LLMIFY THIS MOFO

internal/config/config_test.go

@@ -696,6 +696,31 @@ func TestConfig_ConversionMethods(t *testing.T) {
 	})
 }
 
+func TestConfig_LlamaSwap(t *testing.T) {
+	cfg := createTestConfig()
+	cfg.Transcription.Provider = "llama-swap"
+	cfg.Transcription.Model = "whisper-large-v3-turbo"
+	cfg.Providers = map[string]ProviderConfig{
+		"llama-swap": {APIKey: "test-key", BaseURL: "http://192.168.1.50:8080"},
+	}
+	cfg.LLM = LLMConfig{Enabled: true, Provider: "llama-swap", Model: "qwen3"}
+
+	if err := cfg.Validate(); err != nil {
+		t.Fatalf("Validate() error = %v", err)
+	}
+	if got := cfg.ToTranscriberConfig().BaseURL; got != "http://192.168.1.50:8080" {
+		t.Errorf("transcriber BaseURL = %q", got)
+	}
+	if got := cfg.ToLLMConfig().BaseURL; got != "http://192.168.1.50:8080" {
+		t.Errorf("LLM BaseURL = %q", got)
+	}
+
+	cfg.Providers["llama-swap"] = ProviderConfig{APIKey: "test-key", BaseURL: "http://192.168.1.50:8080/v1"}
+	if err := cfg.Validate(); err == nil || !strings.Contains(err.Error(), "omit the /v1 suffix") {
+		t.Errorf("Validate() error = %v, want /v1 validation error", err)
+	}
+}
+
 func TestValidateModelLanguageCompatibility(t *testing.T) {
 	tests := []struct {
 		name        string

internal/config/convert.go

@@ -32,6 +32,9 @@ func (c *Config) ToTranscriberConfig() transcriber.Config {
 	}
 
 	config.APIKey = c.resolveAPIKeyForProvider(c.Transcription.Provider)
+	if c.Transcription.Provider == provider.ProviderLlamaSwap {
+		config.BaseURL = c.Providers[provider.ProviderLlamaSwap].BaseURL
+	}
 
 	return config
 }
@@ -74,6 +77,9 @@ func (c *Config) ToLLMConfig() LLMAdapterConfig {
 	if c.LLM.Provider != "" {
 		config.APIKey = c.resolveAPIKeyForLLMProvider(c.LLM.Provider)
 	}
+	if c.LLM.Provider == provider.ProviderLlamaSwap {
+		config.BaseURL = c.Providers[provider.ProviderLlamaSwap].BaseURL
+	}
 
 	if c.LLM.CustomPrompt.Enabled && c.LLM.CustomPrompt.Prompt != "" {
 		config.CustomPrompt = c.LLM.CustomPrompt.Prompt
@@ -106,9 +112,10 @@ func (c *Config) IsLLMEnabled() bool {
 
 func (c *Config) ToInjectionConfig() injection.Config {
 	return injection.Config{
-		Backends:         c.Injection.Backends,
-		YdotoolTimeout:   c.Injection.YdotoolTimeout,
-		WtypeTimeout:     c.Injection.WtypeTimeout,
-		ClipboardTimeout: c.Injection.ClipboardTimeout,
+		Backends:          c.Injection.Backends,
+		YdotoolTimeout:    c.Injection.YdotoolTimeout,
+		WtypeTimeout:      c.Injection.WtypeTimeout,
+		ClipboardTimeout:  c.Injection.ClipboardTimeout,
+		CtrlShiftVClasses: c.Injection.CtrlShiftVClasses,
 	}
 }

internal/config/defaults.go

@@ -20,10 +20,11 @@ func DefaultConfig() *Config {
 			Threads:   0,
 		},
 		Injection: InjectionConfig{
-			Backends:         []string{"ydotool", "wtype", "clipboard"},
-			YdotoolTimeout:   5 * time.Second,
-			WtypeTimeout:     5 * time.Second,
-			ClipboardTimeout: 3 * time.Second,
+			Backends:          []string{"ydotool", "wtype", "clipboard"},
+			YdotoolTimeout:    5 * time.Second,
+			WtypeTimeout:      5 * time.Second,
+			ClipboardTimeout:  3 * time.Second,
+			CtrlShiftVClasses: []string{"ghostty"},
 		},
 		Notifications: NotificationsConfig{
 			Enabled: false,

