Release v2.1.1: Token Handling Improvement and DeepSeek Support

- Completely reworked token handling mechanism
- Removed custom token calculation logic
- Direct max_tokens passing to LLM APIs
- Added support for DeepSeek provider
- Integrated deepseek-chat and deepseek-reasoner models

Thanks to @estiens for reporting token handling issues and providing valuable feedback (#1).

README.md

@@ -108,6 +108,7 @@ Transform your smart home experience with powerful AI assistance powered by mult
 - Active API key from:
   - OpenAI ([Get key](https://platform.openai.com/account/api-keys))
   - Anthropic ([Get key](https://console.anthropic.com/))
+  - DeepSeek 🆕 ([Get key](https://platform.deepseek.com/api_keys)) 
   - OpenRouter ([Get key](https://openrouter.ai/keys))
   - Any OpenAI-compatible API provider
 - Python 3.9 or newer
@@ -120,7 +121,7 @@ Transform your smart home experience with powerful AI assistance powered by mult
 - 🔑 **API Key**: Provider-specific authentication
 - 🤖 **Model Selection**: Flexible, provider-specific models
 - 🌡️ **Temperature**: Creativity control (0.0-2.0)
-- 📏 **Max Tokens**: Response length limit (token usage is estimated using a heuristic method based on word count and specific word characteristics, which may differ from actual token usage)
+- 📏 **Max Tokens**: Response length limit (passed directly to the LLM API to control the maximum length of the response)
 - ⏱️ **Request Interval**: API call throttling
 - 💾 **History Size**: Number of messages to retain
 - 🌍 **Custom API Endpoint**: Optional advanced configuration

custom_components/ha_text_ai/__init__.py

@@ -39,11 +39,14 @@
     CONF_CONTEXT_MESSAGES,
     API_PROVIDER_OPENAI,
     API_PROVIDER_ANTHROPIC,
+    API_PROVIDER_DEEPSEEK,
     DEFAULT_MODEL,
+    DEFAULT_DEEPSEEK_MODEL,
     DEFAULT_TEMPERATURE,
     DEFAULT_MAX_TOKENS,
     DEFAULT_OPENAI_ENDPOINT,
     DEFAULT_ANTHROPIC_ENDPOINT,
+    DEFAULT_DEEPSEEK_ENDPOINT,
     DEFAULT_REQUEST_INTERVAL,
     DEFAULT_CONTEXT_MESSAGES,
     API_TIMEOUT,
@@ -235,6 +238,8 @@ async def async_check_api(session, endpoint: str, headers: dict, provider: str)
     try:
         if provider == API_PROVIDER_ANTHROPIC:
             check_url = f"{endpoint}/v1/models"
+        elif provider == API_PROVIDER_DEEPSEEK:
+            check_url = f"{endpoint}/models"  # DeepSeek
         else:  # OpenAI
             check_url = f"{endpoint}/models"
 

custom_components/ha_text_ai/api_client.py

@@ -18,6 +18,8 @@
     API_TIMEOUT,
     API_RETRY_COUNT,
     API_PROVIDER_ANTHROPIC,
+    API_PROVIDER_DEEPSEEK,
+    API_PROVIDER_OPENAI,
     MIN_TEMPERATURE,
     MAX_TEMPERATURE,
     MIN_MAX_TOKENS,
@@ -109,19 +111,49 @@ async def create(
                 return await self._create_anthropic_completion(
                     model, messages, temperature, max_tokens
                 )
+            elif self.api_provider == API_PROVIDER_DEEPSEEK:
+                return await self._create_deepseek_completion(
+                    model, messages, temperature, max_tokens
+                )
             else:
                 return await self._create_openai_completion(
                     model, messages, temperature, max_tokens
                 )
-        except (KeyError, IndexError) as e:
-            if "'choices'" in str(e) or "'message'" in str(e):
-                raise HomeAssistantError("Failed to get a response from the AI model. Please check your internet connection and try again later.")
-            else:
-                raise
         except Exception as e:
             _LOGGER.error("API request failed: %s", str(e))
             raise HomeAssistantError(f"API request failed: {str(e)}")
 
