit-result.test.mjs electron/main/ai/runtime-v2/context-builder.test.mjs electron/main/ai/runtime-v2/agent-loop.test.mjs electron/main/ai/agent/tools/text-window.test.mjs electron/main/ai/agent/tools/knowledge-tools.test.mjs electron/main/ai/agent/tools/chapter-data-access.test.mjs electron/main/ai/agent/tools/chapter-html-edit.test.mjs electron/main/ai/runtime-v2/tools/stage-chapter-edit-core.test.mjs",
+    "test": "set ELECTRON_RUN_AS_NODE=1&& electron --test electron/shared/continuation-import.test.mjs electron/main/workspace-types.test.mjs renderer/src/features/workspace/outlineReorder.test.mjs renderer/src/features/workspace/volumeCollapseState.test.mjs renderer/src/features/workspace/workbenchMenu.test.mjs renderer/src/features/chapters/chapterFormatting.test.mjs renderer/src/features/ai/taskRegistry.test.mjs renderer/src/features/ai/outlineReferences.test.mjs renderer/src/features/tutorials/tutorials.test.mjs renderer/src/features/knowledge/knowledgeCenter.test.mjs electron/main/story-state-store.test.mjs electron/main/knowledge-document-schema.test.mjs electron/main/ai/settings.test.mjs electron/main/ai/codex-cli.test.mjs electron/main/ai/generate.test.mjs electron/main/ai/provider.test.mjs electron/main/ai/protocol-adapter.test.mjs electron/main/ai/proxy-fetch.test.mjs electron/main/ai/knowledge-retrieval.test.mjs electron/main/ai/state-backfill-store.test.mjs electron/main/ai/state-backfill-task-controller.test.mjs electron/main/ai/runtime/background-task-coordinator.test.mjs electron/main/ai/runtime/chapter-processing-store.test.mjs electron/main/ai/tasks/chapter-session-note.test.mjs electron/main/ai/tasks/outline-context.test.mjs electron/main/ai/tasks/story-state-generation.test.mjs electron/main/ai/tasks/worldview-type.test.mjs electron/main/ai/tasks/ranking-analysis.test.mjs electron/main/ai/tasks/qimao-ranking.test.mjs electron/main/ai/tasks/zongheng-ranking.test.mjs electron/main/ai/runtime-v2/conversation-manager.test.mjs electron/main/ai/runtime-v2/staged-changes-store.test.mjs electron/main/ai/runtime-v2/commit-result.test.mjs electron/main/ai/runtime-v2/context-builder.test.mjs electron/main/ai/runtime-v2/agent-loop.test.mjs electron/main/ai/agent/tools/text-window.test.mjs electron/main/ai/agent/tools/knowledge-tools.test.mjs electron/main/ai/agent/tools/chapter-data-access.test.mjs electron/main/ai/agent/tools/chapter-html-edit.test.mjs electron/main/ai/runtime-v2/tools/stage-chapter-edit-core.test.mjs",
     "test:ranking": "set ELECTRON_RUN_AS_NODE=1&& electron --test electron/main/ai/tasks/ranking-analysis.test.mjs",
     "typecheck": "vue-tsc --noEmit",
     "build": "pnpm run typecheck && electron-vite build",
diff --git a/renderer/src/components/home/HomepageSettingsModal.vue b/renderer/src/components/home/HomepageSettingsModal.vue
index 9b645822..01ba3001 100644
--- a/renderer/src/components/home/HomepageSettingsModal.vue
+++ b/renderer/src/components/home/HomepageSettingsModal.vue
@@ -18,7 +18,7 @@ import { useAppStore } from '@/stores/app'
 import { darkModePresets, themePresets } from '@/theme/presets'
 import { toIpcPayload } from '@/utils/ipcPayload'
 import type { AiProfile, AppSettings, DarkModeStyle, ThemeName } from '@/types/app'
-import { isOpenAIChatProtocol, resolveAiProviderProtocol } from '@shared/ai-provider-catalog'
+import { isCodexCliProvider, isOpenAIChatProtocol, resolveAiProviderProtocol } from '@shared/ai-provider-catalog'
 
 const props = defineProps<{
   show: boolean
@@ -53,6 +53,16 @@ const apiProtocolOptions = [
   { label: 'Anthropic Messages', value: 'anthropic' },
   { label: 'OpenAI Chat Completions', value: 'openai-chat' }
 ]
+const codexReasoningEffortOptions = [
+  { label: '默认（沿用 Codex 配置）', value: 'default' },
+  { label: 'Minimal', value: 'minimal' },
+  { label: 'Low', value: 'low' },
+  { label: 'Medium', value: 'medium' },
+  { label: 'High', value: 'high' },
+  { label: 'XHigh', value: 'xhigh' },
+  { label: 'Max', value: 'max' },
+  { label: 'Ultra', value: 'ultra' }
+]
 