internal/config/save.go

@@ -47,6 +47,9 @@ func Save(cfg *Config) error {
 		for name, pc := range cfg.Providers {
 			sb.WriteString(fmt.Sprintf("[providers.%s]\n", name))
 			sb.WriteString(fmt.Sprintf("  api_key = %q\n", pc.APIKey))
+			if pc.BaseURL != "" {
+				sb.WriteString(fmt.Sprintf("  base_url = %q\n", pc.BaseURL))
+			}
 			sb.WriteString("\n")
 		}
 	}
@@ -117,6 +120,14 @@ func Save(cfg *Config) error {
 	sb.WriteString(fmt.Sprintf("  ydotool_timeout = %q\n", cfg.Injection.YdotoolTimeout.String()))
 	sb.WriteString(fmt.Sprintf("  wtype_timeout = %q\n", cfg.Injection.WtypeTimeout.String()))
 	sb.WriteString(fmt.Sprintf("  clipboard_timeout = %q\n", cfg.Injection.ClipboardTimeout.String()))
+	sb.WriteString("  ctrl_shift_v_classes = [")
+	for i, class := range cfg.Injection.CtrlShiftVClasses {
+		if i > 0 {
+			sb.WriteString(", ")
+		}
+		sb.WriteString(fmt.Sprintf("%q", class))
+	}
+	sb.WriteString("]\n")
 	sb.WriteString("\n")
 
 	// Notifications
@@ -221,6 +232,9 @@ keywords = []
 #   api_key = ""               # ElevenLabs API key (or set ELEVENLABS_API_KEY env var)
 # [providers.deepgram]
 #   api_key = ""               # Deepgram API key (or set DEEPGRAM_API_KEY env var)
+# [providers.llama-swap]
+#   api_key = ""               # Or set LLAMA_SWAP_API_KEY
+#   base_url = "http://llama-swap.example:8080" # No /v1 suffix
 
 # ─────────────────────────────────────────────────────────────────────────────
 # Audio Recording
@@ -241,7 +255,7 @@ keywords = []
 # ─────────────────────────────────────────────────────────────────────────────
 
 [transcription]
-  provider = "openai"          # "openai", "groq-transcription", "mistral-transcription", "elevenlabs", "whisper-cpp"
+  provider = "openai"          # Also "llama-swap" for a remote OpenAI-compatible LlamaSwap server
   model = "whisper-1"          # Model: OpenAI="whisper-1", Groq="whisper-large-v3", Mistral="voxtral-mini-latest", ElevenLabs="scribe_v1"
   language = ""                # ISO 639-1 code (e.g., en, es, de). Empty for auto-detect.
   threads = 0                  # CPU threads for local transcription (0 = auto: uses NumCPU-1)
@@ -253,7 +267,7 @@ keywords = []
 
 [llm]
   enabled = true               # Enable LLM post-processing (highly recommended)
-  provider = "openai"          # "openai" or "groq" (must have API key configured above)
+  provider = "openai"          # "openai", "groq", or "llama-swap" (must have API key configured above)
   model = "gpt-4o-mini"        # OpenAI: "gpt-4o-mini", Groq: "llama-3.3-70b-versatile"
 
 [llm.post_processing]
@@ -272,7 +286,7 @@ keywords = []
 # ─────────────────────────────────────────────────────────────────────────────
 
 [injection]
-  backends = ["ydotool", "wtype", "clipboard"]  # Ordered fallback chain (tries each until one succeeds)
+  backends = ["clipboard-paste", "ydotool", "wtype", "clipboard"]  # Ordered fallback chain (tries each until one succeeds)
   ydotool_timeout = "5s"       # Timeout for ydotool commands
   wtype_timeout = "5s"         # Timeout for wtype commands
   clipboard_timeout = "3s"     # Timeout for clipboard operations
@@ -322,12 +336,16 @@ keywords = []
 # - "groq-transcription": Groq Whisper API (very fast, models: whisper-large-v3, whisper-large-v3-turbo)
 # - "mistral-transcription": Mistral Voxtral API (excellent for European languages, model: voxtral-mini-latest)
 # - "elevenlabs": ElevenLabs Scribe API (99 languages, models: scribe_v1, scribe_v2, scribe_v2_realtime)
+# - "llama-swap": Remote OpenAI-compatible LlamaSwap (any configured transcription model)
 #
 # LLM providers (for post-processing):
 # - "openai": GPT models (gpt-4o-mini recommended for cost/quality balance)
 # - "groq": Fast inference (llama-3.3-70b-versatile recommended)
+# - "llama-swap": Any chat model configured in your LlamaSwap server
 #
 # Injection backends:
+# - "clipboard-paste": Temporarily uses the regular clipboard + wtype Ctrl+V, then restores prior text clipboard.
+# - ctrl_shift_v_classes: Window-class substrings that use Ctrl+Shift+V instead (default: ["ghostty"]).
 # - "ydotool": Uses ydotool (requires ydotoold daemon). Best for Chromium/Electron apps.
 # - "wtype": Uses wtype for Wayland. May have issues with some Chromium apps.
 # - "clipboard": Copies to clipboard only (most reliable, requires manual paste).

