add Custom provider

.gitignore

@@ -1,3 +1,5 @@
-.env
+.env
-__pycache__/
+__pycache__/
-outputs/*
+outputs/*
+venv/
+

README.md

@@ -1,59 +1,59 @@
-# FWAL WebUI (Faster Whisper And LLM WebUI) by swrneko
+# FWAL WebUI (Faster Whisper And LLM WebUI) by swrneko
-
+
-<div align="center">
+<div align="center">
-  <img src="https://count.getloli.com/get/@swrneko-faster-whisper-llm?theme=rule34"/>
+  <img src="https://count.getloli.com/get/@swrneko-faster-whisper-llm?theme=rule34"/>
-</div>
+</div>
-
+
-## Screenshots
+## Screenshots
-<div align="center">
+<div align="center">
-  <div>
+  <div>
-    <img src="src/1.png" style="object-fit: cover;"/>
+    <img src="src/1.png" style="object-fit: cover;"/>
-  </div>
+  </div>
-  <div style="align-items: center;">
+  <div style="align-items: center;">
-    <img src="src/2.png" style="width: 49.7%; object-fit: cover;"/>
+    <img src="src/2.png" style="width: 49.7%; object-fit: cover;"/>
-    <img src="src/3.png" style="width: 49.7%; object-fit: cover;"/>
+    <img src="src/3.png" style="width: 49.7%; object-fit: cover;"/>
-  </div>
+  </div>
-</div>
+</div>
-
+
-
+
-## Requirements
+## Requirements
-  - python-conda or miniconda;
+  - python-conda or miniconda;
-  - python 3.10 or above;
+  - python 3.10 or above;
-  - linux (windows not tested but probably working);
+  - linux (windows not tested but probably working);
-
+
-  Python requirements are listed in `requirements.txt`.
+  Python requirements are listed in `requirements.txt`.
-
+
-## Installation
+## Installation
-1. Clone repository:
+1. Clone repository:
-  ```
+  ```
-  git clone https://github.com/swrneko/faster-whisper-n-ionet-llm.git
+  git clone https://github.com/swrneko/faster-whisper-n-ionet-llm.git
-  cd faster-whisper-n-ionet-llm
+  cd faster-whisper-n-ionet-llm
-  ```
+  ```
-  also (if not insatlled)
+  also (if not insatlled)
-  - Insatll conda:
+  - Insatll conda:
-  ```
+  ```
-  wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O miniconda.sh && bash miniconda.sh 
+  wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O miniconda.sh && bash miniconda.sh 
-  ```
+  ```
-
+
-2. Create virtual env:
+2. Create virtual env:
-  ```
+  ```
-  conda create -n faster-whisper-n-ionet-llm python=3.10
+  conda create -n faster-whisper-n-ionet-llm python=3.10
-  conda activate faster-whisper-n-ionet-llm
+  conda activate faster-whisper-n-ionet-llm
-  conda install nvidia::cudnn cuda-version=12
+  conda install nvidia::cudnn cuda-version=12
-  ```
+  ```
-
+
-3. Install requirements:
+3. Install requirements:
-  ```
+  ```
-  pip install -r requirements.txt
+  pip install -r requirements.txt
-  ```
+  ```
-
+
-4. Get api key from [io.net](https://ai.io.net/ai/api-keys) and insert into `.env` file (need to create it in root of repository directory).
+4. Get api key from [io.net](https://ai.io.net/ai/api-keys) and insert into `.env` file (need to create it in root of repository directory).
-It should looks like this:
+It should looks like this:
-  ```
+  ```
-  API_KEY='your_api_key_without_qoutes'  
+  API_KEY='your_api_key_without_qoutes'  
-  ```
+  ```
-
+
-5. Done! Now you can just run it like that:
+5. Done! Now you can just run it like that:
-  ```shell
+  ```shell
-  python app.py
+  python app.py
-  ```
+  ```

