Phase 4 구현: Ollama LLM 연동 (다듬기, 번역, 스트리밍)

- LocalLLMService: Ollama REST API, 스트리밍 NDJSON 파싱, 5초 가용성 폴링
- 시스템 프롬프트: refine/translate/summarize/grammar/expand/custom 6개 액션
- VoiceModeService: 전사→LLM 후처리→텍스트 삽입, LLM 실패 시 원본 폴백
- LLM IPC 핸들러: status/models/process/cancel/serverUrl 8개
- StatusBar: Ollama 연결 상태 + 활성 모델 표시
- Preload: llm API 섹션 추가
- Bootstrap: llm-polling 초기화 단계 추가

CLAUDE.md

@@ -77,10 +77,18 @@ npm run typecheck    # tsc --noEmit
 6. **녹음 UI**: 9개 웨이브 바, cos 분포 가중치, 100ms 애니메이션
 
 ## 현재 상태
-Phase: 3 완료
+Phase: 4 완료
-마지막 완료: Phase 3 — TextInsertService + RecordingTip/ResultPopup 팝업 + Settings 모달
+마지막 완료: Phase 4 — Ollama LLM 연동 (텍스트 다듬기, 번역, 스트리밍)
-다음 작업: Phase 3.5 — 커서 위치 히스토리 팝업 (D3RO 고유 기능)
+다음 작업: Phase 3.5 — 커서 위치 히스토리 팝업 또는 Phase 5 — TTS + DB
-차단 이슈: SoX 미설치 시 AudioCaptureService 동작 불가, @nut-tree/nut-js → @nut-tree-fork/nut-js 포크 사용
+차단 이슈: SoX 미설치, @nut-tree-fork/nut-js 포크 사용
+
+### Phase 4 구현 내용
+- LocalLLMService: Ollama REST API 연동, 스트리밍 NDJSON 파싱, 가용성 폴링(5초)
+- 시스템 프롬프트: refine/translate/summarize/grammar/expand/custom 6개 액션
+- VoiceModeService LLM 연동: 전사→LLM 후처리→텍스트 삽입, LLM 실패 시 원본 폴백
+- LLM IPC 핸들러: llm:getStatus/getModels/setModel/process/cancelProcess/getServerUrl/setServerUrl
+- StatusBar: Ollama 연결 상태 + 활성 모델 표시
+- Bootstrap: LLM 가용성 폴링 초기화 단계 추가
 
 ### Phase 3 구현 내용
 - TextInsertService: clipboard save→set→Ctrl+V→restore (@nut-tree-fork/nut-js, lazy dynamic import)

src/main/bootstrap.ts

@@ -5,6 +5,7 @@ import { initLoggerService, getLogger } from './services/LoggerService'
 import { initConfigService } from './services/ConfigService'
 import { getHotkeyService } from './services/HotkeyService'
 import { getVoiceModeService } from './services/VoiceModeService'
+import { getLocalLLMService } from './services/LocalLLMService'
 import {
   createMainWindow,
   preloadPopupWindows,
@@ -34,7 +35,8 @@ export async function bootstrap(): Promise<void> {
     { name: 'ipc-handlers', critical: true, fn: initIpcHandlers },
     { name: 'popup-preload', critical: false, fn: initPopupWindows },
     { name: 'hotkey', critical: false, fn: initHotkey },
-    { name: 'voice-mode', critical: false, fn: initVoiceMode }
+    { name: 'voice-mode', critical: false, fn: initVoiceMode },
+    { name: 'llm-polling', critical: false, fn: initLLMPolling }
   ]
 
   for (const step of steps) {
@@ -120,3 +122,8 @@ async function initVoiceMode(): Promise<void> {
     setTimeout(() => hideRecordingTip(), 3000)
   })
 }
+
+async function initLLMPolling(): Promise<void> {
+  const llm = getLocalLLMService()
+  llm.startPolling()
+}

src/main/ipc/index.ts

@@ -7,6 +7,7 @@ import { registerSystemHandlers } from './system-handlers'
 import { registerVoiceHandlers } from './voice-handlers'
 import { registerSTTHandlers } from './stt-handlers'
 import { registerHotkeyHandlers } from './hotkey-handlers'
+import { registerLLMHandlers } from './llm-handlers'
 import { getLogger } from '../services/LoggerService'
 