internal/config/types.go

@@ -25,7 +25,8 @@ type Config struct {
 
 // ProviderConfig holds API key for a provider
 type ProviderConfig struct {
-	APIKey string `toml:"api_key"`
+	APIKey  string `toml:"api_key"`
+	BaseURL string `toml:"base_url"` // OpenAI-compatible base URL, without /v1
 }
 
 // LLMConfig configures the LLM post-processing phase
@@ -70,10 +71,11 @@ type TranscriptionConfig struct {
 }
 
 type InjectionConfig struct {
-	Backends         []string      `toml:"backends"`
-	YdotoolTimeout   time.Duration `toml:"ydotool_timeout"`
-	WtypeTimeout     time.Duration `toml:"wtype_timeout"`
-	ClipboardTimeout time.Duration `toml:"clipboard_timeout"`
+	Backends          []string      `toml:"backends"`
+	YdotoolTimeout    time.Duration `toml:"ydotool_timeout"`
+	WtypeTimeout      time.Duration `toml:"wtype_timeout"`
+	ClipboardTimeout  time.Duration `toml:"clipboard_timeout"`
+	CtrlShiftVClasses []string      `toml:"ctrl_shift_v_classes"`
 }
 
 type NotificationsConfig struct {
@@ -139,4 +141,5 @@ type LLMAdapterConfig struct {
 	RemoveFillerWords bool
 	CustomPrompt      string
 	Keywords          []string
+	BaseURL           string
 }

internal/config/validate.go

@@ -2,6 +2,7 @@ package config
 
 import (
 	"fmt"
+	"net/url"
 	"strings"
 
 	"github.com/leonardotrapani/hyprvoice/internal/provider"
@@ -34,6 +35,8 @@ func envVarForProvider(registryName string) string {
 		return "ELEVENLABS_API_KEY"
 	case "deepgram":
 		return "DEEPGRAM_API_KEY"
+	case "llama-swap":
+		return "LLAMA_SWAP_API_KEY"
 	default:
 		return ""
 	}
@@ -82,27 +85,40 @@ func (c *Config) Validate() error {
 				strings.Title(registryName), registryName, envVar)
 		}
 	}
+	if registryName == provider.ProviderLlamaSwap {
+		if err := validateLlamaSwapBaseURL(c.Providers[provider.ProviderLlamaSwap].BaseURL); err != nil {
+			return err
+		}
+	}
 
 	// validate model exists
 	if c.Transcription.Model == "" {
 		return fmt.Errorf("invalid transcription.model: empty")
 	}
 
-	// validate model exists in provider
-	_, err := provider.GetModel(registryName, c.Transcription.Model)
-	if err != nil {
-		models := provider.ModelsOfType(p, provider.Transcription)
-		modelIDs := make([]string, len(models))
-		for i, m := range models {
-			modelIDs[i] = m.ID
+	// LlamaSwap is an OpenAI-compatible router: its model IDs are defined by the
+	// remote server, so they cannot be validated against Hyprvoice's static registry.
+	if registryName == provider.ProviderLlamaSwap {
+		if c.Transcription.Streaming {
+			return fmt.Errorf("llama-swap transcription supports batch mode only (set transcription.streaming = false)")
+		}
+	} else {
+		// validate model exists in provider
+		_, err := provider.GetModel(registryName, c.Transcription.Model)
+		if err != nil {
+			models := provider.ModelsOfType(p, provider.Transcription)
+			modelIDs := make([]string, len(models))
+			for i, m := range models {
+				modelIDs[i] = m.ID
+			}
+			return fmt.Errorf("invalid model for %s: %s (available: %s)", c.Transcription.Provider, c.Transcription.Model, strings.Join(modelIDs, ", "))
 		}
-		return fmt.Errorf("invalid model for %s: %s (available: %s)", c.Transcription.Provider, c.Transcription.Model, strings.Join(modelIDs, ", "))
-	}
 