app.py

@@ -1,195 +1,128 @@
-import gradio as gr
+import gradio as gr
-from pathlib import Path
+from config import *
-
+from services.llm_factory import get_llm_provider
-# Подгрузка сервисов
+from services.fasterWhisper import FasterWhisper
-from services.llm import Llm
+from handlers.gradioHandler import GradioHandlers
-from services.fasterWhisper import FasterWhisper
+from handlers.fileHandlers import FileHandlers
-from services.convertMdToPdf import ConvertMdToPdf
+from handlers.convertMdToPdf import ConvertMdToPdf
-
+from handlers.glueAudio import GlueAudio
-# Загрузка параметров конфигурации
+
-from config import *
+gh = GradioHandlers(get_llm_provider, ConvertMdToPdf, FileHandlers, FasterWhisper, GlueAudio)
-
+
-# Функция транскрибации
+def main():
-def generateByCondition(api_key, llm_model, system_prompt, recognized_text, llm_temperature, is_pipeline_enabled, trigger, isSaveFile, filename, filenamePdf):
+    with gr.Blocks() as demo:
-    llm = Llm(api_key)
+        gr.HTML('<div align=center><h1>Faster Whisper WebUI</h1></div>')
-
+
-    # если чекбокс включен и событие было change → обрабатываем
+        with gr.Row():
-    if is_pipeline_enabled and trigger == "change":
+            with gr.Tab('Actions'):
-        result, md = llm.generate(llm_model, system_prompt, recognized_text, llm_temperature)
+                isPipelineEnabledCheckbox = gr.Checkbox(label='is pipeline enabled', value=True, interactive=True)
-
+
-        # Конвертируем текст с латексом в юникод
+                with gr.Row():
-        pdf, unicodeText = ConvertMdToPdf().convertLatexToText(md)
+                    with gr.Accordion(label='Recognization and integration'):
-
+                        with gr.Column():
-        if isSaveFile:
+                            audioFiles = gr.Files(label='Load audio for transcribe', type="filepath")
-            savePdf(filenamePdf, pdf)
+                            images = gr.Files(label='Upload images', file_types=['image'])
-            saveFile(filename, result)
+                            recognizeBtn = gr.Button('recognize and integrate', variant='primary')
-
+                            with gr.Accordion(label='Recognized text'):
-        return result, unicodeText
+                                recognizedText = gr.TextArea(label='')
-
+
-    # если чекбокс выключен и событие было click → обрабатываем
+                    with gr.Accordion(label='LLM'):
-    if not is_pipeline_enabled and trigger == "click":
+                        with gr.Column():
-        result, md = llm.generate(llm_model, system_prompt, recognized_text, llm_temperature)
+                            refineTextBtn = gr.Button('refine text', variant='secondary', interactive=False)
-
+                            with gr.Accordion(label='Refined text raw'):
-        # Конвертируем текст с латексом в юникод
+                                refinedText = gr.Textbox(label='', show_copy_button=True)
-        pdf, unicodeText = ConvertMdToPdf().convertLatexToText(md)
+                            with gr.Accordion(label='Refined text md formated'):
-
+                                refinedTextMD = gr.Markdown(label='')
-        if isSaveFile:
+
-            savePdf(filenamePdf, pdf)
+            with gr.Tab('Settings'):
-            saveFile(filename, result)
+                with gr.Column():
-
+                    with gr.Accordion('File settings'):
-        return result, unicodeText
+                        saveFileCheckbox = gr.Checkbox(label='save file', value=True, interactive=True)
-
+                        filename = gr.Textbox(label='Output filename', value='output.md', interactive=True)
-    # если нет чекбокса и было событие change
+                        filenamePdf = gr.Textbox(label='Output filename for pdf', value='output.pdf', interactive=True)
-    return gr.skip(), gr.skip()
+
-
+                    with gr.Accordion(label='Faster whisper settings'):
-
+                        with gr.Row():
-def savePdf(filename, pdf):
+                            with gr.Column():
-    directory = Path(OUTPUT_PATH)
+                                device = gr.Dropdown(label='Device', choices=DEVICES, value=DEVICES[1], interactive=True)
-    filePath = directory / filename
+                                compute_type = gr.Dropdown(label='compute_type', choices=COMPUTE_TYPE, value=COMPUTE_TYPE[0], interactive=True)
-    filePath.parent.mkdir(parents=True, exist_ok=True)
+                                fastWhisperModel = gr.Dropdown(label='Model', choices=FAST_WHISPER_MODELS, value=FAST_WHISPER_MODELS[11], interactive=True)
-    pdf.save(filePath)
+                                beamSize = gr.Number(label='beam_size', value=8, interactive=True)
-
+                                noSpeechThreshold  = gr.Number(label='no_speech_threshold', value=0.5, interactive=True)
-# Функция сохранеhния файла
+                                vadFilter = gr.Checkbox(label='vad_filter', value=True, interactive=True)
-def saveFile(filename, text):
+                                wordTimestamps = gr.Checkbox(label='word_timestamps', value=True, interactive=True)
-    directory = Path(OUTPUT_PATH)
+                                conditionOnPreviousText = gr.Checkbox(label='condition_on_previous_text', value=False, interactive=True)
-    filePath = directory / filename
+                            with gr.Column():
-    filePath.parent.mkdir(parents=True, exist_ok=True)
+                                with gr.Accordion(label='Vad parameters'):
-    filePath.write_text(text, encoding='utf-8')
+                                    minSilenceDurationMs = gr.Number(label='min_silence_duration_ms', value=300, interactive=True)
-
+                                    speechPadMs = gr.Number(label='speech_pad_ms', value=200, interactive=True)
-#    ConvertMdToPdf().convert(text)
+                                with gr.Accordion(label='Temperature'):
-
+                                    temp0 = gr.Number(label='temp_0', value=0.0, interactive=True)
-# Функция для динамического обновления кнопки в зависимости от состояния checkbox
+                                    temp1 = gr.Number(label='temp_1', value=0.2, interactive=True)
-def updateButton(isChecked):
+                                    temp2 = gr.Number(label='temp_2', value=0.4, interactive=True)
-    if not isChecked:
+
-        variant = 'primary'
+                    with gr.Accordion(label='LLM settings'):
-    else:
+                        apiKey = gr.Textbox(label='API key (required for io.net, Gemini)', value=DEFAULT_API_KEY, interactive=True)
-        variant = 'secondary'
+                        with gr.Accordion(label='System prompt'):
-
+                            systemPrompt = gr.Textbox(label='', value=DEFAULT_SYSTEM_PROMPT, interactive=True)
-    return gr.update(interactive=not isChecked, variant=variant)
+
-
+                        with gr.Row():
-
+                            llmProvider = gr.Dropdown(label='LLM Provider', choices=LLM_PROVIDERS, value=LLM_PROVIDERS[0], interactive=True)
-def updateTextbox(isChecked):
+                            llmModel = gr.Dropdown(label='Models', choices=LLM_MODELS[LLM_PROVIDERS[0]], value=LLM_MODELS[LLM_PROVIDERS[0]][1], interactive=True)
-    return gr.update(visible=isChecked)
+                            llmTemperature = gr.Number(label='Temperature', value=0.8, interactive=True)
-
+                        
-###################################################
+                        # Настройки Custom провайдера
-# ____ ___.___  ___.          .__                 #
+                        with gr.Accordion(label='Custom Provider Settings', open=True):
-#|    |   \   | \_ |__   ____ |  |   ______  _  __#
+                            customBaseUrl = gr.Textbox(
-#|    |   /   |  | __ \_/ __ \|  |  /  _ \ \/ \/ /#
+                                label='Base URL', 
-#|    |  /|   |  | \_\ \  ___/|  |_(  <_> )     / #
+                                value='http://127.0.0.1:1234/v1/', 
-#|______/ |___|  |___  /\___  >____/\____/ \/\_/  #
+                                interactive=True,
-#                    \/     \/                    #
+                                visible=False # Скрыто по умолчанию
-###################################################
+                            )
-
+
-with gr.Blocks() as demo:
+        # Обработчики событий
-    gr.HTML('''
+        isPipelineEnabledCheckbox.change(gh.updateButton, inputs=[isPipelineEnabledCheckbox], outputs=refineTextBtn)