 const draftSettings = reactive<AppSettings>({
   provider: '',
@@ -60,6 +70,8 @@ const draftSettings = reactive<AppSettings>({
   apiKey: '',
   baseUrl: '',
   apiProtocol: 'auto',
+  codexCliPath: '',
+  codexReasoningEffort: 'default',
   proxyUrl: '',
   temperature: undefined,
   topP: undefined,
@@ -132,6 +144,7 @@ function handleScroll(): void {
 }
 
 const activeProviderPreset = computed(() => getProviderPreset(editingProfile.value?.provider ?? draftSettings.provider))
+const isCodexCli = computed(() => isCodexCliProvider(editingProfile.value?.provider ?? draftSettings.provider))
 const activeApiProtocol = computed(() => resolveAiProviderProtocol(
   editingProfile.value?.provider ?? draftSettings.provider,
   editingProfile.value?.model ?? draftSettings.model,
@@ -152,6 +165,12 @@ const modelSelectOptions = computed(() =>
 const imageModelSelectOptions = computed(() =>
   fetchedImageModels.value.map((m) => ({ label: m.id, value: m.id }))
 )
+const canFetchModels = computed(() => {
+  const profile = editingProfile.value
+  if (!profile || isFetchingModels.value) return false
+  if (isCodexCliProvider(profile.provider)) return true
+  return Boolean(profile.baseUrl.trim() && (profile.provider === 'ollama' || profile.apiKey.trim()))
+})
 const hasPendingChanges = computed(() =>
   draftTheme.value !== appStore.theme
   || JSON.stringify(draftSettings.aiProfiles) !== JSON.stringify(appStore.appSettings.aiProfiles)
@@ -175,6 +194,8 @@ function syncDraftFromStore(): void {
   draftSettings.apiKey = appStore.appSettings.apiKey
   draftSettings.baseUrl = appStore.appSettings.baseUrl
   draftSettings.apiProtocol = appStore.appSettings.apiProtocol ?? 'auto'
+  draftSettings.codexCliPath = appStore.appSettings.codexCliPath
+  draftSettings.codexReasoningEffort = appStore.appSettings.codexReasoningEffort
   draftSettings.proxyUrl = appStore.appSettings.proxyUrl
   proxyTestIp.value = ''
   draftSettings.temperature = appStore.appSettings.temperature
@@ -326,6 +347,8 @@ function handleAddProfile(): void {
     apiKey: '',
     model: defaults.model,
     apiProtocol: 'auto',
+    codexCliPath: '',
+    codexReasoningEffort: 'default',
     temperature: undefined,
     topP: undefined,
     presencePenalty: undefined,
@@ -333,6 +356,9 @@ function handleAddProfile(): void {
   }
   draftSettings.aiProfiles.push(newProfile)
   editingProfileId.value = id
+  if (!draftSettings.activeAiProfileId) {
+    draftSettings.activeAiProfileId = id
+  }
   fetchedModels.value = []
 }
 
