@tyvm/knowhow 0.0.109-dev.05fe5a0 → 0.0.109-dev.2b94ba2

package/src/services/MediaProcessorService.ts

@@ -1,9 +1,10 @@
 import * as fs from "fs";
 import * as path from "path";
-import { exec } from "child_process";
+import { exec, spawn } from "child_process";
 import { promisify } from "util";
 import { fileExists, readFile, mkdir } from "../utils";
 import { AIClient } from "../clients";
+import { Models } from "../types";
 
 const execPromise = promisify(exec);
 
@@ -36,7 +37,7 @@ export interface KeyframeInfo {
  * audio/video processing steps after downloading with ytdl.
  */
 export class MediaProcessorService {
-  constructor(private clients: any) {}
+  constructor(private clients: AIClient) {}
 
   /**
    * Split an audio/video file into fixed-length mp3 chunks using ffmpeg.
@@ -45,7 +46,8 @@ export class MediaProcessorService {
     filePath: string,
     outputDir: string,
     CHUNK_LENGTH_SECONDS = 30,
-    reuseExistingChunks = true
+    reuseExistingChunks = true,
+    onProgress?: (progressFraction: number) => void
   ): Promise<string[]> {
     const parsed = path.parse(filePath);
     const fileName = parsed.name;
@@ -72,8 +74,70 @@ export class MediaProcessorService {
       }
     }
 
-    const command = `ffmpeg -i "${filePath}" -f segment -segment_time ${CHUNK_LENGTH_SECONDS} -map 0:a:0 -acodec mp3 -vn "${outputDirPath}/chunk%04d.mp3"`;
-    await execAsync(command);
+    // Use faster encoding settings:
+    // - mono audio (-ac 1): halves encoding work, Whisper handles mono fine
+    // - low bitrate (-b:a 32k): sufficient for speech, much faster encode + smaller files
+    // - fast preset not available for mp3 encoder, but limiting bitrate helps
+    // - -threads 0: use all available CPU threads for faster processing
+    // If the input is already an mp3, copy the audio stream to avoid re-encoding
+    const inputExt = path.extname(filePath).toLowerCase().replace('.', '');
+    const isAlreadyMp3 = inputExt === 'mp3';
+    const audioCodecArgs = isAlreadyMp3
+      ? '-acodec copy'
+      : '-acodec libmp3lame -ac 1 -b:a 32k -threads 0';
+
+    // Use -progress pipe:1 to get real-time progress from ffmpeg
+    // We need the total duration first to calculate fraction
+    await new Promise<void>((resolve, reject) => {
+      // Get total duration via ffprobe first
+      let totalDurationSeconds = 0;
+      exec(
+        `ffprobe -v error -show_entries format=duration -of default=noprint_wrappers=1:nokey=1 "${filePath}"`,
+        (err, stdout) => {
+          if (!err && stdout.trim()) {
+            totalDurationSeconds = parseFloat(stdout.trim()) || 0;
+          }
+
+          // Now run ffmpeg with progress reporting
+          const args = [
+            '-i', filePath,
+            '-f', 'segment',
+            '-segment_time', String(CHUNK_LENGTH_SECONDS),
+            '-map', '0:a:0',
+            ...audioCodecArgs.split(' '),
+            '-vn',
+            ...(onProgress ? ['-progress', 'pipe:1'] : []),
+            `${outputDirPath}/chunk%04d.mp3`,
+          ];
+
+          const proc = spawn('ffmpeg', args);
+
+          let stdoutBuf = '';
+          proc.stdout?.on('data', (data: Buffer) => {
+            stdoutBuf += data.toString();
+            if (onProgress && totalDurationSeconds > 0) {
+              // ffmpeg -progress outputs key=value lines; look for out_time_ms
+              const match = stdoutBuf.match(/out_time_ms=(\d+)/g);
+              if (match) {
+                const last = match[match.length - 1];
+                const ms = parseInt(last.split('=')[1], 10);
+                const fraction = Math.min(ms / 1000 / totalDurationSeconds, 1);
+                onProgress(fraction);
+                // Keep only tail to avoid unbounded buffer growth
+                stdoutBuf = stdoutBuf.slice(-500);
+              }
+            }
+          });
+
+          proc.on('close', (code) => {
+            if (code === 0) resolve();
+            else reject(new Error(`ffmpeg exited with code ${code}`));
+          });
+          proc.on('error', reject);
+        }
+      );
+    });
+
     await fs.promises.writeFile(doneFilePath, "done");
 