 const logger = getLogger('ipc')
@@ -19,5 +20,6 @@ export function registerAllIpcHandlers(): void {
   registerVoiceHandlers()
   registerSTTHandlers()
   registerHotkeyHandlers()
+  registerLLMHandlers()
   logger.info('All IPC handlers registered')
 }

src/main/ipc/llm-handlers.ts

@@ -0,0 +1,68 @@
+// src/main/ipc/llm-handlers.ts
+
+import { ipcMain } from 'electron'
+import { IPC_CHANNELS } from '@shared/ipc-channels'
+import { ipcSuccess, ipcError, ErrorCode } from '@shared/errors'
+import { getLocalLLMService } from '../services/LocalLLMService'
+import { configGet, configSet } from '../services/ConfigService'
+import type { SetLLMModelParams, SetServerUrlParams, LLMProcessParams } from '@shared/types'
+
+export function registerLLMHandlers(): void {
+  ipcMain.handle(IPC_CHANNELS.LLM.GET_STATUS, async () => {
+    return ipcSuccess(getLocalLLMService().getStatus())
+  })
+
+  ipcMain.handle(IPC_CHANNELS.LLM.GET_MODELS, async () => {
+    try {
+      const models = await getLocalLLMService().getModels()
+      return ipcSuccess(models)
+    } catch {
+      return ipcError(ErrorCode.LLMServerUnreachable, 'Failed to get models')
+    }
+  })
+
+  ipcMain.handle(IPC_CHANNELS.LLM.GET_ACTIVE_MODEL, async () => {
+    return ipcSuccess(configGet('llmModelId'))
+  })
+
+  ipcMain.handle(IPC_CHANNELS.LLM.SET_MODEL, async (_event, params: SetLLMModelParams) => {
+    configSet('llmModelId', params.modelId)
+    return ipcSuccess(undefined)
+  })
+
+  ipcMain.handle(IPC_CHANNELS.LLM.PROCESS, async (_event, params: LLMProcessParams) => {
+    try {
+      const llm = getLocalLLMService()
+      const start = performance.now()
+      const processedText = await llm.processText(
+        params.text,
+        params.action,
+        params.targetLanguage,
+        params.customPrompt
+      )
+      return ipcSuccess({
+        originalText: params.text,
+        processedText,
+        action: params.action,
+        processingTimeMs: Math.round(performance.now() - start),
+        tokenCount: 0
+      })
+    } catch {
+      return ipcError(ErrorCode.LLMProcessingFailed, 'LLM processing failed')
+    }
+  })
+
+  ipcMain.handle(IPC_CHANNELS.LLM.CANCEL_PROCESS, async () => {
+    getLocalLLMService().cancelGeneration()
+    return ipcSuccess(undefined)
+  })
+
+  ipcMain.handle(IPC_CHANNELS.LLM.GET_SERVER_URL, async () => {
+    return ipcSuccess(configGet('ollamaServerUrl'))
+  })
+
+  ipcMain.handle(IPC_CHANNELS.LLM.SET_SERVER_URL, async (_event, params: SetServerUrlParams) => {
+    configSet('ollamaServerUrl', params.url)
+    return ipcSuccess(undefined)
+  })
+}