-    <div align=center>
+        saveFileCheckbox.change(gh.updateTextbox, inputs=saveFileCheckbox, outputs=filename)
-        <h1>
+        saveFileCheckbox.change(gh.updateTextbox, inputs=saveFileCheckbox, outputs=filenamePdf)
-            Faster Whisper WebUI
+
-        </h1>
+        recognizeBtn.click(
-    </div>
+            gh.handleRecognizeBtn, 
-    ''')
+            inputs=[audioFiles, fastWhisperModel, device, compute_type, beamSize, 
-
+                    vadFilter, minSilenceDurationMs, speechPadMs, temp0, temp1, temp2, 
-    with gr.Row():
+                    wordTimestamps, noSpeechThreshold, conditionOnPreviousText, gr.State(GLUED_AUDIO_FILENAME), gr.State(OUTPUT_PATH)],
-        # Вкладка с основным взаимодействием
+            outputs=[recognizedText], 
-        with gr.Tab('Actions'):
+        )
-            isPipelineEnabledCheckbox = gr.Checkbox(label='is pipeline enabled', value=True, interactive=True)
+        
-
+        # --- ИСПРАВЛЕНИЕ: ДОБАВЛЕН customBaseUrl В INPUTS ---
-            with gr.Row():
+        recognizedText.change(
-                with gr.Accordion(label='Recognization and integration'):
+            gh.generateByCondition,
-                    with gr.Column():
+            inputs=[apiKey, llmProvider, llmModel, systemPrompt, recognizedText, llmTemperature, 
-                        audioFile = gr.Audio(label='Load audio for transcribe', type="filepath")
+                    isPipelineEnabledCheckbox, gr.State("change"), saveFileCheckbox, filename, filenamePdf, gr.State(OUTPUT_PATH), customBaseUrl],
-                        images = gr.Files(label='Upload images', file_types=['image'])
+            outputs=[refinedText, refinedTextMD]
-                        recognizeBtn = gr.Button('recognize and integrate', variant='primary')
+        )
-
+
-                        with gr.Accordion(label='Recognized text'):
+        # Обновление выпадающего списка моделей и поля API ключа
-                            recognizedText = gr.TextArea(label='')
+        llmProvider.change(
-
+            gh.update_model_dropdown, 
-                with gr.Accordion(label='LLM'):
+            inputs=llmProvider, 
-                    with gr.Column():
+            outputs=[llmModel, apiKey]
-                        refineTextBtn = gr.Button('refine text', variant='secondary', interactive=False)
+        )
-
+        
-                        with gr.Accordion(label='Refined text raw'):
+        # Переключение видимости URL для Custom провайдера
-                            refinedText = gr.Textbox(label='', show_copy_button=True)
+        llmProvider.change(
-
+            fn=gh.toggle_custom_url,
-                        with gr.Accordion(label='Refined text md formated'):
+            inputs=llmProvider,
-                            refinedTextMD = gr.Markdown(label='')
+            outputs=[customBaseUrl]
-
+        )
-        # Вкладка с настройками
+
-        with gr.Tab('Settings'):
+        refineTextBtn.click(
-            with gr.Column():
+            gh.generateByCondition,
-                # Первое поле на всю ширину в акордионе настроек
+            inputs=[apiKey, llmProvider, llmModel, systemPrompt, recognizedText, llmTemperature, 
-                with gr.Accordion('File settings'):
+                    isPipelineEnabledCheckbox, gr.State("click"), saveFileCheckbox, filename, filenamePdf, gr.State(OUTPUT_PATH), customBaseUrl],
-                    saveFileCheckbox = gr.Checkbox(label='save file', value=True, interactive=True)
+            outputs=[refinedText, refinedTextMD]
-                    filename = gr.Textbox(label='Output filename', value='output.txt', interactive=True)
+        )
-                    filenamePdf = gr.Textbox(label='Output filename for pdf', value='output.pdf', interactive=True)
+
-
+    demo.launch()
-                # Акордион настроек faster whisper
+
-                with gr.Accordion(label='Faster whisper settings'):
+if __name__ == '__main__':
-                    with gr.Row():
+    main()
-                        # Левая колонка в акордионе
-                        with gr.Column():
-                            device = gr.Dropdown(label='Device', choices=["cpu", "cuda"], value="cuda", interactive=True)
-                            compute_type = gr.Dropdown(label='compute_type', choices=["auto", "int8", "float16", "float32"], value="auto", interactive=True)
-                            fastWhisperModel = gr.Dropdown(label='Model', choices=FAST_WHISPER_MODELS, value=FAST_WHISPER_MODELS[11], interactive=True)
-
-                            beamSize = gr.Number(label='beam_size', value=8, interactive=True)
-                            noSpeechThreshold  = gr.Number(label='no_speech_threshold', value=0.5, interactive=True)
-                            vadFilter = gr.Checkbox(label='vad_filter', value=True, interactive=True)
-                            wordTimestamps = gr.Checkbox(label='word_timestamps', value=True, interactive=True)
-                            conditionOnPreviousText = gr.Checkbox(label='condition_on_previous_text', value=False, interactive=True)
-
-                            
-
-                        # Правая колонка в акордионе
-                        with gr.Column():
-                            with gr.Accordion(label='Vad parameters'):
-                                minSilenceDurationMs = gr.Number(label='min_silence_duration_ms', value=300, interactive=True)
-                                speechPadMs = gr.Number(label='speech_pad_ms', value=200, interactive=True)
-
-                            with gr.Accordion(label='Temperature'):
-                                temp0 = gr.Number(label='temp_0', value=0.0, interactive=True)
-                                temp1 = gr.Number(label='temp_1', value=0.2, interactive=True)
-                                temp2 = gr.Number(label='temp_2', value=0.4, interactive=True)
-
-                # Нижний акордион настроек для api ключа llm
-                with gr.Accordion(label='ai.io.net api settings'):
-                    apiKey = gr.Textbox(label='API key', value=DEFAULT_API_KEY, interactive=True)
-
-                    with gr.Accordion(label='System prompt'):
-                        systemPrompt = gr.Textbox(label='', value=DEFAULT_SYSTEM_PROMPT, interactive=True)
-
-                    with gr.Row():
-                        llmModel = gr.Dropdown(label='models', choices=LLM_MODELS, value=LLM_MODELS[1], interactive=True)
-                        llmTemperature = gr.Number(label='Temperature', value=0.8, interactive=True )
-
-    ######################################################################
-    #.____                 .__         ___.          .__                 #
-    #|    |    ____   ____ |__| ____   \_ |__   ____ |  |   ______  _  __#
-    #|    |   /  _ \ / ___\|  |/ ___\   | __ \_/ __ \|  |  /  _ \ \/ \/ /#
-    #|    |__(  <_> ) /_/  >  \  \___   | \_\ \  ___/|  |_(  <_> )     / #
-    #|_______ \____/\___  /|__|\___  >  |___  /\___  >____/\____/ \/\_/  #
-    #        \/    /_____/         \/       \/     \/                    #
-    ######################################################################
-
-    isPipelineEnabledCheckbox.change(updateButton, inputs=[isPipelineEnabledCheckbox], outputs=refineTextBtn)
-    saveFileCheckbox.change(updateTextbox, inputs=saveFileCheckbox, outputs=filename)
-    saveFileCheckbox.change(updateTextbox, inputs=saveFileCheckbox, outputs=filenamePdf)
-
-    recognizeBtn.click(FasterWhisper().recognize, outputs=[recognizedText], inputs=[fastWhisperModel, device, compute_type, audioFile, beamSize, vadFilter, minSilenceDurationMs, speechPadMs, temp0, temp1, temp2, wordTimestamps, noSpeechThreshold, conditionOnPreviousText])
-    
-    # Если пайплайн включен то тогда делаем автоматически
-    # автоматический пайплайн
-    recognizedText.change(
-        generateByCondition,
-        inputs=[apiKey, llmModel, systemPrompt, recognizedText, llmTemperature, isPipelineEnabledCheckbox, gr.State("change"), saveFileCheckbox, filename, filenamePdf],
-        outputs=[refinedText, refinedTextMD]
-    )
-
-    # ручной запуск по кнопке
-    refineTextBtn.click(
-        generateByCondition,
-        inputs=[apiKey, llmModel, systemPrompt, recognizedText, llmTemperature, isPipelineEnabledCheckbox, gr.State("click"), saveFileCheckbox, filename, filenamePdf],
-        outputs=[refinedText, refinedTextMD]
-    )
-
-demo.launch()