+    async def _create_deepseek_completion(
+        self,
+        model: str,
+        messages: List[Dict[str, str]],
+        temperature: float,
+        max_tokens: int,
+    ) -> Dict[str, Any]:
+        """Create completion using DeepSeek API."""
+        url = f"{self.endpoint}/chat/completions"
+        payload = {
+            "model": model,
+            "messages": messages,
+            "temperature": temperature,
+            "max_tokens": max_tokens,
+            "stream": False
+        }
+
+        data = await self._make_request(url, payload)
+        return {
+            "choices": [
+                {
+                    "message": {"content": data["choices"][0]["message"]["content"]},
+                }
+            ],
+            "usage": {
+                "prompt_tokens": data["usage"]["prompt_tokens"],
+                "completion_tokens": data["usage"]["completion_tokens"],
+                "total_tokens": data["usage"]["total_tokens"],
+            },
+        }
+
     async def _create_openai_completion(
         self,
         model: str,

custom_components/ha_text_ai/config_flow.py

@@ -28,13 +28,16 @@
     CONF_CONTEXT_MESSAGES,
     API_PROVIDER_OPENAI,
     API_PROVIDER_ANTHROPIC,
+    API_PROVIDER_DEEPSEEK,
     API_PROVIDERS,
     DEFAULT_MODEL,
+    DEFAULT_DEEPSEEK_MODEL,
     DEFAULT_TEMPERATURE,
     DEFAULT_MAX_TOKENS,
     DEFAULT_REQUEST_INTERVAL,
     DEFAULT_OPENAI_ENDPOINT,
     DEFAULT_ANTHROPIC_ENDPOINT,
+    DEFAULT_DEEPSEEK_ENDPOINT,
     DEFAULT_CONTEXT_MESSAGES,
     MIN_TEMPERATURE,
     MAX_TEMPERATURE,
@@ -90,17 +93,22 @@ async def async_step_provider(self, user_input: Optional[Dict[str, Any]] = None)
         self._errors = {}
 
         if user_input is None:
-            default_endpoint = (
-                DEFAULT_OPENAI_ENDPOINT if self._provider == API_PROVIDER_OPENAI
-                else DEFAULT_ANTHROPIC_ENDPOINT
-            )
+            # Выбор endpoint по провайдеру
+            default_endpoint = {
+                API_PROVIDER_OPENAI: DEFAULT_OPENAI_ENDPOINT,
+                API_PROVIDER_ANTHROPIC: DEFAULT_ANTHROPIC_ENDPOINT,
+                API_PROVIDER_DEEPSEEK: DEFAULT_DEEPSEEK_ENDPOINT,
+            }.get(self._provider, DEFAULT_OPENAI_ENDPOINT)
+
+            # Выбор модели по умолчанию по провайдеру
+            default_model = DEFAULT_DEEPSEEK_MODEL if self._provider == API_PROVIDER_DEEPSEEK else DEFAULT_MODEL
 
             return self.async_show_form(
                 step_id="provider",
                 data_schema=vol.Schema({
                     vol.Required(CONF_NAME, default="my_assistant"): str,
                     vol.Required(CONF_API_KEY): str,
-                    vol.Required(CONF_MODEL, default=DEFAULT_MODEL): str,
+                    vol.Required(CONF_MODEL, default=default_model): str,
                     vol.Required(CONF_API_ENDPOINT, default=default_endpoint): str,
                     vol.Optional(CONF_TEMPERATURE, default=DEFAULT_TEMPERATURE): vol.All(
                         vol.Coerce(float),
@@ -156,15 +164,13 @@ async def async_step_provider(self, user_input: Optional[Dict[str, Any]] = None)
             if not await self._async_validate_api(input_copy):
                 return self.async_show_form(
                     step_id="provider",
-                    data_schema=vol.Schema({
-                    }),
+                    data_schema=vol.Schema({}),
                     errors=self._errors
                 )
         except Exception as e:
             return self.async_show_form(
                 step_id="provider",
-                data_schema=vol.Schema({
-                }),
+                data_schema=vol.Schema({}),
                 errors={"base": str(e)}
             )
 
@@ -250,14 +256,16 @@ async def _create_entry(self, user_input: Dict[str, Any]) -> FlowResult:
 
         unique_id = f"{DOMAIN}_{normalized_name}_{self._provider}".lower()
 