     const folderFiles = await fs.promises.readdir(outputDirPath);
@@ -298,8 +362,9 @@ export class MediaProcessorService {
     });
     const image = `data:image/jpeg;base64,${base64}`;
     return this.clients.createCompletion("openai", {
-      model: "gpt-4o",
+      model: Models.openai.GPT_4o,
       max_tokens: 2500,
+      timeout: 20000,
       messages: [
         {
           role: "user",
@@ -315,7 +380,8 @@ export class MediaProcessorService {
   async *streamProcessVideo(
     filePath: string,
     reusePreviousTranscript = true,
-    chunkTime = 30
+    chunkTime = 30,
+    onChunkingProgress?: (fraction: number) => void
   ) {
     const parsed = path.parse(filePath);
     const videoJson = `${parsed.dir}/${parsed.name}/video.json`;
@@ -324,7 +390,8 @@ export class MediaProcessorService {
     const transcriptions = this.streamProcessAudio(
       filePath,
       reusePreviousTranscript,
-      chunkTime
+      chunkTime,
+      onChunkingProgress
     );
 
     console.log("Extracting keyframes...");
@@ -352,7 +419,8 @@ export class MediaProcessorService {
   async *streamProcessAudio(
     filePath: string,
     reusePreviousTranscript = true,
-    chunkTime = 30
+    chunkTime = 30,
+    onChunkingProgress?: (fraction: number) => void
   ): AsyncGenerator<TranscriptChunk> {
     const parsed = path.parse(filePath);
     const outputPath = `${parsed.dir}/${parsed.name}/transcript.json`;
@@ -382,7 +450,8 @@ export class MediaProcessorService {
       filePath,
       parsed.dir,
       chunkTime,
-      reusePreviousTranscript
+      reusePreviousTranscript,
+      onChunkingProgress
     );
 
     for await (const chunk of this.streamTranscription(

package/src/services/modules/index.ts

@@ -31,14 +31,10 @@ export class ModulesService {
     // puts packages), then cwd node_modules, then global node_modules.
     // This allows modules installed via `knowhow modules install` to be found
     // even when knowhow itself is installed globally.
-    const cwdPaths = (require as any).resolve
-      ? require.resolve.paths?.("") || []
-      : [];
     const resolvePaths = [
       path.join(process.cwd(), ".knowhow", "node_modules"),
       path.join(os.homedir(), ".knowhow", "node_modules"),
       path.join(process.cwd(), "node_modules"),
-      ...cwdPaths,
     ];
 
     for (const modulePath of allModulePaths) {
@@ -57,22 +53,31 @@ export class ModulesService {
           resolvedPath = modulePath; // fall back to normal require resolution
         }
       }
-      const rawModule = require(resolvedPath);
-      const importedModule = (rawModule.default || rawModule) as KnowhowModule;
-      context.Events?.log(
-        "ModulesService",
-        `🔌 Loading module: ${modulePath} (resolved: ${resolvedPath})`
-      );
-      await importedModule.init({
-        config,
-        cwd: process.cwd(),
-        context: context as ModuleContext,
-      });
-      context.Events?.log(
-        "ModulesService",
-        `✅ Module initialized: ${modulePath} (tools: ${importedModule.tools.length}, agents: ${importedModule.agents.length}, plugins: ${importedModule.plugins.length}, clients: ${importedModule.clients.length})`
-      );
 