config.py

@@ -1,48 +1,93 @@
-import os
+import os
-from dotenv import load_dotenv
+from dotenv import load_dotenv
-
+
-load_dotenv()
+load_dotenv()
-
+
-FAST_WHISPER_MODELS = ['tiny', 'base', 'small', 'medium', 'large-v1', 'large-v2', 'large-v3', 'large', 'distil-large-v2', 'distil-large-v3', 'distil-large-v3.5', 'large-v3-turbo', 'turbo']
+FAST_WHISPER_MODELS = ['tiny', 'base', 'small', 'medium', 'large-v1', 'large-v2', 'large-v3', 'large', 'distil-large-v2', 'distil-large-v3', 'distil-large-v3.5', 'large-v3-turbo', 'turbo']
-LLM_MODELS = ['openai/gpt-oss-120b', 'Qwen/Qwen3-235B-A22B-Thinking-2507', 'deepseek-ai/DeepSeek-R1-0528', 'meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8', 'openai/gpt-oss-20b', 'Intel/Qwen3-Coder-480B-A35B-Instruct-int4-mixed-ar', 'meta-llama/Llama-3.2-90B-Vision-Instruct', 'mistralai/Mistral-Nemo-Instruct-2407', 'Qwen/Qwen2.5-VL-32B-Instruct', 'meta-llama/Llama-3.3-70B-Instruct', 'mistralai/Devstral-Small-2505', 'mistralai/Magistral-Small-2506', 'mistralai/Mistral-Large-Instruct-2411', 'CohereForAI/aya-expanse-32b']
+DEVICES = ['cpu', 'cuda']
-
+COMPUTE_TYPE = ['auto', 'int8', 'float16', 'float32']
-
+
-# Стандартный API ключ
+# Стандартный API ключ
-DEFAULT_API_KEY=os.getenv('API_KEY')
+IO_API_KEY=os.getenv('IO_API_KEY')
-# Задаем выходную директорию
+GEMINI_API_KEY=os.getenv('GEMINI_API_KEY')
-OUTPUT_PATH='outputs'
+DEFAULT_API_KEY=IO_API_KEY
-
+
-DEFAULT_SYSTEM_PROMPT='''You are a diligent university student who has recorded a lecture as an audio file and later transcribed it into raw text. 
+
-Your task is to rewrite this unstructured transcript into a clear, logically organized, and detailed lecture summary (lecture notes).  
+# Словарь провайдеров и их моделей
-
+LLM_PROVIDERS = ['io.net', 'Gemini', 'gpt4free', 'Custom']
-Guidelines:
+LLM_MODELS = {
-1. Structure:
+    'io.net': [
-   - Organize the text into a hierarchy of sections and subsections.  
+        'openai/gpt-oss-120b', 'Qwen/Qwen3-235B-A22B-Thinking-2507', 
-   - Use headings, bullet points, or numbering where appropriate.  
+        'deepseek-ai/DeepSeek-R1-0528', 'meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8', 
-   - Present the material in a logical flow (from introduction → main points → details → examples → conclusion).  
+        'openai/gpt-oss-20b', 'Intel/Qwen3-Coder-480B-A35B-Instruct-int4-mixed-ar', 
-
+        'meta-llama/Llama-3.2-90B-Vision-Instruct', 'mistralai/Mistral-Nemo-Instruct-2407', 
-2. Clarity & Cohesion:
+        'Qwen/Qwen2.5-VL-32B-Instruct', 'meta-llama/Llama-3.3-70B-Instruct', 
-   - Remove filler words, repetitions, and irrelevant fragments.  
+        'mistralai/Devstral-Small-2505', 'mistralai/Magistral-Small-2506', 
-   - Rewrite incomplete sentences into full, grammatically correct sentences.  
+        'mistralai/Mistral-Large-Instruct-2411', 'CohereForAI/aya-expanse-32b'
-   - Ensure smooth transitions between topics, making the summary feel continuous and well-connected.  
+    ],
-
+    'Gemini': [
-3. Depth & Detail: 
+        'gemini-2.5-pro',
-   - Capture all important concepts, definitions, examples, and explanations from the lecture.  
+        'gemini-2.5-flash',
-   - Expand shorthand or fragmented thoughts into full, precise explanations.  
+        'gemini-2.5-flash-lite'
-   - Where appropriate, rephrase or clarify confusing passages for better understanding.  
+    ],
-
+    'gpt4free': [
-4. Accuracy:
+        'default',
-   - Preserve the lecturer’s original meaning, intent, and terminology.  
+        'gpt-4',
-   - Avoid adding personal opinions or new information that was not in the lecture.  
+        'sonar-reasoning',
-
+        'command-r-plus',
-5. Style:
+        'llama-3.3-70b',
-   - Write in a formal, academic tone suitable for study notes.  
+        'hermes-3-llama-3.1-405b'
-   - Aim for readability: concise sentences, but thorough coverage of concepts.  
+        'qwen-3-235b',
-   - Use emphasis (e.g., bold or italic text) only when it improves comprehension.  
+        'gpt-4o-mini',
-
+        'deepseek-r1',
-Final Output: A cohesive, detailed, and well-structured lecture summary, suitable for later studying and revision.
+        'PollinationsAI:gpt-5-nano'
-Use only russian language!
+    ],
-USE LATEX IN DOLLAR SIGN ($)!
+    'Custom': [
-EXTRA BIG LENTH OF CONSPECT!
+        'qwen/qwen3-vl-30b',
-'''
+        'qwen/qwen3-coder-30b',
-
+        'openai/gpt-oss-20b', 
+        'qwen3-vl-8b-thinking', 
+        'qwen/qwen3-vl-8b',
+    ],
+}
+
+# Задаем выходную директорию
+OUTPUT_PATH='outputs'
+
+GLUED_AUDIO_FILENAME='glued.mp3'
+
+DEFAULT_SYSTEM_PROMPT='''You are a diligent university student who has recorded a lecture as an audio file and later transcribed it into raw text. 
+Your task is to rewrite this unstructured transcript into a clear, logically organized, and detailed lecture summary (lecture notes).  
+
+Guidelines:
+1. Structure:
+   - Organize the text into a hierarchy of sections and subsections.  
+   - Use headings, bullet points, or numbering where appropriate.  
+   - Present the material in a logical flow (from introduction → main points → details → examples → conclusion).  
+
+2. Clarity & Cohesion:
+   - Remove filler words, repetitions, and irrelevant fragments.  
+   - Rewrite incomplete sentences into full, grammatically correct sentences.  
+   - Ensure smooth transitions between topics, making the summary feel continuous and well-connected.  
+
+3. Depth & Detail: 
+   - Capture all important concepts, definitions, examples, and explanations from the lecture.  
+   - Expand shorthand or fragmented thoughts into full, precise explanations.  
+   - Where appropriate, rephrase or clarify confusing passages for better understanding.  
+
+4. Accuracy:
+   - Preserve the lecturer’s original meaning, intent, and terminology.  
+   - Avoid adding personal opinions or new information that was not in the lecture.  
+
+5. Style:
+   - Write in a formal, academic tone suitable for study notes.  
+   - Aim for readability: concise sentences, but thorough coverage of concepts.  
+   - Use emphasis (e.g., bold or italic text) only when it improves comprehension.  
+
+Final Output: A cohesive, detailed, and well-structured lecture summary, suitable for later studying and revision.
+Use only russian language!
+USE LATEX IN DOLLAR SIGN ($)!
+EXTRA BIG LENTH OF CONSPECT!
+MAKE AS LONG AS POSIBLE AND AS BE GOOD!
+'''
+