src/main/services/LocalLLMService.ts

@@ -0,0 +1,433 @@
+// src/main/services/LocalLLMService.ts
+// Ollama REST API를 통해 로컬 LLM과 상호작용한다.
+// 설계서 01의 ILocalLLMService 구현.
+
+import { EventEmitter } from 'events'
+import { getLogger } from './LoggerService'
+import { configGet } from './ConfigService'
+import { D3ROError, ErrorCode } from '@shared/errors'
+import type { LLMStatus, LLMModel, LLMAction, LLMConnectionState } from '@shared/types'
+
+const logger = getLogger('LocalLLMService')
+
+// ============================================================
+// 내부 타입
+// ============================================================
+
+const enum LLMState {
+  Unavailable = 'unavailable',
+  Available = 'available',
+  Generating = 'generating',
+  Error = 'error'
+}
+
+interface GenerateOptions {
+  model?: string
+  temperature?: number
+  maxTokens?: number
+  systemPrompt?: string
+  stream?: boolean
+}
+
+interface GenerateResult {
+  text: string
+  model: string
+  promptTokens: number
+  completionTokens: number
+  totalDuration: number
+}
+
+interface OllamaGenerateResponse {
+  model: string
+  response: string
+  done: boolean
+  total_duration?: number
+  prompt_eval_count?: number
+  eval_count?: number
+}
+
+interface OllamaTagsResponse {
+  models: Array<{
+    name: string
+    size: number
+    parameter_size: string
+    quantization_level: string
+    modified_at: string
+  }>
+}
+
+interface LocalLLMEvents {
+  token: (payload: { token: string; done: boolean }) => void
+  complete: (payload: { result: GenerateResult }) => void
+  'availability-changed': (payload: { available: boolean }) => void
+  error: (payload: { error: D3ROError }) => void
+}
+
+// ============================================================
+// 시스템 프롬프트 (설계서 Phase 4 참조)
+// ============================================================
+
+const SYSTEM_PROMPTS: Record<string, string> = {
+  refine: `다음 음성 전사 텍스트를 자연스럽고 격식 있는 문어체로 다듬어주세요.
+원래 의미를 유지하면서 문법 오류를 수정하고, 불필요한 반복이나 필러를 제거하세요.
+다듬어진 텍스트만 출력하세요. 설명이나 부가 문구를 붙이지 마세요.`,
+
+  translate: `다음 텍스트를 {{targetLanguage}}로 번역해주세요.
+자연스럽고 정확한 번역만 출력하세요. 원문이나 설명을 붙이지 마세요.`,
+
+  summarize: `다음 텍스트의 핵심 내용을 3줄 이내로 요약해주세요.
+요약문만 출력하세요.`,
+
+  grammar: `다음 텍스트의 문법 오류만 수정해주세요.
+원래 의미와 톤을 유지하면서 문법 오류만 수정하세요.
+수정된 텍스트만 출력하세요.`,
+
+  expand: `다음 텍스트를 더 자세하고 풍부하게 확장해주세요.
+확장된 텍스트만 출력하세요.`
+}
+
+// ============================================================
+// LocalLLMService
+// ============================================================
+
+class LocalLLMService extends EventEmitter {
+  private _state = LLMState.Unavailable
+  private _pollInterval: ReturnType<typeof setInterval> | null = null
+  private _available = false
+  private _abortController: AbortController | null = null
+  private _disposed = false
+
+  get state(): LLMState {
+    return this._state
+  }
+
+  /**
+   * Ollama 가용성 폴링을 시작한다 (5초 간격).
+   */
+  startPolling(): void {
+    this._checkAvailability()
+    this._pollInterval = setInterval(() => {
+      if (this._state !== LLMState.Generating) {
+        this._checkAvailability()
+      }
+    }, 5000)
+    logger.info('Ollama availability polling started')
+  }
+
+  stopPolling(): void {
+    if (this._pollInterval) {
+      clearInterval(this._pollInterval)
+      this._pollInterval = null
+    }
+  }
+
+  /**
+   * 비스트리밍 텍스트 생성.
+   */
+  async generate(prompt: string, options?: GenerateOptions): Promise<GenerateResult> {
+    if (!this._available) {
+      throw new D3ROError(ErrorCode.LLMServerUnreachable, 'Ollama 서버에 연결할 수 없습니다')
+    }
+
+    const serverUrl = configGet('ollamaServerUrl')
+    const model = options?.model ?? configGet('llmModelId') ?? 'qwen3:4b'
+
+    this._state = LLMState.Generating
+
+    try {
+      const response = await fetch(`${serverUrl}/api/generate`, {
+        method: 'POST',
+        headers: { 'Content-Type': 'application/json' },
+        body: JSON.stringify({
+          model,
+          prompt,
+          system: options?.systemPrompt,
+          stream: false,
+          options: {
+            temperature: options?.temperature ?? 0.3,
+            num_predict: options?.maxTokens ?? 2048
