Add Requesty as an OpenAI-compatible LLM provider

.env.example

@@ -94,7 +94,7 @@ TTS_VOICE=am_puck
 # =============================================================================
 # LLM Configuration
 # =============================================================================
-# Choose your LLM provider: "ollama" (local), "groq" (cloud), "openrouter" (cloud), or "openai_compatible"
+# Choose your LLM provider: "ollama" (local), "groq" (cloud), "openrouter" (cloud), "requesty" (cloud), or "openai_compatible"
 LLM_PROVIDER=ollama
 
 # =============================================================================
@@ -140,6 +140,14 @@ GROQ_MODEL=llama-3.3-70b-versatile
 OPENROUTER_API_KEY=
 OPENROUTER_MODEL=google/gemini-2.0-flash-001
 
+# =============================================================================
+# LLM - Requesty (Cloud)
+# =============================================================================
+# OpenAI-compatible LLM gateway. Get your API key from https://app.requesty.ai/api-keys
+# Required when LLM_PROVIDER=requesty
+REQUESTY_API_KEY=
+REQUESTY_MODEL=openai/gpt-4o-mini
+
 # =============================================================================
 # LLM - OpenAI-Compatible (Any Provider)
 # =============================================================================

README.md

@@ -23,7 +23,7 @@ ollama pull coreworxlab/caal-qwen3.5-9b
 
 **Infinitely extensible.** Any [n8n](https://n8n.io/) workflow becomes a voice-activated tool. Control [Home Assistant](https://www.home-assistant.io/) devices, query APIs, automate your life — then share your tools with the community via the [CAAL Tool Registry](https://github.com/CoreWorxLab/caal-tools). Tools follow a suite convention — fewer tools, better accuracy, more reliable routing.
 
-**Local by default.** Runs fully on your network with [Ollama](https://ollama.ai/). No accounts, no telemetry, no cloud dependency. Want to use [Groq](https://groq.com/), [OpenRouter](https://openrouter.ai/), or any OpenAI-compatible API? Your choice. Your credentials and tool executions never leave your network regardless.
+**Local by default.** Runs fully on your network with [Ollama](https://ollama.ai/). No accounts, no telemetry, no cloud dependency. Want to use [Groq](https://groq.com/), [OpenRouter](https://openrouter.ai/), [Requesty](https://requesty.ai/), or any OpenAI-compatible API? Your choice. Your credentials and tool executions never leave your network regardless.
 
 ---
 
@@ -33,7 +33,7 @@ ollama pull coreworxlab/caal-qwen3.5-9b
 - **Tool Chaining** — Sequential multi-tool calls in one prompt. The model uses real data from each step to inform the next
 - **Home Assistant** — Voice control across lights, covers, locks, climate, media, and more via `hass`
 - **n8n Workflows** — Any workflow becomes a tool. Visual, inspectable, shareable, auditable through n8n's execution history
-- **Flexible Providers** — Ollama, Groq, OpenRouter, or any OpenAI-compatible API. Speaches or Groq for STT. Kokoro or Piper for TTS
+- **Flexible Providers** — Ollama, Groq, OpenRouter, Requesty, or any OpenAI-compatible API. Speaches or Groq for STT. Kokoro or Piper for TTS
 - **Short-Term Memory** — Store and recall information across sessions
 - **Internationalization** — English, French, Italian, with more coming
 - **Wake Word** — "Hey Cal" via OpenWakeWord
@@ -165,6 +165,7 @@ CAAL is built in the open. If you build a tool, we see the PR. If you find a bug
 - [Ollama](https://ollama.ai/) - Local LLM server
 - [Groq](https://groq.com/) - Fast cloud LLM inference
 - [OpenRouter](https://openrouter.ai/) - Unified API for 200+ models
+- [Requesty](https://requesty.ai/) - OpenAI-compatible LLM gateway
 - [n8n](https://n8n.io/) - Workflow automation
 - [Home Assistant](https://www.home-assistant.io/) - Smart home platform
 

