LLMIFY THIS MOFO

docs/config.md

@@ -91,12 +91,16 @@ Hyprvoice uses a unified provider system where API keys are configured once and
 
 [providers.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" # Do not include /v1
 ```
 
 **API key resolution order:**
 
 1. `[providers.X]` section in config
-2. Environment variable (`OPENAI_API_KEY`, `GROQ_API_KEY`, etc.)
+2. Environment variable (`OPENAI_API_KEY`, `GROQ_API_KEY`, `LLAMA_SWAP_API_KEY`, etc.)
 
 ## Transcription Providers
 
@@ -104,6 +108,29 @@ Hyprvoice supports multiple transcription backends. See [docs/providers.md](./pr
 
 ### Cloud Providers
 
+### LlamaSwap (network-hosted OpenAI-compatible models)
+
+LlamaSwap proxies OpenAI-compatible endpoints, including `/v1/audio/transcriptions` and `/v1/chat/completions`. Configure its host once, then set the model IDs exactly as they appear in LlamaSwap's `/v1/models` response:
+
+```toml
+[providers.llama-swap]
+api_key = "your-llama-swap-api-key"
+base_url = "http://192.168.1.50:8080" # No trailing /v1
+
+[transcription]
+provider = "llama-swap"
+model = "whisper-large-v3-turbo"
+language = ""
+streaming = false
+
+[llm]
+enabled = true
+provider = "llama-swap"
+model = "your-chat-model-id"
+```
+
+The transcription and chat model IDs are deliberately not restricted by Hyprvoice: LlamaSwap selects from the models configured on your server. Hyprvoice submits transcription after recording ends, so set `streaming = false`.
+
 ### OpenAI Whisper API
 
 Cloud-based transcription using OpenAI's Whisper API:
@@ -455,7 +482,7 @@ Configurable text injection with multiple backends:
 
 ```toml
 [injection]
-backends = ["ydotool", "wtype", "clipboard"]  # Ordered fallback chain
+backends = ["clipboard-paste", "ydotool", "wtype", "clipboard"]  # Ordered fallback chain
 ydotool_timeout = "5s"
 wtype_timeout = "5s"
 clipboard_timeout = "3s"
@@ -464,6 +491,7 @@ clipboard_timeout = "3s"
 ### Injection Backends
 
 - **`ydotool`**: Uses ydotool (requires `ydotoold` daemon for ydotool v1.0.0+). Most compatible with Chromium/Electron apps.
+- **`clipboard-paste`**: Temporarily copies text to the regular clipboard, sends Ctrl+V through wtype, then restores the prior text clipboard. This works in browsers and Electron apps and is layout-independent; recommended for Dvorak/Colemak users. Requires `wl-clipboard` and `wtype`.
 - **`wtype`**: Uses wtype for Wayland. May have issues with some Chromium-based apps (known upstream bug).
 - **`clipboard`**: Copies text to clipboard only. Most reliable, but requires manual paste.