+          }
+        }),
+        signal: AbortSignal.timeout(120000)
+      })
+
+      if (!response.ok) {
+        throw new D3ROError(
+          ErrorCode.LLMProcessingFailed,
+          `Ollama responded with ${response.status}: ${response.statusText}`
+        )
+      }
+
+      const data = (await response.json()) as OllamaGenerateResponse
+
+      const result: GenerateResult = {
+        text: data.response,
+        model: data.model,
+        promptTokens: data.prompt_eval_count ?? 0,
+        completionTokens: data.eval_count ?? 0,
+        totalDuration: data.total_duration ? data.total_duration / 1e6 : 0
+      }
+
+      this._state = LLMState.Available
+      this.emit('complete', { result })
+      return result
+    } catch (error) {
+      this._state = LLMState.Available
+      if (error instanceof D3ROError) throw error
+      throw new D3ROError(
+        ErrorCode.LLMProcessingFailed,
+        `LLM generation failed: ${error instanceof Error ? error.message : String(error)}`
+      )
+    }
+  }
+
+  /**
+   * 스트리밍 텍스트 생성. NDJSON 파싱.
+   * 반환된 AbortController로 취소 가능.
+   */
+  async *streamGenerate(
+    prompt: string,
+    options?: Omit<GenerateOptions, 'stream'>
+  ): AsyncGenerator<string, GenerateResult> {
+    if (!this._available) {
+      throw new D3ROError(ErrorCode.LLMServerUnreachable, 'Ollama 서버에 연결할 수 없습니다')
+    }
+
+    const serverUrl = configGet('ollamaServerUrl')
+    const model = options?.model ?? configGet('llmModelId') ?? 'qwen3:4b'
+
+    this._state = LLMState.Generating
+    this._abortController = new AbortController()
+
+    try {
+      const response = await fetch(`${serverUrl}/api/generate`, {
+        method: 'POST',
+        headers: { 'Content-Type': 'application/json' },
+        body: JSON.stringify({
+          model,
+          prompt,
+          system: options?.systemPrompt,
+          stream: true,
+          options: {
+            temperature: options?.temperature ?? 0.3,
+            num_predict: options?.maxTokens ?? 2048
+          }
+        }),
+        signal: this._abortController.signal
+      })
+
+      if (!response.ok || !response.body) {
+        throw new D3ROError(
+          ErrorCode.LLMProcessingFailed,
+          `Ollama responded with ${response.status}`
+        )
+      }
+
+      const reader = response.body.getReader()
+      const decoder = new TextDecoder()
+      let buffer = ''
+      let fullText = ''
+      let lastChunk: OllamaGenerateResponse | null = null
+
+      while (true) {
+        const { done, value } = await reader.read()
+        if (done) break
+
+        buffer += decoder.decode(value, { stream: true })
+        const lines = buffer.split('\n')
+        buffer = lines.pop() ?? ''
+
+        for (const line of lines) {
+          if (!line.trim()) continue
+          try {
+            const chunk = JSON.parse(line) as OllamaGenerateResponse
+            fullText += chunk.response
+            this.emit('token', { token: chunk.response, done: chunk.done })
+            yield chunk.response
+
+            if (chunk.done) {
+              lastChunk = chunk
+            }
+          } catch {
+            logger.warn(`Failed to parse NDJSON line: ${line.substring(0, 100)}`)
+          }
+        }
+      }
+
+      this._state = LLMState.Available
+      this._abortController = null
+
+      const result: GenerateResult = {
+        text: fullText,
+        model: lastChunk?.model ?? model,
+        promptTokens: lastChunk?.prompt_eval_count ?? 0,
+        completionTokens: lastChunk?.eval_count ?? 0,
+        totalDuration: lastChunk?.total_duration ? lastChunk.total_duration / 1e6 : 0
+      }
+
+      this.emit('complete', { result })
+      return result
+    } catch (error) {
+      this._state = LLMState.Available
+      this._abortController = null
+      if (error instanceof D3ROError) throw error
+      if (error instanceof DOMException && error.name === 'AbortError') {
+        throw new D3ROError(ErrorCode.LLMProcessingCancelled, 'LLM generation cancelled')
+      }
+      throw new D3ROError(
+        ErrorCode.LLMProcessingFailed,