handlers/convertMdToPdf.py

@@ -1,29 +1,36 @@
-import re
+import re
-from pylatexenc.latex2text import LatexNodes2Text
+from pylatexenc.latex2text import LatexNodes2Text
-from markdown_pdf import MarkdownPdf
+from markdown_pdf import MarkdownPdf
-from markdown_pdf import Section
+from markdown_pdf import Section
-
+
-class ConvertMdToPdf:
+class ConvertMdToPdf:
-    # Конвертирует md в pdf
+    # Конвертирует md в pdf
-    def convertLatexToText(self, text:str):
+    def convertLatexToText(self, text:str):
-        # Обрабатываем только математические выражения
+        '''
-        text = re.sub(
+        Функция для конвертации LaTeX в текст;
-            r'\$\$(.*?)\$\$|\$(.*?)\$', 
+
-            self.replace_math, 
+        Args:
-            text, 
+            :param text: текст содержащий LaTeX.
-            flags=re.DOTALL
+        '''
-        )
+
-        pdf = MarkdownPdf(toc_level=0, optimize=True)
+        # Обрабатываем только математические выражения
-        pdf.add_section(Section(text))
+        text = re.sub(
-        return pdf, text
+            r'\$\$(.*?)\$\$|\$(.*?)\$', 
-
+            self.replace_math, 
-    def replace_math(self, match):
+            text, 
-        math_content = match.group(1) or match.group(2)  # $$...$$ или $...$
+            flags=re.DOTALL
-        try:
+        )
-            # Преобразуем только математическое выражение
+        pdf = MarkdownPdf(toc_level=0, optimize=True)
-            converted = LatexNodes2Text().latex_to_text(math_content)
+        pdf.add_section(Section(text))
-
+        return pdf, text
-            return converted
+
-        except:
+    def replace_math(self, match):
-            return math_content  # В случае ошибки оставляем как есть
+        math_content = match.group(1) or match.group(2)  # $$...$$ или $...$
-
+        try:
+            # Преобразуем только математическое выражение
+            converted = LatexNodes2Text().latex_to_text(math_content)
+
+            return converted
+        except:
+            return math_content  # В случае ошибки оставляем как есть
+

handlers/fileHandlers.py

@@ -0,0 +1,31 @@
+from pathlib import Path
+
+# Для аннотации типов
+from markdown_pdf import MarkdownPdf 
+from pydub import AudioSegment
+
+class FileHandlers:
+    # Функция сохранения файла
+    def saveFile(self, filename, content, output_path, format='mp3'):
+        '''
+        Сохраняет текст, pdf из markdown_pdf или склеенный аудиофайл в файл с указанным названием и директорией.
+
+        Args:
+            :param filename: название файла;
+            :param content: содержание файла;
+            :param output_path: выходная диретория файла.
+        '''
+        # Создание объекта директории
+        directory = Path(output_path)
+        filePath = directory / filename # Добавление пути директории
+        filePath.parent.mkdir(parents=True, exist_ok=True) # Создание директории если не существует
+
+        # Сохранение для разных типов
+        if type(content) == MarkdownPdf:
+            return content.save(filePath)
+
+        elif type(content) == str:
+            return filePath.write_text(content, encoding='utf-8')
+
+        elif type(content) == AudioSegment:
+            return content.export(filePath, format=format)