-	// validate language-model compatibility using effective language (transcription overrides general)
-	effectiveLanguage := c.resolveEffectiveLanguage()
-	if err := ValidateModelLanguageCompatibility(registryName, c.Transcription.Model, effectiveLanguage); err != nil {
-		return err
+		// validate language-model compatibility using effective language (transcription overrides general)
+		effectiveLanguage := c.resolveEffectiveLanguage()
+		if err := ValidateModelLanguageCompatibility(registryName, c.Transcription.Model, effectiveLanguage); err != nil {
+			return err
+		}
 	}
 
 	// LLM validation
@@ -121,20 +137,26 @@ func (c *Config) Validate() error {
 			return fmt.Errorf("invalid llm.provider: %s (available: %s)", c.LLM.Provider, strings.Join(providers, ", "))
 		}
 
-		// validate LLM model exists
-		llmModel, err := provider.GetModel(c.LLM.Provider, c.LLM.Model)
-		if err != nil {
-			models := provider.ModelsOfType(llmProvider, provider.LLM)
-			modelIDs := make([]string, len(models))
-			for i, m := range models {
-				modelIDs[i] = m.ID
+		if c.LLM.Provider == provider.ProviderLlamaSwap {
+			if err := validateLlamaSwapBaseURL(c.Providers[provider.ProviderLlamaSwap].BaseURL); err != nil {
+				return err
+			}
+		} else {
+			// validate LLM model exists
+			llmModel, err := provider.GetModel(c.LLM.Provider, c.LLM.Model)
+			if err != nil {
+				models := provider.ModelsOfType(llmProvider, provider.LLM)
+				modelIDs := make([]string, len(models))
+				for i, m := range models {
+					modelIDs[i] = m.ID
+				}
+				return fmt.Errorf("invalid llm.model: %s (available for %s: %s)", c.LLM.Model, c.LLM.Provider, strings.Join(modelIDs, ", "))
 			}
-			return fmt.Errorf("invalid llm.model: %s (available for %s: %s)", c.LLM.Model, c.LLM.Provider, strings.Join(modelIDs, ", "))
-		}
 
-		// verify model is actually an LLM
-		if llmModel.Type != provider.LLM {
-			return fmt.Errorf("invalid llm.model: %s is not an LLM model", c.LLM.Model)
+			// verify model is actually an LLM
+			if llmModel.Type != provider.LLM {
+				return fmt.Errorf("invalid llm.model: %s is not an LLM model", c.LLM.Model)
+			}
 		}
 
 		// validate LLM API key
@@ -151,10 +173,10 @@ func (c *Config) Validate() error {
 	if len(c.Injection.Backends) == 0 {
 		return fmt.Errorf("invalid injection.backends: empty (must have at least one backend)")
 	}
-	validBackends := map[string]bool{"ydotool": true, "wtype": true, "clipboard": true}
+	validBackends := map[string]bool{"clipboard-paste": true, "ydotool": true, "wtype": true, "clipboard": true}
 	for _, backend := range c.Injection.Backends {
 		if !validBackends[backend] {
-			return fmt.Errorf("invalid injection.backends: unknown backend %q (must be ydotool, wtype, or clipboard)", backend)
+			return fmt.Errorf("invalid injection.backends: unknown backend %q (must be clipboard-paste, ydotool, wtype, or clipboard)", backend)
 		}
 	}
 	if c.Injection.YdotoolTimeout <= 0 {
@@ -175,6 +197,20 @@ func (c *Config) Validate() error {
 	return nil
 }
 
+func validateLlamaSwapBaseURL(baseURL string) error {
+	if baseURL == "" {
+		return fmt.Errorf("llama-swap base_url required in providers.llama-swap.base_url")
+	}
+	u, err := url.Parse(baseURL)
+	if err != nil || u.Scheme == "" || u.Host == "" {
+		return fmt.Errorf("invalid llama-swap base_url: %q (expected http://host:port, without /v1)", baseURL)
+	}
+	if strings.TrimRight(u.Path, "/") == "/v1" {
+		return fmt.Errorf("invalid llama-swap base_url: omit the /v1 suffix")
+	}
+	return nil
+}
+
 // ValidateModelLanguageCompatibility validates that a model supports the given language.
 // Returns error if the language is not supported, nil if supported or if langCode is empty (auto).
 func ValidateModelLanguageCompatibility(registryProvider, modelID, langCode string) error {