@@ -348,6 +374,8 @@ function handleCopyProfile(): void {
     apiKey: source.apiKey,
     model: source.model,
     apiProtocol: source.apiProtocol ?? 'auto',
+    codexCliPath: source.codexCliPath ?? '',
+    codexReasoningEffort: source.codexReasoningEffort ?? 'default',
     temperature: source.temperature,
     topP: source.topP,
     presencePenalty: source.presencePenalty,
@@ -384,6 +412,8 @@ function updateEditingProfile(updates: Partial<AiProfile>): void {
     if (updates.apiKey !== undefined) draftSettings.apiKey = updates.apiKey
     if (updates.baseUrl !== undefined) draftSettings.baseUrl = updates.baseUrl
     if (updates.apiProtocol !== undefined) draftSettings.apiProtocol = updates.apiProtocol
+    if (updates.codexCliPath !== undefined) draftSettings.codexCliPath = updates.codexCliPath
+    if (updates.codexReasoningEffort !== undefined) draftSettings.codexReasoningEffort = updates.codexReasoningEffort
     if ('temperature' in updates) draftSettings.temperature = updates.temperature
     if ('topP' in updates) draftSettings.topP = updates.topP
     if ('presencePenalty' in updates) draftSettings.presencePenalty = updates.presencePenalty
@@ -440,6 +470,8 @@ function buildProfilePayload(): AppSettings {
     apiKey: profile.apiKey,
     baseUrl: profile.baseUrl,
     apiProtocol: profile.apiProtocol ?? 'auto',
+    codexCliPath: profile.codexCliPath ?? '',
+    codexReasoningEffort: profile.codexReasoningEffort ?? 'default',
     temperature: profile.temperature,
     topP: profile.topP,
     presencePenalty: profile.presencePenalty,
@@ -497,6 +529,8 @@ async function saveSettings(): Promise<void> {
     apiKey: activeProfile?.apiKey ?? draftSettings.apiKey,
     baseUrl: activeProfile?.baseUrl ?? draftSettings.baseUrl,
     apiProtocol: activeProfile?.apiProtocol ?? draftSettings.apiProtocol ?? 'auto',
+    codexCliPath: activeProfile?.codexCliPath ?? draftSettings.codexCliPath,
+    codexReasoningEffort: activeProfile?.codexReasoningEffort ?? draftSettings.codexReasoningEffort,
     temperature: activeProfile?.temperature ?? draftSettings.temperature,
     topP: activeProfile?.topP ?? draftSettings.topP,
     presencePenalty: activeProfile?.presencePenalty ?? draftSettings.presencePenalty,
@@ -602,7 +636,7 @@ async function saveSettings(): Promise<void> {
                   @update:value="(value) => handleProviderChange(value ?? 'openai-compatible')"
                 />
               </n-form-item>
-              <n-form-item label="API Key">
+              <n-form-item v-if="!isCodexCli" label="API Key">
                 <n-input
                   type="password"
                   show-password-on="click"
@@ -612,6 +646,16 @@ async function saveSettings(): Promise<void> {
                 />
               </n-form-item>
             </div>
+            <n-form-item v-if="isCodexCli" label="Codex CLI 路径">
+              <div class="preset-field">
+                <n-input
+                  :value="editingProfile.codexCliPath ?? ''"
+                  placeholder="留空自动查找，或填写 codex 可执行文件/所在目录"
+                  @update:value="(value) => updateEditingProfile({ codexCliPath: value })"
+                />
+                <span class="preset-hint">请先在终端执行 codex login；应用代理会自动传给 Codex CLI。</span>
+              </div>
+            </n-form-item>
             <n-form-item v-if="activeProviderPreset.customBaseUrl" label="Base URL">
               <n-input
                 :value="editingProfile.baseUrl"
@@ -634,14 +678,14 @@ async function saveSettings(): Promise<void> {
                   <n-input
                     v-else
                     :value="editingProfile.model"
-                    placeholder="填写 Key 后可点右侧按钮拉取或手动输入"
+                    :placeholder="isCodexCli ? '默认使用 Codex 配置，或拉取/输入模型名称' : '填写 Key 后可点右侧按钮拉取或手动输入'"
                     @update:value="(value) => updateEditingProfile({ model: value })"
                   />
                   <n-button
                     quaternary
                     class="model-fetch-btn"
                     title="获取模型列表"
-                    :disabled="isFetchingModels || !editingProfile.baseUrl.trim() || (editingProfile.provider !== 'ollama' && !editingProfile.apiKey.trim())"
+                    :disabled="!canFetchModels"
                     @click="handleFetchModels"
                   >
                     <template #icon>
@@ -655,7 +699,14 @@ async function saveSettings(): Promise<void> {
             <div class="provider-hint-block">
               <p>{{ activeProviderPreset.hint }}</p>
             </div>
-            <details class="advanced-settings">
+            <n-form-item v-if="isCodexCli" label="推理强度">
+              <n-select
+                :options="codexReasoningEffortOptions"
+                :value="editingProfile.codexReasoningEffort ?? 'default'"
+                @update:value="(value) => updateEditingProfile({ codexReasoningEffort: value ?? 'default' })"
+              />
+            </n-form-item>
+            <details v-else class="advanced-settings">
               <summary>API 高级设置</summary>
               <div class="advanced-settings-body">
                 <n-form-item label="API 协议">
diff --git a/renderer/src/features/workspace/storeHelpers.ts b/renderer/src/features/workspace/storeHelpers.ts
index 855bb0f0..f68df205 100644
--- a/renderer/src/features/workspace/storeHelpers.ts
+++ b/renderer/src/features/workspace/storeHelpers.ts
@@ -236,6 +236,8 @@ export const defaultAppSettings: AppSettings = {
   apiKey: '',
   baseUrl: 'https://api.deepseek.com/v1',
   apiProtocol: 'auto',
+  codexCliPath: '',
+  codexReasoningEffort: 'default',
   proxyUrl: '',
   aiProfiles: [],
   activeAiProfileId: '',
@@ -348,6 +350,12 @@ function normalizeAiProfile(profile: AiProfile): AiProfile {
       || profile.apiProtocol === 'anthropic'
         ? profile.apiProtocol
         : 'auto',
+    codexCliPath: String(profile.codexCliPath ?? '').trim(),
+    codexReasoningEffort:
+      typeof profile.codexReasoningEffort === 'string'
+      && ['default', 'minimal', 'low', 'medium', 'high', 'xhigh', 'max', 'ultra'].includes(profile.codexReasoningEffort)
+        ? profile.codexReasoningEffort
+        : 'default',
     temperature: normalizeOptionalNumber(profile.temperature, 0, 2),
     topP: normalizeOptionalNumber(profile.topP, 0, 1),
     presencePenalty: normalizeOptionalNumber(profile.presencePenalty, -2, 2),
@@ -371,6 +379,12 @@ export function normalizeAppSettings(settings?: Partial<AppSettings> | null): Ap
   const topP = normalizeOptionalNumber(source.topP, 0, 1)
   const presencePenalty = normalizeOptionalNumber(source.presencePenalty, -2, 2)
   const frequencyPenalty = normalizeOptionalNumber(source.frequencyPenalty, -2, 2)
+  const codexCliPath = sanitizeSettingString(source.codexCliPath, '')
+  const codexReasoningEffort =
+    typeof source.codexReasoningEffort === 'string'
+    && ['default', 'minimal', 'low', 'medium', 'high', 'xhigh', 'max', 'ultra'].includes(source.codexReasoningEffort)
+      ? source.codexReasoningEffort
+      : 'default'
 
