open-agents-ai 0.94.0 → 0.95.0

package/dist/index.js

@@ -32523,13 +32523,66 @@ var init_voice = __esm({
         id: "glados",
         label: "GLaDOS",
         onnxUrl: "https://raw.githubusercontent.com/robit-man/EGG/main/voice/glados_piper_medium.onnx",
-        configUrl: "https://raw.githubusercontent.com/robit-man/EGG/main/voice/glados_piper_medium.onnx.json"
+        configUrl: "https://raw.githubusercontent.com/robit-man/EGG/main/voice/glados_piper_medium.onnx.json",
+        backend: "onnx"
       },
       overwatch: {
         id: "overwatch",
         label: "Overwatch",
         onnxUrl: "https://raw.githubusercontent.com/robit-man/combine_overwatch_onnx/main/overwatch.onnx",
-        configUrl: "https://raw.githubusercontent.com/robit-man/combine_overwatch_onnx/main/overwatch.onnx.json"
+        configUrl: "https://raw.githubusercontent.com/robit-man/combine_overwatch_onnx/main/overwatch.onnx.json",
+        backend: "onnx"
+      },
+      // MLX Audio models — macOS Apple Silicon only
+      kokoro: {
+        id: "kokoro",
+        label: "Kokoro (MLX)",
+        onnxUrl: "",
+        configUrl: "",
+        backend: "mlx",
+        mlxModelId: "mlx-community/Kokoro-82M-bf16",
+        mlxVoice: "af_heart",
+        mlxLangCode: "a"
+      },
+      "kokoro:af_heart": {
+        id: "kokoro:af_heart",
+        label: "Kokoro Heart (MLX)",
+        onnxUrl: "",
+        configUrl: "",
+        backend: "mlx",
+        mlxModelId: "mlx-community/Kokoro-82M-bf16",
+        mlxVoice: "af_heart",
+        mlxLangCode: "a"
+      },
+      "kokoro:af_bella": {
+        id: "kokoro:af_bella",
+        label: "Kokoro Bella (MLX)",
+        onnxUrl: "",
+        configUrl: "",
+        backend: "mlx",
+        mlxModelId: "mlx-community/Kokoro-82M-bf16",
+        mlxVoice: "af_bella",
+        mlxLangCode: "a"
+      },
+      "kokoro:am_adam": {
+        id: "kokoro:am_adam",
+        label: "Kokoro Adam (MLX)",
+        onnxUrl: "",
+        configUrl: "",
+        backend: "mlx",
+        mlxModelId: "mlx-community/Kokoro-82M-bf16",
+        mlxVoice: "am_adam",
+        mlxLangCode: "a"
+      },
+      "kokoro:bf_emma": {
+        id: "kokoro:bf_emma",
+        label: "Kokoro Emma (MLX)",
+        onnxUrl: "",
+        configUrl: "",
+        backend: "mlx",
+        mlxModelId: "mlx-community/Kokoro-82M-bf16",
+        mlxVoice: "bf_emma",
+        mlxLangCode: "b"
       }
     };
     VoiceEngine = class {
@@ -32551,6 +32604,12 @@ var init_voice = __esm({
       speakQueue = [];
       speaking = false;
       phonemizeFn = null;
+      /** True when current model uses MLX Audio backend */
+      mlxActive = false;
+      /** Cached: is mlx-audio pip package installed? */
+      mlxInstalled = null;
+      /** Path to python3 binary (resolved once) */
+      python3Path = null;
       // -------------------------------------------------------------------------
       // Public API
       // -------------------------------------------------------------------------
@@ -32560,13 +32619,21 @@ var init_voice = __esm({
           this.killPlayback();
           return "Voice feedback disabled.";
         }
+        const model = VOICE_MODELS[this.modelId];
+        const isMlx = model?.backend === "mlx";
         try {
-          await this.ensureRuntime();
-          await this.ensureModel(this.modelId);
-          await this.loadSession();
+          if (isMlx) {
+            await this.ensureMlxAudio();
+            this.mlxActive = true;
+          } else {
+            await this.ensureRuntime();
+            await this.ensureModel(this.modelId);
+            await this.loadSession();
+            this.mlxActive = false;
+          }
           this.enabled = true;
           this.ready = true;
-          return `Voice feedback enabled (${VOICE_MODELS[this.modelId]?.label ?? this.modelId}).`;
+          return `Voice feedback enabled (${model?.label ?? this.modelId}).`;
         } catch (err) {
           return `Failed to enable voice: ${err instanceof Error ? err.message : String(err)}`;
         }
@@ -32574,24 +32641,36 @@ var init_voice = __esm({
       async setModel(id) {
         const key = id.toLowerCase();
         if (!VOICE_MODELS[key]) {
-          return `Unknown voice model: "${id}". Available: ${Object.keys(VOICE_MODELS).join(", ")}`;
-        }
+          const onnxModels = Object.keys(VOICE_MODELS).filter((k) => VOICE_MODELS[k].backend !== "mlx");
+          const mlxModels = Object.keys(VOICE_MODELS).filter((k) => VOICE_MODELS[k].backend === "mlx");
+          return `Unknown voice model: "${id}". Available:
+  ONNX: ${onnxModels.join(", ")}
+  MLX (macOS): ${mlxModels.join(", ")}`;
+        }
+        const model = VOICE_MODELS[key];
+        const isMlx = model.backend === "mlx";