+      let importedModule: KnowhowModule;
+      try {
+        const rawModule = require(resolvedPath);
+        importedModule = (rawModule.default || rawModule) as KnowhowModule;
+        context.Events?.log(
+          "ModulesService",
+          `🔌 Loading module: ${modulePath} (resolved: ${resolvedPath})`
+        );
+        await importedModule.init({
+          config,
+          cwd: process.cwd(),
+          context: context as ModuleContext,
+        });
+        context.Events?.log(
+          "ModulesService",
+          `✅ Module initialized: ${modulePath} (tools: ${importedModule.tools.length}, agents: ${importedModule.agents.length}, plugins: ${importedModule.plugins.length}, clients: ${importedModule.clients.length})`
+        );
+      } catch (err: any) {
+        process.stderr.write(
+          `\n⚠️  Failed to load module "${modulePath}": ${err.message}\n` +
+          `   Run "knowhow modules setup --global" or "knowhow modules install ${modulePath} --global" to fix this.\n\n`
+        );
+        continue;
+      }
       // Only register tools/agents/plugins/clients if the relevant services
       // are available in context (they may not be during early CLI command registration)
       if (context.Agents) {

package/tests/unit/clients/AIClient.test.ts

@@ -0,0 +1,446 @@
+/**
+ * Integration tests for AIClient — verifies that retry, timeout, and
+ * AbortSignal options flow correctly through AIClient into the underlying
+ * GenericClient mock.
+ *
+ * We bypass all real provider initialisation by calling:
+ *   aiClient.registerClient(provider, mockClient)
+ *   aiClient.registerModels(provider, [model])
+ */
+
+// Prevent real _initDefaultProviders from firing (it reads env vars / files)
+jest.mock("../../../src/config", () => ({
+  getConfig: jest.fn().mockResolvedValue({ modules: [] }),
+  getGlobalConfig: jest.fn().mockResolvedValue({ modules: [] }),
+  getConfigSync: jest.fn().mockReturnValue({}),
+}));
+jest.mock("../../../src/services/KnowhowClient", () => ({
+  loadKnowhowJwt: jest.fn().mockReturnValue(null),
+  KNOWHOW_API_URL: "https://mock.local",
+}));
+
+import { AIClient } from "../../../src/clients/index";
+import type { GenericClient } from "../../../src/clients/types";
+
+// ─── Helpers ────────────────────────────────────────────────────────────────
+
+/** Build a minimal mock CompletionResponse */
+const mockCompletion = () => ({
+  choices: [{ message: { role: "assistant" as const, content: "hello" } }],
+  model: "mock-model",
+  usage: { prompt_tokens: 10, completion_tokens: 5 },
+});
+
+/** Build a minimal mock ImageGenerationResponse */
+const mockImage = () => ({
+  created: Date.now(),
+  data: [{ url: "https://mock.local/image.png" }],
+});
+
+/** Build a minimal mock EmbeddingResponse */
+const mockEmbedding = () => ({
+  data: [{ object: "embedding", embedding: [0.1, 0.2], index: 0 }],
+  model: "mock-embed",
+  usage: { prompt_tokens: 5, total_tokens: 5 },
+});
+
+/** Build a minimal mock AudioGenerationResponse */
+const mockAudio = () => ({
+  audio: Buffer.from("fake-audio"),
+  format: "mp3",
+});
+
+/**
+ * Create an AIClient with a registered mock provider.
+ * Returns the AIClient and the mocked GenericClient.
+ */
+function setupClient(overrides: Partial<GenericClient> = {}) {
+  const mockGenericClient: GenericClient = {
+    setKey: jest.fn(),
+    createChatCompletion: jest
+      .fn()
+      .mockResolvedValue(mockCompletion()),
+    createEmbedding: jest.fn().mockResolvedValue(mockEmbedding()),
+    createImageGeneration: jest.fn().mockResolvedValue(mockImage()),
+    createAudioGeneration: jest.fn().mockResolvedValue(mockAudio()),
+    createAudioTranscription: jest
+      .fn()
+      .mockResolvedValue({ text: "transcribed" }),
+    createVideoGeneration: jest.fn().mockResolvedValue({
+      created: Date.now(),
+      data: [{ url: "https://mock.local/video.mp4" }],
+    }),
+    getModels: jest.fn().mockResolvedValue([]),
+    ...overrides,
+  };
+
+  const aiClient = new AIClient();
+  // Register our mock bypassing all env/network checks
+  aiClient.registerClient("mock", mockGenericClient);