+        default_model = DEFAULT_DEEPSEEK_MODEL if self._provider == API_PROVIDER_DEEPSEEK else DEFAULT_MODEL
+
         entry_data = {
             CONF_API_PROVIDER: self._provider,
             CONF_NAME: instance_name,
             "normalized_name": normalized_name,
             CONF_API_KEY: user_input.get(CONF_API_KEY),
             CONF_API_ENDPOINT: user_input.get(CONF_API_ENDPOINT),
             "unique_id": unique_id,
-            CONF_MODEL: user_input.get(CONF_MODEL, DEFAULT_MODEL),
+            CONF_MODEL: user_input.get(CONF_MODEL, default_model),
             CONF_TEMPERATURE: user_input.get(CONF_TEMPERATURE, DEFAULT_TEMPERATURE),
             CONF_MAX_TOKENS: user_input.get(CONF_MAX_TOKENS, DEFAULT_MAX_TOKENS),
             CONF_REQUEST_INTERVAL: user_input.get(CONF_REQUEST_INTERVAL, DEFAULT_REQUEST_INTERVAL),
@@ -302,7 +310,7 @@ async def async_step_init(self, user_input: Optional[Dict[str, Any]] = None) ->
             data_schema=vol.Schema({
                 vol.Optional(
                     CONF_MODEL,
-                    default=current_data.get(CONF_MODEL, DEFAULT_MODEL)
+                    default=current_data.get(CONF_MODEL, default_model)
                 ): str,
                 vol.Optional(
                     CONF_TEMPERATURE,

custom_components/ha_text_ai/const.py

@@ -7,7 +7,7 @@
 @source: https://github.com/smkrv/ha-text-ai
 """
 import os
-import json 
+import json
 from typing import Final
 import voluptuous as vol
 from homeassistant.const import Platform, CONF_API_KEY, CONF_NAME
@@ -21,10 +21,12 @@
 CONF_API_PROVIDER: Final = "api_provider"
 API_PROVIDER_OPENAI: Final = "openai"
 API_PROVIDER_ANTHROPIC: Final = "anthropic"
+API_PROVIDER_DEEPSEEK: Final = "deepseek"
 
 API_PROVIDERS: Final = [
     API_PROVIDER_OPENAI,
-    API_PROVIDER_ANTHROPIC
+    API_PROVIDER_ANTHROPIC,
+    API_PROVIDER_DEEPSEEK
 ]
 
 # Read version from manifest.json
@@ -46,6 +48,7 @@
 # Default endpoints
 DEFAULT_OPENAI_ENDPOINT: Final = "https://api.openai.com/v1"
 DEFAULT_ANTHROPIC_ENDPOINT: Final = "https://api.anthropic.com"
+DEFAULT_DEEPSEEK_ENDPOINT: Final = "https://api.deepseek.com"
 
 # Configuration constants
 CONF_MODEL: Final = "model"
@@ -65,6 +68,7 @@
 
 # Default values
 DEFAULT_MODEL: Final = "gpt-4o-mini"
+DEFAULT_DEEPSEEK_MODEL: Final = "deepseek-chat"
 DEFAULT_TEMPERATURE: Final = 0.1
 DEFAULT_MAX_TOKENS: Final = 1000
 DEFAULT_REQUEST_INTERVAL: Final = 1.0
@@ -80,7 +84,7 @@
 MIN_TEMPERATURE: Final = 0.0
 MAX_TEMPERATURE: Final = 2.0
 MIN_MAX_TOKENS: Final = 1
-MAX_MAX_TOKENS: Final = 4096
+MAX_MAX_TOKENS: Final = 100000
 MIN_REQUEST_INTERVAL: Final = 0.1
 MAX_REQUEST_INTERVAL: Final = 60.0
 

custom_components/ha_text_ai/coordinator.py

@@ -796,46 +796,6 @@ def _get_safe_initial_state(self) -> Dict[str, Any]:
             "normalized_name": self.normalized_name,
         }
 