handlers/glueAudio.py

@@ -0,0 +1,61 @@
+import subprocess
+from pathlib import Path
+
+
+class GlueAudio():
+    def glue(self, audio_files: list, output_path: str, output_filename: str) -> Path:
+        """
+        Склеивает аудиофайлы РАЗНЫХ форматов с помощью FFmpeg и filter_complex.
+        Это универсальный и эффективный по памяти метод.
+
+        Args:
+            audio_files (list): Список путей к исходным аудиофайлам.
+            output_path (str): Директория для сохранения итогового файла.
+            output_filename (str): Имя итогового склеенного файла.
+
+        Returns:
+            Path: Путь к созданному склеенному файлу.
+        """
+        output_dir = Path(output_path)
+        output_dir.mkdir(parents=True, exist_ok=True)
+        final_audio_path = output_dir / output_filename
+
+        if not audio_files:
+            raise ValueError("Список аудиофайлов для склейки пуст.")
+
+        # 1. Формируем часть команды с входными файлами (-i file1 -i file2 ...)
+        input_args = []
+        for file_path in audio_files:
+            input_args.extend(['-i', str(Path(file_path).resolve())])
+
+        # 2. Формируем строку для filter_complex
+        num_files = len(audio_files)
+        stream_specifiers = "".join([f"[{i}:a]" for i in range(num_files)])
+        filter_complex_str = f"{stream_specifiers}concat=n={num_files}:v=0:a=1[outa]"
+
+        # 3. Собираем полную команду
+        command = [
+            'ffmpeg',
+            *input_args,  # Распаковываем список входных файлов
+            '-filter_complex', filter_complex_str,
+            '-map', '[outa]',
+            '-c:a', 'libmp3lame',
+            '-q:a', '2',
+            str(final_audio_path),
+            '-y'
+        ]
+
+        try:
+            # 4. Выполняем команду
+            print(f"Выполнение команды FFmpeg: {' '.join(command)}")
+            subprocess.run(command, check=True, capture_output=True, text=True)
+            print("FFmpeg успешно завершил склейку.")
+
+        except FileNotFoundError:
+            raise FileNotFoundError("FFmpeg не найден. Убедитесь, что он установлен и доступен в системной переменной PATH.")
+        except subprocess.CalledProcessError as e:
+            print("Ошибка при выполнении FFmpeg!")
+            print("Stderr:", e.stderr)
+            raise RuntimeError(f"Ошибка FFmpeg при склейке файлов: {e.stderr}")
+        
+        return final_audio_path

handlers/gradioHandler.py

@@ -0,0 +1,87 @@
+from config import LLM_MODELS, GEMINI_API_KEY, IO_API_KEY
+import gradio as gr 
+
+class GradioHandlers:
+    def __init__(self, llm_factory, ConvertMdToPdf, FileHandlers, FasterWhisper, GlueAudio):
+        self.fh = FileHandlers()
+        self.ga = GlueAudio()
+        self.ConvertMdToPdf = ConvertMdToPdf()
+        self.FasterWhisper = FasterWhisper()
+        self.llm_factory = llm_factory 
+
+    def handleRecognizeBtn(self, audioFiles, model, device, compute_type, beamSize, vadFilter, 
+            minSilenceDurationMs, speechPadMs, temp0, temp1, temp2, 
+            wordTimestamps, noSpeechThreshold, conditionOnPreviousText, filename, outPath):
+        try:
+            glued_audio_path = self.ga.glue(
+                audio_files=[f.name for f in audioFiles], 
+                output_path=outPath,
+                output_filename=filename
+            )
+        except (FileNotFoundError, RuntimeError) as e:
+            gr.Warning(str(e))
+            return "" 
+
+        return self.FasterWhisper.recognize(model, device, compute_type, str(glued_audio_path), beamSize, vadFilter, minSilenceDurationMs, speechPadMs, temp0, temp1, temp2, wordTimestamps, noSpeechThreshold, conditionOnPreviousText)
+
+    # Добавил аргумент custom_base_url в конец
+    def generateByCondition(self, api_key, llm_provider, 
+                            llm_model, system_prompt, recognized_text, 
+                            llm_temperature, is_pipeline_enabled, trigger, 
+                            isSaveFile, filename, filenamePdf, output_path, custom_base_url):
+        
+        try:
+            if llm_provider == "Custom":
+                # Передаем base_url только для Custom
+                provider = self.llm_factory(llm_provider, api_key, base_url=custom_base_url)
+            else:
+                provider = self.llm_factory(llm_provider, api_key)
+        except ValueError as e:
+            gr.Warning(str(e))
+            return gr.skip(), gr.skip()
+
+        def process():
+            # Добавлена обработка ошибок генерации
+            try:
+                result, md = provider.generate(llm_model, system_prompt, recognized_text, llm_temperature)
+            except Exception as e:
+                raise gr.Error(f"Ошибка генерации LLM: {e}")
+
+            pdf, unicodeText = self.ConvertMdToPdf.convertLatexToText(md)
+            if isSaveFile:
+                self.fh.saveFile(filenamePdf, pdf, output_path)
+                self.fh.saveFile(filename, result, output_path)
+            return result, unicodeText
+
+        if (is_pipeline_enabled and trigger == "change") or (not is_pipeline_enabled and trigger == "click"):
+            return process()
+
+        return gr.skip(), gr.skip()
+    
+    def update_model_dropdown(self, provider):
+        models = LLM_MODELS.get(provider, [])
+        default_value = models[0] if models else None
+        
+        # Обновляем список моделей и настройки поля API Key
+        if provider == 'io.net': 
+            return gr.update(choices=models, value=default_value), gr.update(label='API key', value=IO_API_KEY, interactive=True, visible=True)
+        if provider == 'Gemini': 
+            return gr.update(choices=models, value=default_value), gr.update(label='API key', value=GEMINI_API_KEY, interactive=True, visible=True)
+        if provider == 'gpt4free': 
+            return gr.update(choices=models, value=default_value), gr.update(label='API key (not required)', value="", interactive=False, visible=True)
+        if provider == "Custom": 
+            return gr.update(choices=models, value=default_value), gr.update(label='API key (optional)', value="", interactive=True, visible=True) # Для Custom ключ может понадобиться
+
+    def updateButton(self, isChecked):
+        variant = 'secondary' if isChecked else 'primary'
+        return gr.update(interactive=not isChecked, variant=variant)
+    
+    def toggle_custom_url(self, provider):
+        """Показывает поле Base URL только если выбран Custom"""
+        return gr.update(visible=(provider == 'Custom'))
+    
+    def update_custom_url(self, base_url):
+        return None 
+
+    def updateTextbox(self, isChecked):
+        return gr.update(visible=isChecked)