+  aiClient.registerModels("mock", ["mock-model", "mock-embed"]);
+
+  return { aiClient, mockGenericClient };
+}
+
+// ─── Tests ──────────────────────────────────────────────────────────────────
+
+describe("AIClient — retry / timeout / AbortSignal", () => {
+  afterEach(() => {
+    jest.useRealTimers();
+  });
+
+  // ── createCompletion ──────────────────────────────────────────────────────
+
+  describe("createCompletion", () => {
+    it("returns a completion on success", async () => {
+      const { aiClient } = setupClient();
+      const result = await aiClient.createCompletion("mock", {
+        model: "mock-model",
+        messages: [{ role: "user", content: "hi" }],
+      });
+      expect(result.choices[0].message.content).toBe("hello");
+    });
+
+    it("forwards the AbortSignal to createChatCompletion", async () => {
+      const { aiClient, mockGenericClient } = setupClient();
+      const controller = new AbortController();
+
+      await aiClient.createCompletion("mock", {
+        model: "mock-model",
+        messages: [],
+        signal: controller.signal,
+      });
+
+      const callArgs = (mockGenericClient.createChatCompletion as jest.Mock)
+        .mock.calls[0][0];
+      expect(callArgs.signal).toBeInstanceOf(AbortSignal);
+    });
+
+    it("retries on 429 and succeeds", async () => {
+      jest.useFakeTimers();
+      const { aiClient, mockGenericClient } = setupClient({
+        createChatCompletion: jest
+          .fn()
+          .mockRejectedValueOnce(new Error("429 rate limited"))
+          .mockResolvedValueOnce(mockCompletion()),
+      });
+
+      const promise = aiClient.createCompletion("mock", {
+        model: "mock-model",
+        messages: [],
+        maxRetries: 2,
+        backoffMs: 50,
+      });
+      await jest.runAllTimersAsync();
+      const result = await promise;
+
+      expect(result.choices[0].message.content).toBe("hello");
+      expect(mockGenericClient.createChatCompletion).toHaveBeenCalledTimes(2);
+    });
+
+    it("aborts immediately when external signal is pre-aborted", async () => {
+      const controller = new AbortController();
+      controller.abort();
+
+      const { aiClient, mockGenericClient } = setupClient();
+      await expect(
+        aiClient.createCompletion("mock", {
+          model: "mock-model",
+          messages: [],
+          signal: controller.signal,
+        })
+      ).rejects.toMatchObject({ name: "AbortError" });
+
+      expect(mockGenericClient.createChatCompletion).not.toHaveBeenCalled();
+    });
+
+    it("cancels in-flight request when external signal is aborted", async () => {
+      const controller = new AbortController();
+      const { aiClient, mockGenericClient } = setupClient({
+        createChatCompletion: jest.fn().mockImplementation((opts: any) => {
+          return new Promise((_, reject) => {
+            opts.signal?.addEventListener("abort", () =>
+              reject(opts.signal.reason)
+            );
+          });
+        }),
+      });
+
+      const promise = aiClient.createCompletion("mock", {
+        model: "mock-model",
+        messages: [],
+        signal: controller.signal,
+      });
+
+      setImmediate(() =>
+        controller.abort(new DOMException("User cancelled", "AbortError"))
+      );
+
+      await expect(promise).rejects.toMatchObject({ name: "AbortError" });
+      expect(mockGenericClient.createChatCompletion).toHaveBeenCalledTimes(1);
+    });
+
+    it("times out per-attempt and retries", async () => {
+      jest.useFakeTimers();
+      const { aiClient, mockGenericClient } = setupClient({
+        createChatCompletion: jest
+          .fn()
+          .mockImplementationOnce((opts: any) => {
+            return new Promise((_, reject) => {
+              opts.signal?.addEventListener("abort", () =>
+                reject(opts.signal.reason)
+              );
+            });
+          })
+          .mockResolvedValueOnce(mockCompletion()),
+      });
+
+      const promise = aiClient.createCompletion("mock", {
+        model: "mock-model",
+        messages: [],
+        timeout: 1000,
+        maxRetries: 2,
+        backoffMs: 10,
+      });