-    def _calculate_context_tokens(self, messages: List[Dict[str, str]], model: Optional[str] = None) -> int:
-        total_tokens = 0
-
-        # Compile regular expressions for performance
-        number_pattern = re.compile(r'[0-9]')
-        special_char_pattern = re.compile(r'[^\w\s]')
-        whitespace_pattern = re.compile(r'\s+')
-
-        try:
-            for message in messages:
-                text = message.get('content', '')
-                if text:
-                    # Normalize whitespace
-                    text = whitespace_pattern.sub(' ', text.strip())
-
-                    # Advanced token estimation heuristics
-                    words = text.split()
-                    for word in words:
-                        # Complexity-based token calculation
-                        if len(word) > 8:  # Long words
-                            total_tokens += 2
-                        elif number_pattern.search(word):  # Words with numbers
-                            total_tokens += 1.5
-                        elif special_char_pattern.search(word):  # Words with special characters
-                            total_tokens += 1.5
-                        elif word.isupper():  # Acronyms and technical terms
-                            total_tokens += 1.5
-                        else:  # Regular words
-                            total_tokens += 1
-
-                    # Additional correction for technical texts
-                    total_tokens = math.ceil(total_tokens * 1.2)
-
-            return int(total_tokens)
-
-        except Exception as e:
-            _LOGGER.error(f"Token calculation error: {e}. Processed {len(messages)} messages.")
-            # Safe fallback with minimal token estimation
-            return len(messages) * 100
-
     async def async_ask_question(
         self,
         question: str,
@@ -876,9 +836,7 @@ async def async_process_question(
             system_prompt: Optional[str] = None,
             context_messages: Optional[int] = None,
         ) -> dict:
-            """
-            Enhanced question processing with intelligent token management.
-            """
+            """Process question with context management."""
             if self.client is None:
                 raise HomeAssistantError("AI client not initialized")
 
@@ -900,62 +858,23 @@ async def async_process_question(
                 if temp_system_prompt:
                     messages.append({"role": "system", "content": temp_system_prompt})
 
-                # Context history management
+                # Add context from history
                 context_history = self._conversation_history[-temp_context_messages:]
-
-                # Comprehensive token calculation
-                context_tokens = self._calculate_context_tokens(
-                    [{"content": entry["question"]} for entry in context_history] +
-                    [{"content": entry["response"]} for entry in context_history] +
-                    [{"content": question}],
-                    temp_model
-                )
-
-                # Dynamic token allocation
-                available_tokens = max(0, temp_max_tokens - context_tokens)
-
-                # Context trimming if over token limit
-                if context_tokens > temp_max_tokens:
-                    _LOGGER.warning(
-                        f"Token limit exceeded. "
-                        f"Context: {context_tokens}, "
-                        f"Max: {temp_max_tokens}"
-                    )
-
-                    # Intelligent context reduction
-                    while context_tokens > temp_max_tokens // 2 and context_history:
-                        context_history.pop(0)
-                        context_tokens = self._calculate_context_tokens(
-                            [{"content": entry["question"]} for entry in context_history] +
-                            [{"content": entry["response"]} for entry in context_history] +
-                            [{"content": question}],
-                            temp_model
-                        )
-
-                # Rebuild messages with trimmed context
                 for entry in context_history:
                     messages.append({"role": "user", "content": entry["question"]})
                     messages.append({"role": "assistant", "content": entry["response"]})
 
+                # Add current question
                 messages.append({"role": "user", "content": question})
 
-                # Detailed token logging
-                _LOGGER.debug(
-                    f"Token Analysis: "
-                    f"Context Tokens: {context_tokens}, "
-                    f"Max Tokens: {temp_max_tokens}, "
-                    f"Available Tokens: {available_tokens}"
-                )
-
-                # Prepare API call with dynamic token management
+                # Process message
                 kwargs = {
                     "model": temp_model,
                     "temperature": temp_temperature,
-                    "max_tokens": min(temp_max_tokens, available_tokens),
+                    "max_tokens": temp_max_tokens,
                     "messages": messages,
                 }
 
-                # Process message
                 response = await self.async_process_message(question, **kwargs)
 
                 # Update metrics

custom_components/ha_text_ai/manifest.json

@@ -23,6 +23,6 @@
   "single_config_entry": false,
   "ssdp": [],
   "usb": [],
-  "version": "2.1.0",
+  "version": "2.1.1",
   "zeroconf": []
 }

custom_components/ha_text_ai/sensor.py

@@ -155,7 +155,7 @@ def __init__(
             name=self._attr_name,
             manufacturer="Community",
             model=f"{model} ({api_provider} provider)",
-            sw_version=VERSION, 
+            sw_version=VERSION,
         )
 
         _LOGGER.debug(