+        `LLM streaming failed: ${error instanceof Error ? error.message : String(error)}`
+      )
+    }
+  }
+
+  /**
+   * 텍스트를 LLM 액션에 따라 처리한다.
+   */
+  async processText(
+    text: string,
+    action: LLMAction,
+    targetLanguage?: string,
+    customPrompt?: string
+  ): Promise<string> {
+    let systemPrompt: string
+
+    if (action === 'custom' && customPrompt) {
+      systemPrompt = customPrompt
+    } else if (action === 'translate') {
+      systemPrompt = SYSTEM_PROMPTS.translate.replace(
+        '{{targetLanguage}}',
+        targetLanguage ?? 'English'
+      )
+    } else {
+      systemPrompt = SYSTEM_PROMPTS[action] ?? SYSTEM_PROMPTS.refine
+    }
+
+    const result = await this.generate(text, { systemPrompt })
+    return result.text.trim()
+  }
+
+  cancelGeneration(): void {
+    if (this._abortController) {
+      this._abortController.abort()
+      this._abortController = null
+      logger.info('LLM generation cancelled')
+    }
+  }
+
+  /**
+   * Ollama에 설치된 모델 목록을 조회한다.
+   */
+  async getModels(): Promise<LLMModel[]> {
+    const serverUrl = configGet('ollamaServerUrl')
+
+    try {
+      const response = await fetch(`${serverUrl}/api/tags`, {
+        signal: AbortSignal.timeout(5000)
+      })
+
+      if (!response.ok) return []
+
+      const data = (await response.json()) as OllamaTagsResponse
+
+      return data.models.map((m) => ({
+        id: m.name,
+        name: m.name,
+        sizeBytes: m.size,
+        parameterSize: m.parameter_size ?? '',
+        quantization: m.quantization_level ?? '',
+        modifiedAt: m.modified_at
+      }))
+    } catch {
+      return []
+    }
+  }
+
+  getStatus(): LLMStatus {
+    const connectionState: LLMConnectionState = this._available
+      ? this._state === LLMState.Generating
+        ? 'connecting'
+        : 'connected'
+      : 'disconnected'
+
+    return {
+      connectionState,
+      serverUrl: configGet('ollamaServerUrl'),
+      activeModel: configGet('llmModelId'),
+      serverVersion: null
+    }
+  }
+
+  isAvailable(): boolean {
+    return this._available
+  }
+
+  dispose(): void {
+    this._disposed = true
+    this.stopPolling()
+    this.cancelGeneration()
+    this.removeAllListeners()
+    logger.info('LocalLLMService disposed')
+  }
+
+  // ── 가용성 체크 ────────────────────────────────────────
+
+  private async _checkAvailability(): Promise<void> {
+    if (this._disposed) return
+    const serverUrl = configGet('ollamaServerUrl')
+
+    try {
+      const response = await fetch(`${serverUrl}/api/tags`, {
+        signal: AbortSignal.timeout(3000)
+      })
+
+      const wasAvailable = this._available
+      this._available = response.ok
+
+      if (!wasAvailable && this._available) {
+        this._state = LLMState.Available
+        this.emit('availability-changed', { available: true })
+        logger.info('Ollama server connected')
+      } else if (wasAvailable && !this._available) {
+        this._state = LLMState.Unavailable
+        this.emit('availability-changed', { available: false })
+        logger.warn('Ollama server disconnected')
+      }
+    } catch {
+      if (this._available) {
+        this._available = false
+        this._state = LLMState.Unavailable
+        this.emit('availability-changed', { available: false })
+        logger.warn('Ollama server unreachable')
+      }
+    }
+  }
+
+  // ── EventEmitter 타입 오버라이드 ───────────────────────
+
+  override on<K extends keyof LocalLLMEvents>(event: K, listener: LocalLLMEvents[K]): this {
+    return super.on(event, listener)
+  }
+
+  override off<K extends keyof LocalLLMEvents>(event: K, listener: LocalLLMEvents[K]): this {
+    return super.off(event, listener)
+  }
+
+  override emit<K extends keyof LocalLLMEvents>(
+    event: K,
+    ...args: Parameters<LocalLLMEvents[K]>
+  ): boolean {
+    return super.emit(event, ...args)
+  }
+}
+
+// ── 싱글톤 ─────────────────────────────────────────────
+
+let instance: LocalLLMService | null = null
+
+export function getLocalLLMService(): LocalLLMService {
+  if (!instance) {
+    instance = new LocalLLMService()
+  }
+  return instance
+}