+      await jest.runAllTimersAsync();
+      const result = await promise;
+
+      expect(result.choices[0].message.content).toBe("hello");
+      expect(mockGenericClient.createChatCompletion).toHaveBeenCalledTimes(2);
+    });
+  });
+
+  // ── createEmbedding ───────────────────────────────────────────────────────
+
+  describe("createEmbedding", () => {
+    it("forwards the AbortSignal to createEmbedding on the client", async () => {
+      const { aiClient, mockGenericClient } = setupClient();
+      const controller = new AbortController();
+
+      await aiClient.createEmbedding("mock", {
+        input: "test text",
+        model: "mock-embed",
+        signal: controller.signal,
+      });
+
+      const callArgs = (mockGenericClient.createEmbedding as jest.Mock).mock
+        .calls[0][0];
+      expect(callArgs.signal).toBeInstanceOf(AbortSignal);
+    });
+
+    it("retries on 500 and succeeds", async () => {
+      jest.useFakeTimers();
+      const { aiClient, mockGenericClient } = setupClient({
+        createEmbedding: jest
+          .fn()
+          .mockRejectedValueOnce(new Error("500 Internal Server Error"))
+          .mockResolvedValueOnce(mockEmbedding()),
+      });
+
+      const promise = aiClient.createEmbedding("mock", {
+        input: "test",
+        model: "mock-embed",
+        maxRetries: 2,
+        backoffMs: 50,
+      });
+      await jest.runAllTimersAsync();
+      const result = await promise;
+
+      expect(result.data[0].embedding).toEqual([0.1, 0.2]);
+      expect(mockGenericClient.createEmbedding).toHaveBeenCalledTimes(2);
+    });
+  });
+
+  // ── createImageGeneration ─────────────────────────────────────────────────
+
+  describe("createImageGeneration", () => {
+    it("forwards the AbortSignal to createImageGeneration on the client", async () => {
+      const { aiClient, mockGenericClient } = setupClient();
+      const controller = new AbortController();
+
+      await aiClient.createImageGeneration("mock", {
+        model: "mock-model",
+        prompt: "a cat",
+        signal: controller.signal,
+      });
+
+      const callArgs = (mockGenericClient.createImageGeneration as jest.Mock)
+        .mock.calls[0][0];
+      expect(callArgs.signal).toBeInstanceOf(AbortSignal);
+    });
+
+    it("retries on 429 and succeeds", async () => {
+      jest.useFakeTimers();
+      const { aiClient, mockGenericClient } = setupClient({
+        createImageGeneration: jest
+          .fn()
+          .mockRejectedValueOnce(new Error("429 Too Many Requests"))
+          .mockResolvedValueOnce(mockImage()),
+      });
+
+      const promise = aiClient.createImageGeneration("mock", {
+        model: "mock-model",
+        prompt: "a cat",
+        maxRetries: 2,
+        backoffMs: 50,
+      });
+      await jest.runAllTimersAsync();
+      const result = await promise;
+
+      expect(result.data[0].url).toBe("https://mock.local/image.png");
+      expect(mockGenericClient.createImageGeneration).toHaveBeenCalledTimes(2);
+    });
+
+    it("aborts when external signal fires mid-request", async () => {
+      const controller = new AbortController();
+      const { aiClient, mockGenericClient } = setupClient({
+        createImageGeneration: jest.fn().mockImplementation((opts: any) => {
+          return new Promise((_, reject) => {
+            opts.signal?.addEventListener("abort", () =>
+              reject(opts.signal.reason)
+            );
+          });
+        }),
+      });
+
+      const promise = aiClient.createImageGeneration("mock", {
+        model: "mock-model",
+        prompt: "a cat",
+        signal: controller.signal,
+      });
+      setImmediate(() =>
+        controller.abort(new DOMException("User cancelled", "AbortError"))
+      );
+
+      await expect(promise).rejects.toMatchObject({ name: "AbortError" });
+      expect(mockGenericClient.createImageGeneration).toHaveBeenCalledTimes(1);
+    });
+  });
+
+  // ── createAudioGeneration ─────────────────────────────────────────────────
+
+  describe("createAudioGeneration", () => {
+    it("forwards the AbortSignal to createAudioGeneration on the client", async () => {