   let aiProfiles = Array.isArray(source.aiProfiles)
     ? source.aiProfiles.map(normalizeAiProfile).filter((profile) => profile.id)
@@ -379,25 +393,29 @@ export function normalizeAppSettings(settings?: Partial<AppSettings> | null): Ap
 
   if (aiProfiles.length === 0 && (apiKey || model !== defaultAppSettings.model)) {
     const migratedId = `profile-${Date.now()}`
-    aiProfiles = [{ id: migratedId, name: provider || 'Default', provider, baseUrl, apiKey, model, apiProtocol, temperature, topP, presencePenalty, frequencyPenalty }]
+    aiProfiles = [{ id: migratedId, name: provider || 'Default', provider, baseUrl, apiKey, model, apiProtocol, codexCliPath, codexReasoningEffort, temperature, topP, presencePenalty, frequencyPenalty }]
     activeAiProfileId = migratedId
   }
 
-  if (activeAiProfileId && !aiProfiles.find(p => p.id === activeAiProfileId)) {
+  if (!aiProfiles.find(p => p.id === activeAiProfileId)) {
     activeAiProfileId = aiProfiles[0]?.id ?? ''
   }
 
+  const activeProfile = aiProfiles.find(p => p.id === activeAiProfileId)
+
   return {
-    provider,
-    model,
-    apiKey,
-    baseUrl,
-    apiProtocol,
+    provider: activeProfile?.provider ?? provider,
+    model: activeProfile?.model ?? model,
+    apiKey: activeProfile?.apiKey ?? apiKey,
+    baseUrl: activeProfile?.baseUrl ?? baseUrl,
+    apiProtocol: activeProfile?.apiProtocol ?? apiProtocol,
+    codexCliPath: activeProfile?.codexCliPath ?? codexCliPath,
+    codexReasoningEffort: activeProfile?.codexReasoningEffort ?? codexReasoningEffort,
     proxyUrl: sanitizeSettingString(source.proxyUrl, defaultAppSettings.proxyUrl),
-    temperature,
-    topP,
-    presencePenalty,
-    frequencyPenalty,
+    temperature: activeProfile?.temperature ?? temperature,
+    topP: activeProfile?.topP ?? topP,
+    presencePenalty: activeProfile?.presencePenalty ?? presencePenalty,
+    frequencyPenalty: activeProfile?.frequencyPenalty ?? frequencyPenalty,
     aiProfiles,
     activeAiProfileId,
     imageProvider: sanitizeSettingString(source.imageProvider, defaultAppSettings.imageProvider),
diff --git a/renderer/src/pages/ContinuationImportPage.vue b/renderer/src/pages/ContinuationImportPage.vue
index 539c792d..7302f3f2 100644
--- a/renderer/src/pages/ContinuationImportPage.vue
+++ b/renderer/src/pages/ContinuationImportPage.vue
@@ -35,6 +35,7 @@ import {
 