         this.modelId = key;
         this.session = null;
         this.config = null;
         this.ready = false;
         if (this.enabled) {
           try {
-            await this.ensureModel(key);
-            await this.loadSession();
+            if (isMlx) {
+              await this.ensureMlxAudio();
+              this.mlxActive = true;
+            } else {
+              await this.ensureModel(key);
+              await this.loadSession();
+              this.mlxActive = false;
+            }
             this.ready = true;
-            return `Switched to ${VOICE_MODELS[key].label} voice.`;
+            return `Switched to ${model.label} voice.`;
           } catch (err) {
             this.enabled = false;
             return `Failed to load ${key}: ${err instanceof Error ? err.message : String(err)}`;
           }
         }
-        return `Voice model set to ${VOICE_MODELS[key].label}. Enable with /voice.`;
+        return `Voice model set to ${model.label}. Enable with /voice.`;
       }
       /**
        * Speak text asynchronously (non-blocking) at full volume.
@@ -32636,6 +32715,9 @@ var init_voice = __esm({
        * Used for sending TTS audio to Telegram and WebSocket clients.
        */
       async synthesizeToBuffer(text) {
+        if (this.mlxActive) {
+          return this.synthesizeMlxToBuffer(text);
+        }
         if (!this.session || !this.config || !this.ort)
           return null;
         const cleaned = text.replace(/\*/g, "");
@@ -32668,6 +32750,14 @@ var init_voice = __esm({
        * Used for streaming to WebSocket clients.
        */
       async synthesizeToPCM(text) {
+        if (this.mlxActive) {
+          const wav = await this.synthesizeMlxToBuffer(text);
+          if (!wav || wav.length <= 44)
+            return null;
+          const sampleRate = wav.readUInt32LE(24);
+          const pcm = Buffer.from(wav.buffer, wav.byteOffset + 44, wav.length - 44);
+          return { pcm, sampleRate };
+        }
         if (!this.session || !this.config || !this.ort)
           return null;
         const cleaned = text.replace(/\*/g, "");
@@ -32705,6 +32795,8 @@ var init_voice = __esm({
       }
       /** Get the current model's sample rate */
       get sampleRate() {
+        if (this.mlxActive)
+          return 24e3;
         return this.config?.audio?.sample_rate ?? 22050;
       }
       dispose() {
@@ -32713,6 +32805,7 @@ var init_voice = __esm({
         this.session = null;
         this.ort = null;
         this.config = null;
+        this.mlxActive = false;
       }
       // -------------------------------------------------------------------------
       // Text chunking for long TTS input
@@ -32775,6 +32868,10 @@ var init_voice = __esm({
       // Synthesis pipeline
       // -------------------------------------------------------------------------
       async synthesizeAndPlay(text, volume = 1, pitchFactor = 1) {
+        if (this.mlxActive) {
+          await this.synthesizeWithMlx(text, volume, pitchFactor);
+          return;
+        }
         if (!this.session || !this.config || !this.ort)
           return;
         const cleaned = text.replace(/\*/g, "");
@@ -33021,6 +33118,196 @@ var init_voice = __esm({
         }
       }
       // -------------------------------------------------------------------------
+      // MLX Audio backend (macOS Apple Silicon)
+      // -------------------------------------------------------------------------
+      /** Check if we're on macOS Apple Silicon */
+      isMlxCapable() {
+        return platform3() === "darwin" && process.arch === "arm64";
+      }
+      /** Resolve python3 binary path */
+      findPython3() {
+        if (this.python3Path)
+          return this.python3Path;
+        for (const bin of ["python3", "python"]) {
+          try {
+            const path = execSync27(`which ${bin}`, { stdio: "pipe", timeout: 5e3 }).toString().trim();
+            if (path) {
+              this.python3Path = path;
+              return path;
+            }
+          } catch {
+          }
+        }
+        return null;
+      }
+      /** Check if mlx-audio is installed */
+      checkMlxInstalled() {
+        if (this.mlxInstalled !== null)
+          return this.mlxInstalled;
+        const py = this.findPython3();
+        if (!py) {
+          this.mlxInstalled = false;
+          return false;
+        }
+        try {
+          execSync27(`${py} -c "import mlx_audio"`, { stdio: "pipe", timeout: 1e4 });
+          this.mlxInstalled = true;
+          return true;
+        } catch {
+          this.mlxInstalled = false;
+          return false;
+        }
+      }
+      /**
+       * Ensure mlx-audio is installed. Auto-installs via pip on first use.