src/main/services/VoiceModeService.ts

@@ -13,6 +13,7 @@ import { getHotkeyService } from './HotkeyService'
 import type { HotkeyConfig } from './HotkeyService'
 import { configGet } from './ConfigService'
 import { getTextInsertService } from './TextInsertService'
+import { getLocalLLMService } from './LocalLLMService'
 import { D3ROError, ErrorCode } from '@shared/errors'
 import { TIMING } from '@shared/constants'
 import { RecognitionState, AudioState } from '@shared/types'
@@ -407,8 +408,14 @@ class VoiceModeService extends EventEmitter {
 
       this.emit('transcription-update', { text: result.text, isFinal: true })
 
-      // Phase 2: LLM 후처리 없이 바로 완료
+      // LLM 후처리
+      const llmAction = configGet('defaultLLMAction')
+      if (llmAction !== 'refine' || !getLocalLLMService().isAvailable()) {
+        // LLM 미가용이거나 기본 액션이면 원본 텍스트로 완료
         this._completeSession(result.text)
+      } else {
+        await this._processWithLLM(result.text)
+      }
     } catch (error) {
       if (this._isInTerminalState()) return
       this._handleError(
@@ -420,6 +427,35 @@ class VoiceModeService extends EventEmitter {
     }
   }
 
+  // ── LLM 후처리 ─────────────────────────────────────────
+
+  private async _processWithLLM(transcribedText: string): Promise<void> {
+    if (this._isInTerminalState()) return
+
+    // RECOGNIZING 상태 유지 (UI에서 thinking으로 표시됨)
+    try {
+      const llm = getLocalLLMService()
+      const action = configGet('defaultLLMAction')
+
+      logger.info(`Processing with LLM (action: ${action})`)
+
+      const processedText = await llm.processText(transcribedText, action)
+
+      if (this._isInTerminalState()) return
+
+      if (this._session) {
+        this._session.processedText = processedText
+      }
+
+      this._completeSession(processedText)
+    } catch (error) {
+      if (this._isInTerminalState()) return
+      logger.warn(`LLM processing failed, using original text: ${error instanceof Error ? error.message : String(error)}`)
+      // LLM 실패 시 원본 텍스트로 폴백
+      this._completeSession(transcribedText)
+    }
+  }
+
   // ── 세션 완료/취소 ─────────────────────────────────────
 
   private async _completeSession(finalText: string): Promise<void> {

src/main/services/index.ts

@@ -25,3 +25,4 @@ export { getHotkeyService } from './HotkeyService'
 export { getVoiceModeService } from './VoiceModeService'
 export { getAudioCaptureService } from './AudioCaptureService'
 export { getTextInsertService } from './TextInsertService'
+export { getLocalLLMService } from './LocalLLMService'

src/preload/index.ts

@@ -39,6 +39,14 @@ import type {
   SetHotkeyParams,
   SetEnabledParams,
   HotkeyTriggeredEvent,
+  LLMStatus,
+  LLMModel,
+  LLMProcessParams,
+  LLMProcessResult,
+  SetLLMModelParams,
+  SetServerUrlParams,
+  LLMStatusChangedEvent,
+  LLMProcessProgressEvent,
   PermissionStatus
 } from '@shared/types'
 import type { IPCResult } from '@shared/errors'
@@ -142,6 +150,25 @@ const electronAPI = {
       on(IPC_CHANNELS.HOTKEY.TRIGGERED, cb)
   },
 
+  // ── LLM ────────────────────────────────────────────────
+  llm: {
+    getStatus: () => invoke<LLMStatus>(IPC_CHANNELS.LLM.GET_STATUS),
+    getModels: () => invoke<LLMModel[]>(IPC_CHANNELS.LLM.GET_MODELS),
+    getActiveModel: () => invoke<string | null>(IPC_CHANNELS.LLM.GET_ACTIVE_MODEL),
+    setModel: (params: SetLLMModelParams) =>
+      invoke<void>(IPC_CHANNELS.LLM.SET_MODEL, params),
+    process: (params: LLMProcessParams) =>
+      invoke<LLMProcessResult>(IPC_CHANNELS.LLM.PROCESS, params),
+    cancelProcess: () => invoke<void>(IPC_CHANNELS.LLM.CANCEL_PROCESS),
+    getServerUrl: () => invoke<string>(IPC_CHANNELS.LLM.GET_SERVER_URL),
+    setServerUrl: (params: SetServerUrlParams) =>
+      invoke<void>(IPC_CHANNELS.LLM.SET_SERVER_URL, params),
+    onStatusChanged: (cb: (e: LLMStatusChangedEvent) => void): Unsubscribe =>
+      on(IPC_CHANNELS.LLM.STATUS_CHANGED, cb),
+    onProcessProgress: (cb: (e: LLMProcessProgressEvent) => void): Unsubscribe =>
+      on(IPC_CHANNELS.LLM.PROCESS_PROGRESS, cb)
+  },
+
   // ── Window ─────────────────────────────────────────────
   window: {
     minimize: () => send(IPC_CHANNELS.WINDOW.MINIMIZE),