 import { useAppStore } from '@/stores/app'
 import { createDefaultWorkflowDocuments } from '@/features/novelWorkflow/documents'
+import { isCodexCliProvider } from '@shared/ai-provider-catalog'
 import type {
   CharacterCard,
   CharacterRelationship,
@@ -192,6 +193,7 @@ const totalCharacterCount = computed(() => chapters.value.reduce((sum, chapter)
 const volumeCount = computed(() => new Set(chapters.value.map((chapter) => chapter.volumeTitle.trim() || '正文')).size)
 const hasAiSettings = computed(() => {
   const settings = appStore.appSettings
+  if (isCodexCliProvider(settings.provider)) return Boolean(settings.model)
   const isLocal = /^https?:\/\/(?:127\.0\.0\.1|localhost)(?::\d+)?(?:\/|$)/i.test(settings.baseUrl.trim())
   return Boolean(settings.model && settings.baseUrl && (settings.apiKey || settings.provider === 'ollama' || isLocal))
 })
diff --git a/renderer/src/stores/app.ts b/renderer/src/stores/app.ts
index bb1bd159..c8452030 100644
--- a/renderer/src/stores/app.ts
+++ b/renderer/src/stores/app.ts
@@ -2804,6 +2804,8 @@ export const useAppStore = defineStore('app', () => {
     appSettings.value.apiKey = profile.apiKey
     appSettings.value.baseUrl = profile.baseUrl
     appSettings.value.apiProtocol = profile.apiProtocol ?? 'auto'
+    appSettings.value.codexCliPath = profile.codexCliPath ?? ''
+    appSettings.value.codexReasoningEffort = profile.codexReasoningEffort ?? 'default'
     appSettings.value.temperature = profile.temperature
     appSettings.value.topP = profile.topP
     appSettings.value.presencePenalty = profile.presencePenalty
@@ -2839,6 +2841,8 @@ export const useAppStore = defineStore('app', () => {
       if (updates.apiKey !== undefined) appSettings.value.apiKey = updates.apiKey
       if (updates.baseUrl !== undefined) appSettings.value.baseUrl = updates.baseUrl
       if (updates.apiProtocol !== undefined) appSettings.value.apiProtocol = updates.apiProtocol
+      if (updates.codexCliPath !== undefined) appSettings.value.codexCliPath = updates.codexCliPath
+      if (updates.codexReasoningEffort !== undefined) appSettings.value.codexReasoningEffort = updates.codexReasoningEffort
       if ('temperature' in updates) appSettings.value.temperature = updates.temperature
       if ('topP' in updates) appSettings.value.topP = updates.topP
       if ('presencePenalty' in updates) appSettings.value.presencePenalty = updates.presencePenalty
diff --git a/renderer/src/types/app.ts b/renderer/src/types/app.ts
index 6410fe5b..f8752e17 100644
--- a/renderer/src/types/app.ts
+++ b/renderer/src/types/app.ts
@@ -875,6 +875,8 @@ export interface AiProfile {
   apiKey: string
   model: string
   apiProtocol?: 'auto' | 'openai-responses' | 'openai-chat' | 'anthropic'
+  codexCliPath?: string
+  codexReasoningEffort?: 'default' | 'minimal' | 'low' | 'medium' | 'high' | 'xhigh' | 'max' | 'ultra'
   temperature?: number
   topP?: number
   presencePenalty?: number
@@ -894,6 +896,10 @@ export interface AppSettings {
   baseUrl: string
   /** API 线协议；auto 时按供应商和模型目录解析 */
   apiProtocol?: 'auto' | 'openai-responses' | 'openai-chat' | 'anthropic'
+  /** Codex CLI 可执行文件或所在目录 */
+  codexCliPath: string
+  /** Codex CLI 推理强度 */
+  codexReasoningEffort: 'default' | 'minimal' | 'low' | 'medium' | 'high' | 'xhigh' | 'max' | 'ultra'
   /** AI 网络请求使用的 HTTP/HTTPS 代理地址 */
   proxyUrl: string
   /** 可选：模型采样温度，留空时使用服务端默认值 */