src/caal/chat/api.py

@@ -179,6 +179,15 @@ def _get_runtime_settings() -> dict:
             user_settings.get("openrouter_model")
             or os.getenv("OPENROUTER_MODEL", "openai/gpt-4")
         ),
+        # Requesty
+        "requesty_api_key": (
+            settings.get("requesty_api_key")
+            or os.getenv("REQUESTY_API_KEY", "")
+        ),
+        "requesty_model": (
+            user_settings.get("requesty_model")
+            or os.getenv("REQUESTY_MODEL", "openai/gpt-4o-mini")
+        ),
         # Shared
         "max_turns": settings.get(
             "max_turns", int(os.getenv("OLLAMA_MAX_TURNS", "20"))
@@ -536,6 +545,7 @@ async def reload_chat() -> ReloadResponse:
             or runtime.get("groq_model")
             or runtime.get("openai_model")
             or runtime.get("openrouter_model")
+            or runtime.get("requesty_model")
             or ""
         ),
         tools_loaded=tools_loaded,

src/caal/llm/providers/__init__.py

@@ -9,6 +9,7 @@
     - GroqProvider: Groq cloud API
     - OpenAICompatibleProvider: Any OpenAI-compatible server
     - OpenRouterProvider: OpenRouter cloud API (400+ models)
+    - RequestyProvider: Requesty cloud API (OpenAI-compatible gateway)
 
 Example:
     >>> from caal.llm.providers import create_provider
@@ -26,6 +27,10 @@
     >>> # Create OpenRouter provider
     >>> provider = create_provider("openrouter", model="openai/gpt-4",
     ...                            api_key="sk-...")
+    >>>
+    >>> # Create Requesty provider
+    >>> provider = create_provider("requesty", model="openai/gpt-4o-mini",
+    ...                            api_key="sk-...")
 """
 
 from __future__ import annotations
@@ -39,6 +44,7 @@
 from .ollama_provider import OllamaProvider
 from .openai_compatible_provider import OpenAICompatibleProvider
 from .openrouter_provider import OpenRouterProvider
+from .requesty_provider import RequestyProvider
 
 __all__ = [
     "LLMProvider",
@@ -48,6 +54,7 @@
     "GroqProvider",
     "OpenAICompatibleProvider",
     "OpenRouterProvider",
+    "RequestyProvider",
     "create_provider",
 ]
 
@@ -62,7 +69,7 @@ def create_provider(
 
     Args:
         provider_name: Provider identifier ("ollama", "groq", "openai_compatible",
-            or "openrouter")
+            "openrouter", or "requesty")
         **kwargs: Provider-specific configuration options
 
     Returns:
@@ -94,10 +101,12 @@ def create_provider(
         return OpenAICompatibleProvider(**kwargs)
     elif provider_name == "openrouter":
         return OpenRouterProvider(**kwargs)
+    elif provider_name == "requesty":
+        return RequestyProvider(**kwargs)
     else:
         raise ValueError(
             f"Unknown LLM provider: {provider_name}. "
-            f"Supported providers: ollama, groq, openai_compatible, openrouter"
+            f"Supported providers: ollama, groq, openai_compatible, openrouter, requesty"
         )
 
 
@@ -109,14 +118,17 @@ def create_provider_from_settings(settings: dict[str, Any]) -> LLMProvider:
 
     Args:
         settings: Runtime settings dict with keys like:
-            - llm_provider: "ollama", "groq", "openai_compatible", or "openrouter"
+            - llm_provider: "ollama", "groq", "openai_compatible", "openrouter",
+              or "requesty"
             - ollama_model: Ollama model name
             - groq_model: Groq model name
             - openai_model: OpenAI-compatible model name
             - openai_base_url: OpenAI-compatible server URL
             - openai_api_key: OpenAI-compatible API key (optional)
             - openrouter_model: OpenRouter model name
             - openrouter_api_key: OpenRouter API key (required)
+            - requesty_model: Requesty model name
+            - requesty_api_key: Requesty API key (required)
             - temperature: Sampling temperature
             - num_ctx: Context window size (Ollama only)
 
@@ -168,8 +180,21 @@ def create_provider_from_settings(settings: dict[str, Any]) -> LLMProvider:
             api_key=api_key,
             temperature=settings.get("temperature", 0.7),
         )
+    elif provider_name == "requesty":
+        # API key from settings, fallback to environment variable
+        api_key = settings.get("requesty_api_key") or os.environ.get("REQUESTY_API_KEY")
+        if not api_key:
+            raise ValueError(
+                "Requesty API key required. Set requesty_api_key in settings "
+                "or REQUESTY_API_KEY environment variable."
+            )
+        return RequestyProvider(
+            model=settings.get("requesty_model", "openai/gpt-4o-mini"),
+            api_key=api_key,
+            temperature=settings.get("temperature", 0.7),
+        )
     else:
         raise ValueError(
             f"Unknown LLM provider: {provider_name}. "
-            f"Supported providers: ollama, groq, openai_compatible, openrouter"
+            f"Supported providers: ollama, groq, openai_compatible, openrouter, requesty"
         )