+      const { aiClient, mockGenericClient } = setupClient();
+      const controller = new AbortController();
+
+      await aiClient.createAudioGeneration("mock", {
+        model: "mock-model",
+        input: "Hello world",
+        voice: "alloy",
+        signal: controller.signal,
+      });
+
+      const callArgs = (mockGenericClient.createAudioGeneration as jest.Mock)
+        .mock.calls[0][0];
+      expect(callArgs.signal).toBeInstanceOf(AbortSignal);
+    });
+
+    it("retries on ECONNRESET and succeeds", async () => {
+      jest.useFakeTimers();
+      const { aiClient, mockGenericClient } = setupClient({
+        createAudioGeneration: jest
+          .fn()
+          .mockRejectedValueOnce(new Error("ECONNRESET"))
+          .mockResolvedValueOnce(mockAudio()),
+      });
+
+      const promise = aiClient.createAudioGeneration("mock", {
+        model: "mock-model",
+        input: "Hello",
+        voice: "alloy",
+        maxRetries: 2,
+        backoffMs: 50,
+      });
+      await jest.runAllTimersAsync();
+      const result = await promise;
+
+      expect(result.format).toBe("mp3");
+      expect(mockGenericClient.createAudioGeneration).toHaveBeenCalledTimes(2);
+    });
+  });
+
+  // ── createVideoGeneration ─────────────────────────────────────────────────
+
+  describe("createVideoGeneration", () => {
+    it("forwards the AbortSignal to createVideoGeneration on the client", async () => {
+      const { aiClient, mockGenericClient } = setupClient();
+      const controller = new AbortController();
+
+      await aiClient.createVideoGeneration("mock", {
+        model: "mock-model",
+        prompt: "a sunset",
+        signal: controller.signal,
+      });
+
+      const callArgs = (mockGenericClient.createVideoGeneration as jest.Mock)
+        .mock.calls[0][0];
+      expect(callArgs.signal).toBeInstanceOf(AbortSignal);
+    });
+
+    it("retries on 503 and succeeds", async () => {
+      jest.useFakeTimers();
+      const { aiClient, mockGenericClient } = setupClient({
+        createVideoGeneration: jest
+          .fn()
+          .mockRejectedValueOnce(new Error("503 Service Unavailable"))
+          .mockResolvedValueOnce({
+            created: Date.now(),
+            data: [{ url: "https://mock.local/video.mp4" }],
+          }),
+      });
+
+      const promise = aiClient.createVideoGeneration("mock", {
+        model: "mock-model",
+        prompt: "a sunset",
+        maxRetries: 2,
+        backoffMs: 50,
+      });
+      await jest.runAllTimersAsync();
+      const result = await promise;
+
+      expect(result.data[0].url).toBe("https://mock.local/video.mp4");
+      expect(mockGenericClient.createVideoGeneration).toHaveBeenCalledTimes(2);
+    });
+  });
+
+  // ── createAudioTranscription ──────────────────────────────────────────────
+
+  describe("createAudioTranscription", () => {
+    it("forwards the AbortSignal to createAudioTranscription on the client", async () => {
+      const { aiClient, mockGenericClient } = setupClient();
+      const controller = new AbortController();
+
+      await aiClient.createAudioTranscription("mock", {
+        file: Buffer.from("fake-audio"),
+        signal: controller.signal,
+      });
+
+      const callArgs = (
+        mockGenericClient.createAudioTranscription as jest.Mock
+      ).mock.calls[0][0];
+      expect(callArgs.signal).toBeInstanceOf(AbortSignal);
+    });
+
+    it("retries on timeout error and succeeds", async () => {
+      jest.useFakeTimers();
+      const { aiClient, mockGenericClient } = setupClient({
+        createAudioTranscription: jest
+          .fn()
+          .mockRejectedValueOnce(new Error("timeout"))
+          .mockResolvedValueOnce({ text: "hello world" }),
+      });
+
+      const promise = aiClient.createAudioTranscription("mock", {
+        file: Buffer.from("fake-audio"),
+        maxRetries: 2,
+        backoffMs: 50,
+      });
+      await jest.runAllTimersAsync();
+      const result = await promise;
+
+      expect(result.text).toBe("hello world");
+      expect(mockGenericClient.createAudioTranscription).toHaveBeenCalledTimes(2);
+    });
+  });
+});