handlers/metadataHandler.py

@@ -0,0 +1,21 @@
+import os
+from PIL import Image, ExifTags
+import subprocess
+import json
+from datetime import datetime
+
+class MetadataHandler:
+    def get_image_timestamp(self, image_path: str) -> datetime | None: 
+        """Извлекает метку времени из метаданных изображения, если она доступна."""
+        try:
+            image = Image.open(image_path)
+            exif_data = image._getexif()
+            if exif_data:
+                for tag, value in exif_data.items():
+                    decoded_tag = ExifTags.TAGS.get(tag, tag)
+                    if decoded_tag == 'DateTimeOriginal':
+                        return value
+            return None
+        except Exception as e:
+            print(f"Error extracting metadata from image: {e}")
+            return None

pandoc.txt

@@ -0,0 +1,7 @@
+pandoc "out.md" -o output1310.pdf \
+  --pdf-engine=xelatex \
+  -V geometry:margin=2.5cm \
+  -V fontsize=12pt \
+  -V mainfont="Times New Roman" \
+  -V colorlinks=true \
+  -V linkcolor=blue\

requirements.txt

@@ -1,102 +1,102 @@
-aiofiles==24.1.0
+aiofiles==24.1.0
-annotated-types==0.7.0
+annotated-types==0.7.0
-anyio==4.10.0
+anyio==4.10.0
-av==15.1.0
+av==15.1.0
-beautifulsoup4==4.13.5
+beautifulsoup4==4.13.5
-Brotli==1.1.0
+Brotli==1.1.0
-bs4==0.0.2
+bs4==0.0.2
-certifi==2025.8.3
+certifi==2025.8.3
-cffi==2.0.0
+cffi==2.0.0
-charset-normalizer==3.4.3
+charset-normalizer==3.4.3
-click==8.2.1
+click==8.2.1
-coloredlogs==15.0.1
+coloredlogs==15.0.1
-colour==0.1.5
+colour==0.1.5
-cssselect2==0.8.0
+cssselect2==0.8.0
-ctranslate2==4.6.0
+ctranslate2==4.6.0
-distro==1.9.0
+distro==1.9.0
-dotenv==0.9.9
+dotenv==0.9.9
-exceptiongroup==1.3.0
+exceptiongroup==1.3.0
-fastapi==0.116.1
+fastapi==0.116.1
-faster-whisper==1.2.0
+faster-whisper==1.2.0
-ffmpeg-python==0.2.0
+ffmpeg-python==0.2.0
-ffmpy==0.6.1
+ffmpy==0.6.1
-filelock==3.19.1
+filelock==3.19.1
-flatbuffers==25.2.10
+flatbuffers==25.2.10
-flatlatex==0.15
+flatlatex==0.15
-fonttools==4.59.2
+fonttools==4.59.2
-fsspec==2025.9.0
+fsspec==2025.9.0
-future==1.0.0
+future==1.0.0
-gradio==5.44.1
+gradio==5.44.1
-gradio_client==1.12.1
+gradio_client==1.12.1
-groovy==0.1.2
+groovy==0.1.2
-h11==0.16.0
+h11==0.16.0
-hf-xet==1.1.9
+hf-xet==1.1.9
-httpcore==1.0.9
+httpcore==1.0.9
-httpx==0.28.1
+httpx==0.28.1
-huggingface-hub==0.34.4
+huggingface-hub==0.34.4
-humanfriendly==10.0
+humanfriendly==10.0
-idna==3.10
+idna==3.10
-iso639-lang==2.6.3
+iso639-lang==2.6.3
-Jinja2==3.1.6
+Jinja2==3.1.6
-jiter==0.10.0
+jiter==0.10.0
-joblib==1.5.2
+joblib==1.5.2
-langdetect==1.0.9
+langdetect==1.0.9
-littleutils==0.2.4
+littleutils==0.2.4
-markdown-it-py==3.0.0
+markdown-it-py==3.0.0
-markdown_pdf==1.9
+markdown_pdf==1.9
-MarkupSafe==3.0.2
+MarkupSafe==3.0.2
-mdurl==0.1.2
+mdurl==0.1.2
-mpmath==1.3.0
+mpmath==1.3.0
-nltk==3.9.1
+nltk==3.9.1
-numpy==2.2.6
+numpy==2.2.6
-onnxruntime==1.22.1
+onnxruntime==1.22.1
-openai==1.106.1
+openai==1.106.1
-orjson==3.11.3
+orjson==3.11.3
-outdated==0.2.2
+outdated==0.2.2
-packaging==25.0
+packaging==25.0
-pandas==2.3.2
+pandas==2.3.2
-pillow==11.3.0
+pillow==11.3.0
-protobuf==6.32.0
+protobuf==6.32.0
-pycparser==2.22
+pycparser==2.22
-pydantic==2.11.7
+pydantic==2.11.7
-pydantic_core==2.33.2
+pydantic_core==2.33.2
-pydub==0.25.1
+pydub==0.25.1
-pydyf==0.11.0
+pydyf==0.11.0
-Pygments==2.19.2
+Pygments==2.19.2
-pylatexenc==2.10
+pylatexenc==2.10
-pymultidictionary==1.3.2
+pymultidictionary==1.3.2
-PyMuPDF==1.26.4
+PyMuPDF==1.26.4
-pyperclip==1.9.0
+pyperclip==1.9.0
-pyphen==0.17.2
+pyphen==0.17.2
-python-dateutil==2.9.0.post0
+python-dateutil==2.9.0.post0
-python-dotenv==1.1.1
+python-dotenv==1.1.1
-python-multipart==0.0.20
+python-multipart==0.0.20
-pytz==2025.2
+pytz==2025.2
-PyYAML==6.0.2
+PyYAML==6.0.2
-regex==2025.9.1
+regex==2025.9.1
-requests==2.32.5
+requests==2.32.5
-rich==14.1.0
+rich==14.1.0
-ruff==0.12.12
+ruff==0.12.12
-safehttpx==0.1.6
+safehttpx==0.1.6
-semantic-version==2.10.0
+semantic-version==2.10.0
-shellingham==1.5.4
+shellingham==1.5.4
-six==1.17.0
+six==1.17.0
-sniffio==1.3.1
+sniffio==1.3.1
-soupsieve==2.8
+soupsieve==2.8
-starlette==0.47.3
+starlette==0.47.3
-sympy==1.14.0
+sympy==1.14.0
-tinycss2==1.4.0
+tinycss2==1.4.0
-tinyhtml5==2.0.0
+tinyhtml5==2.0.0
-tkmacosx==1.0.5
+tkmacosx==1.0.5
-tokenizers==0.22.0
+tokenizers==0.22.0
-tomlkit==0.13.3
+tomlkit==0.13.3
-tqdm==4.67.1
+tqdm==4.67.1
-typer==0.17.4
+typer==0.17.4
-typing-inspection==0.4.1
+typing-inspection==0.4.1
-typing_extensions==4.15.0
+typing_extensions==4.15.0
-tzdata==2025.2
+tzdata==2025.2
-urllib3==2.5.0
+urllib3==2.5.0
-uvicorn==0.35.0
+uvicorn==0.35.0
-webencodings==0.5.1
+webencodings==0.5.1
-websockets==15.0.1
+websockets==15.0.1
-zopfli==0.2.3.post1
+zopfli==0.2.3.post1