src/caal/llm/providers/requesty_provider.py

@@ -0,0 +1,75 @@
+"""Requesty LLM provider implementation.
+
+Provides access to many cloud models through Requesty's unified,
+OpenAI-compatible LLM gateway. Inherits from OpenAICompatibleProvider
+since Requesty exposes an OpenAI-compatible API.
+
+See: https://docs.requesty.ai
+"""
+
+from __future__ import annotations
+
+import logging
+
+from .openai_compatible_provider import OpenAICompatibleProvider
+
+__all__ = ["RequestyProvider"]
+
+logger = logging.getLogger(__name__)
+
+REQUESTY_BASE_URL = "https://router.requesty.ai/v1"
+
+
+class RequestyProvider(OpenAICompatibleProvider):
+    """Requesty cloud LLM provider.
+
+    Provides access to many models including GPT-4, Claude, Llama, Mistral,
+    and many others through a single API. Inherits all chat, streaming, and
+    tool handling from OpenAICompatibleProvider.
+
+    Differences from base OpenAICompatibleProvider:
+        - Fixed base URL (router.requesty.ai)
+        - API key is required
+        - Attribution headers for model providers
+
+    Args:
+        model: Requesty model ID (e.g., "openai/gpt-4o-mini", "anthropic/claude-sonnet-4-5")
+        api_key: Requesty API key (required)
+        temperature: Sampling temperature (0.0-2.0)
+        max_tokens: Maximum tokens to generate
+
+    See: https://app.requesty.ai/router/list
+    """
+
+    def __init__(
+        self,
+        model: str,
+        api_key: str,
+        temperature: float = 0.7,
+        max_tokens: int = 4096,
+    ) -> None:
+        if not api_key:
+            raise ValueError(
+                "Requesty API key required. Pass api_key parameter "
+                "or configure in settings."
+            )
+
+        super().__init__(
+            model=model,
+            base_url=REQUESTY_BASE_URL,
+            api_key=api_key,
+            temperature=temperature,
+            max_tokens=max_tokens,
+        )
+
+        # Add Requesty-specific attribution headers
+        self._client.default_headers.update({
+            "HTTP-Referer": "https://github.com/caal-project/caal",
+            "X-Title": "CAAL Voice Assistant",
+        })
+
+        logger.debug(f"RequestyProvider initialized: {model}")
+
+    @property
+    def provider_name(self) -> str:
+        return "requesty"

src/caal/settings.py

@@ -111,6 +111,9 @@ def validate_url(url: str) -> tuple[bool, str]:
     # OpenRouter settings (cloud API)
     "openrouter_api_key": "",     # OpenRouter API key (required for openrouter provider)
     "openrouter_model": "",       # Model name (empty = use default)
+    # Requesty settings (cloud API)
+    "requesty_api_key": "",       # Requesty API key (required for requesty provider)
+    "requesty_model": "",         # Model name (empty = use default)
 }
 
 # Per-language Piper TTS voice mapping

voice_agent.py

@@ -176,6 +176,15 @@ def get_runtime_settings() -> dict:
             user_settings.get("openrouter_model")
             or os.getenv("OPENROUTER_MODEL", "openai/gpt-4")
         ),
+        # Requesty settings
+        "requesty_api_key": (
+            settings.get("requesty_api_key")
+            or os.getenv("REQUESTY_API_KEY", "")
+        ),
+        "requesty_model": (
+            user_settings.get("requesty_model")
+            or os.getenv("REQUESTY_MODEL", "openai/gpt-4o-mini")
+        ),
         # Shared settings
         "max_turns": settings.get("max_turns", int(os.getenv("OLLAMA_MAX_TURNS", "20"))),
         "tool_cache_size": settings.get("tool_cache_size", int(os.getenv("TOOL_CACHE_SIZE", "3"))),
@@ -384,6 +393,10 @@ async def entrypoint(ctx: agents.JobContext) -> None:
         logger.info(
             f"  LLM: OpenRouter ({runtime.get('openrouter_model', '?')})"
         )
+    elif llm_provider == "requesty":
+        logger.info(
+            f"  LLM: Requesty ({runtime.get('requesty_model', '?')})"
+        )
     logger.info(f"  MCP: {list(mcp_servers.keys()) or 'None'}")
     logger.info(
         f"  Turn detection: interruptions={runtime['allow_interruptions']}, "