fix: MetalRT-only mode no longer requires llama.cpp models

When engine preference is "metalrt", the pipeline now treats llama.cpp
STT/LLM/TTS init failures as warnings instead of fatal errors. This
allows users who only selected MetalRT in setup to run without needing
any GGUF models, Zipformer, or Piper TTS files on disk.

Made-with: Cursor

Author: shubhammalhotra28

CMakeLists.txt

@@ -5,7 +5,7 @@ if(POLICY CMP0177)
 endif()
 set(CMAKE_POLICY_VERSION_MINIMUM 3.5 CACHE STRING "" FORCE)
 
-project(rcli VERSION 0.2.9 LANGUAGES C CXX)
+project(rcli VERSION 0.3.0 LANGUAGES C CXX)
 
 set(CMAKE_CXX_STANDARD 17)
 set(CMAKE_CXX_STANDARD_REQUIRED ON)

src/pipeline/orchestrator.cpp

@@ -43,23 +43,26 @@ bool Orchestrator::init(const PipelineConfig& config) {
             pool_->total_size() / (1024.0*1024.0),
             pool_->utilization_pct());
 
+    // llama.cpp STT/LLM/TTS — required for llamacpp/auto, optional for metalrt-only
+    bool need_llamacpp = (config.llm_backend != LlmBackend::METALRT);
+
     if (!stt_.init(config.stt)) {
-        LOG_ERROR("Pipeline", "STT init failed");
-        return false;
+        if (need_llamacpp) { LOG_ERROR("Pipeline", "STT init failed"); return false; }
+        LOG_WARN("Pipeline", "llama.cpp STT not available (MetalRT will handle STT)");
     }
 
     if (!offline_stt_.init(config.offline_stt)) {
         LOG_WARN("Pipeline", "Offline STT init failed (will use streaming STT)");
     }
 
     if (!llm_.init(config.llm)) {
-        LOG_ERROR("Pipeline", "LLM init failed");
-        return false;
+        if (need_llamacpp) { LOG_ERROR("Pipeline", "LLM init failed"); return false; }
+        LOG_WARN("Pipeline", "llama.cpp LLM not available (MetalRT will handle LLM)");
     }
 
     if (!tts_.init(config.tts)) {
-        LOG_ERROR("Pipeline", "TTS init failed");
-        return false;
+        if (need_llamacpp) { LOG_ERROR("Pipeline", "TTS init failed"); return false; }
+        LOG_WARN("Pipeline", "llama.cpp TTS not available (MetalRT will handle TTS)");
     }
 
     if (!audio_.init(config.audio, capture_rb_.get(), playback_rb_.get())) {
@@ -72,13 +75,13 @@ bool Orchestrator::init(const PipelineConfig& config) {
     }
 
     tools_.register_defaults();
-    tools_.set_model_profile(&llm_.profile());
 
-    // Cache tool-aware system prompt (includes tool definitions) in KV cache.
-    // When external tool defs are set, they'll be included automatically.
-    std::string tool_system = llm_.profile().build_tool_system_prompt(
-        config.system_prompt, tools_.get_tool_definitions_json());
-    llm_.cache_system_prompt(tool_system);
+    if (llm_.is_initialized()) {
+        tools_.set_model_profile(&llm_.profile());
+        std::string tool_system = llm_.profile().build_tool_system_prompt(
+            config.system_prompt, tools_.get_tool_definitions_json());
+        llm_.cache_system_prompt(tool_system);
+    }
 
     // --- MetalRT backend (optional) ---
     if (config.llm_backend == LlmBackend::METALRT ||