+       * Only works on macOS Apple Silicon.
+       */
+      async ensureMlxAudio() {
+        if (!this.isMlxCapable()) {
+          throw new Error("MLX Audio requires macOS on Apple Silicon (M1/M2/M3/M4). Use an ONNX voice model instead: glados, overwatch");
+        }
+        const py = this.findPython3();
+        if (!py) {
+          throw new Error("python3 not found. Install Python 3.10+ to use MLX Audio voices. Try: brew install python@3.12");
+        }
+        if (this.checkMlxInstalled())
+          return;
+        renderInfo("Installing MLX Audio for voice synthesis (first time setup)...");
+        try {
+          execSync27(`${py} -m pip install mlx-audio --quiet`, {
+            stdio: "pipe",
+            timeout: 3e5
+            // 5 min — may need to compile
+          });
+          this.mlxInstalled = true;
+          renderInfo("MLX Audio installed successfully.");
+        } catch (err) {
+          try {
+            execSync27(`${py} -m pip install mlx-audio --user --quiet`, {
+              stdio: "pipe",
+              timeout: 3e5
+            });
+            this.mlxInstalled = true;
+            renderInfo("MLX Audio installed successfully (user site-packages).");
+          } catch (err2) {
+            throw new Error(`Failed to install mlx-audio. Try manually: pip install mlx-audio
+Error: ${err2 instanceof Error ? err2.message : String(err2)}`);
+          }
+        }
+      }
+      /**
+       * Synthesize and play audio using MLX Audio CLI.
+       * Generates a WAV file via `python3 -m mlx_audio.tts.generate`, then plays it.
+       */
+      async synthesizeWithMlx(text, volume = 1, _pitchFactor = 1) {
+        const py = this.findPython3();
+        if (!py)
+          return;
+        const cleaned = text.replace(/\*/g, "").trim();
+        if (!cleaned)
+          return;
+        const model = VOICE_MODELS[this.modelId];
+        if (!model)
+          return;
+        const mlxModelId = model.mlxModelId ?? "mlx-community/Kokoro-82M-bf16";
+        const mlxVoice = model.mlxVoice ?? "af_heart";
+        const mlxLangCode = model.mlxLangCode ?? "a";
+        const wavPath = join41(tmpdir6(), `oa-mlx-${Date.now()}.wav`);
+        const pyScript = [
+          "import sys, json",
+          "from mlx_audio.tts import generate as tts_gen",
+          "text = json.loads(sys.argv[1])",
+          `tts_gen.main(["--model", ${JSON.stringify(mlxModelId)}, "--text", text, "--voice", ${JSON.stringify(mlxVoice)}, "--lang_code", ${JSON.stringify(mlxLangCode)}, "--audio_path", ${JSON.stringify(wavPath)}])`
+        ].join("; ");
+        try {
+          execSync27(`${py} -c ${JSON.stringify(pyScript)} ${JSON.stringify(JSON.stringify(cleaned))}`, { stdio: "pipe", timeout: 6e4, cwd: tmpdir6() });
+        } catch (err) {
+          try {
+            const safeText = cleaned.replace(/'/g, "'\\''");
+            execSync27(`${py} -m mlx_audio.tts.generate --model ${mlxModelId} --text '${safeText}' --voice ${mlxVoice} --lang_code ${mlxLangCode} --audio_path ${JSON.stringify(wavPath)}`, { stdio: "pipe", timeout: 6e4, cwd: tmpdir6() });
+          } catch (err2) {
+            return;
+          }
+        }
+        if (!existsSync31(wavPath))
+          return;
+        if (volume !== 1) {
+          try {
+            const wavData = readFileSync22(wavPath);
+            if (wavData.length > 44) {
+              const header = wavData.subarray(0, 44);
+              const samples = new Int16Array(wavData.buffer, wavData.byteOffset + 44, (wavData.length - 44) / 2);