src/renderer/components/AppLayout.tsx

@@ -18,6 +18,7 @@ import MenuBookIcon from '@mui/icons-material/MenuBook'
 import SettingsIcon from '@mui/icons-material/Settings'
 import { DashboardPage } from '../pages/DashboardPage'
 import { SettingsModal } from './SettingsModal'
+import { StatusBar } from './StatusBar'
 
 type Route = 'dashboard' | 'history' | 'dictionary'
 
@@ -34,7 +35,8 @@ export function AppLayout(): React.ReactElement {
   const [settingsOpen, setSettingsOpen] = useState(false)
 
   return (
-    <Box sx={{ display: 'flex', height: '100vh' }}>
+    <Box sx={{ display: 'flex', height: '100vh', flexDirection: 'column' }}>
+    <Box sx={{ display: 'flex', flex: 1, overflow: 'hidden' }}>
       {/* Sidebar Drawer */}
       <Drawer
         variant="permanent"
@@ -107,7 +109,8 @@ export function AppLayout(): React.ReactElement {
           </Typography>
         )}
       </Box>
-
+    </Box>
+      <StatusBar />
       <SettingsModal open={settingsOpen} onClose={() => setSettingsOpen(false)} />
     </Box>
   )

src/renderer/components/StatusBar.tsx

@@ -0,0 +1,69 @@
+// src/renderer/components/StatusBar.tsx
+// 하단 상태 표시: Ollama 연결 상태
+
+import { useState, useEffect } from 'react'
+import { Box, Chip } from '@mui/material'
+import CircleIcon from '@mui/icons-material/Circle'
+import type { LLMStatus } from '@shared/types'
+
+export function StatusBar(): React.ReactElement {
+  const [llmStatus, setLlmStatus] = useState<LLMStatus | null>(null)
+
+  useEffect(() => {
+    // 초기 상태 조회
+    window.electronAPI.llm.getStatus().then((result) => {
+      if (result.success) setLlmStatus(result.data)
+    })
+
+    // 상태 변경 구독
+    const unsub = window.electronAPI.llm.onStatusChanged((event) => {
+      setLlmStatus(event.status)
+    })
+
+    // 5초마다 폴링 (main에서 이벤트를 보내지 않을 수 있으므로)
+    const interval = setInterval(() => {
+      window.electronAPI.llm.getStatus().then((result) => {
+        if (result.success) setLlmStatus(result.data)
+      })
+    }, 5000)
+
+    return () => {
+      unsub()
+      clearInterval(interval)
+    }
+  }, [])
+
+  const connected = llmStatus?.connectionState === 'connected'
+
+  return (
+    <Box
+      sx={{
+        display: 'flex',
+        alignItems: 'center',
+        gap: 1,
+        px: 2,
+        py: 0.5,
+        borderTop: 1,
+        borderColor: 'divider',
+        bgcolor: 'background.paper'
+      }}
+    >
+      <Chip
+        icon={<CircleIcon sx={{ fontSize: 8 }} />}
+        label={connected ? 'Ollama Connected' : 'Ollama Offline'}
+        size="small"
+        variant="outlined"
+        color={connected ? 'success' : 'default'}
+        sx={{ height: 22, '& .MuiChip-label': { fontSize: 11 } }}
+      />
+      {llmStatus?.activeModel && (
+        <Chip
+          label={llmStatus.activeModel}
+          size="small"
+          variant="outlined"
+          sx={{ height: 22, '& .MuiChip-label': { fontSize: 11 } }}
+        />
+      )}
+    </Box>
+  )
+}