+              for (let i = 0; i < samples.length; i++) {
+                samples[i] = Math.round(samples[i] * volume);
+              }
+              const scaled = Buffer.concat([header, Buffer.from(samples.buffer, samples.byteOffset, samples.byteLength)]);
+              writeFileSync11(wavPath, scaled);
+            }
+          } catch {
+          }
+        }
+        if (this.onPCMOutput) {
+          try {
+            const wavData = readFileSync22(wavPath);
+            if (wavData.length > 44) {
+              const pcm = Buffer.from(wavData.buffer, wavData.byteOffset + 44, wavData.length - 44);
+              const sampleRate = wavData.readUInt32LE(24);
+              this.onPCMOutput(pcm, sampleRate);
+            }
+          } catch {
+          }
+        }
+        await this.playWav(wavPath);
+        try {
+          unlinkSync4(wavPath);
+        } catch {
+        }
+      }
+      /**
+       * Synthesize text to a WAV buffer using MLX Audio (no playback).
+       * Used by synthesizeToBuffer / synthesizeToPCM for MLX models.
+       */
+      async synthesizeMlxToBuffer(text) {
+        const py = this.findPython3();
+        if (!py)
+          return null;
+        const cleaned = text.replace(/\*/g, "").trim();
+        if (!cleaned)
+          return null;
+        const model = VOICE_MODELS[this.modelId];
+        if (!model)
+          return null;
+        const mlxModelId = model.mlxModelId ?? "mlx-community/Kokoro-82M-bf16";
+        const mlxVoice = model.mlxVoice ?? "af_heart";
+        const mlxLangCode = model.mlxLangCode ?? "a";
+        const wavPath = join41(tmpdir6(), `oa-mlx-buf-${Date.now()}.wav`);
+        const pyScript = [
+          "import sys, json",
+          "from mlx_audio.tts import generate as tts_gen",
+          "text = json.loads(sys.argv[1])",
+          `tts_gen.main(["--model", ${JSON.stringify(mlxModelId)}, "--text", text, "--voice", ${JSON.stringify(mlxVoice)}, "--lang_code", ${JSON.stringify(mlxLangCode)}, "--audio_path", ${JSON.stringify(wavPath)}])`
+        ].join("; ");
+        try {
+          execSync27(`${py} -c ${JSON.stringify(pyScript)} ${JSON.stringify(JSON.stringify(cleaned))}`, { stdio: "pipe", timeout: 6e4, cwd: tmpdir6() });
+        } catch {
+          try {
+            const safeText = cleaned.replace(/'/g, "'\\''");
+            execSync27(`${py} -m mlx_audio.tts.generate --model ${mlxModelId} --text '${safeText}' --voice ${mlxVoice} --lang_code ${mlxLangCode} --audio_path ${JSON.stringify(wavPath)}`, { stdio: "pipe", timeout: 6e4, cwd: tmpdir6() });
+          } catch {
+            return null;
+          }
+        }
+        if (!existsSync31(wavPath))
+          return null;
+        try {
+          const data = readFileSync22(wavPath);
+          unlinkSync4(wavPath);
+          return data;
+        } catch {
+          return null;
+        }
+      }
+      // -------------------------------------------------------------------------
       // Setup: ONNX runtime installation
       // -------------------------------------------------------------------------
       async ensureRuntime() {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "open-agents-ai",
-  "version": "0.94.0",
+  "version": "0.95.0",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/index.js",
@@ -56,7 +56,10 @@
     "micropayments",
     "factory-ai",
     "droid",
-    "code-delegation"
+    "code-delegation",
+    "mlx-audio",
+    "apple-silicon",
+    "kokoro-tts"
   ],
   "author": "robit-man",
   "license": "MIT",