services/fasterWhisper.py

@@ -1,34 +1,39 @@
-from faster_whisper import WhisperModel
+from faster_whisper import WhisperModel
-
+
-class FasterWhisper:
+class FasterWhisper:
-    def recognize(self, model, device, compute_type, audioFile, beamSize, vadFilter, minSilenceDurationMs, speechPadMs, temp0, temp1, temp2, wordTimestamps, noSpeechThreshold, conditionOnPreviousText):
+    def recognize(self, model, device, compute_type, 
-        model = WhisperModel(model, device=device, compute_type=compute_type) # Задаем модель
+                  audioFile, beamSize, vadFilter, 
-
+                  minSilenceDurationMs, speechPadMs, 
-        segments, _ = model.transcribe( # Распознаем текст
+                  temp0, temp1, temp2, wordTimestamps, 
-            audioFile,
+                  noSpeechThreshold, conditionOnPreviousText):
-            beam_size=beamSize,
+        model = WhisperModel(model, device=device, compute_type=compute_type) # Задаем модель
-            vad_filter=vadFilter,
+
-            vad_parameters={
+        segments, _ = model.transcribe( # Распознаем текст
-                "min_silence_duration_ms": minSilenceDurationMs,
+            audioFile,
-                "speech_pad_ms": speechPadMs
+            beam_size=beamSize,
-            },
+            vad_filter=vadFilter,
-            temperature= [temp0, temp1, temp2],
+            vad_parameters={
-            word_timestamps=wordTimestamps,
+                "min_silence_duration_ms": minSilenceDurationMs,
-            no_speech_threshold=noSpeechThreshold,
+                "speech_pad_ms": speechPadMs
-            condition_on_previous_text=conditionOnPreviousText
+            },
-        )
+            temperature= [temp0, temp1, temp2],
-
+            word_timestamps=wordTimestamps,
-        text = ''
+            no_speech_threshold=noSpeechThreshold,
-
+            condition_on_previous_text=conditionOnPreviousText
-        for seg in segments:
+        )
-            text += f"[{self.format_timestamp(seg.start)} -> {self.format_timestamp(seg.end)}] {seg.text}" + '\n'
+
-        
+        text = ''
-        return(text)
+
-
+        for seg in segments:
-    def format_timestamp(self, seconds: float) -> str:
+            text += f"[{self.format_timestamp(seg.start)} -> {self.format_timestamp(seg.end)}] {seg.text}" + '\n'
-        millis = int(seconds * 1000)
+            print(f"[{self.format_timestamp(seg.start)} -> {self.format_timestamp(seg.end)}] {seg.text}")
-        hours = millis // (3600 * 1000)
+        
-        minutes = (millis % (3600 * 1000)) // (60 * 1000)
+        return text
-        seconds_int = (millis % (60 * 1000)) // 1000
+
-        millis = millis % 1000
+    def format_timestamp(self, seconds: float) -> str:
-        return f"{hours:02d}:{minutes:02d}:{seconds_int:02d},{millis:03d}"
+        millis = int(seconds * 1000)
+        hours = millis // (3600 * 1000)
+        minutes = (millis % (3600 * 1000)) // (60 * 1000)
+        seconds_int = (millis % (60 * 1000)) // 1000
+        millis = millis % 1000
+        return f"{hours:02d}:{minutes:02d}:{seconds_int:02d},{millis:03d}"

services/llm.py

@@ -1,31 +0,0 @@
-import openai
-
-class Llm:
-    def __init__(self, apiKey:str):
-        self.client = openai.OpenAI(
-            api_key=apiKey,
-            base_url='https://api.intelligence.io.solutions/api/v1/'
-        )
-
-    def generate(self, model:str, systemPrompt:str, userPrompt:str, temp:float):
-        '''
-        prompt[user_promtp, system_ptompt]
-        Function generate text by prompt
-        '''
-
-        # Получаем ответ от нейросети
-        response = self.client.chat.completions.create(
-            model=model,
-            messages=[
-                {'role': 'system', 'content': systemPrompt},
-                {'role': 'user', 'content': userPrompt},
-            ],
-            temperature=temp,
-            stream=False
-        ) 
-
-        # Достаем текст
-        text = str(response.choices[0].message.content)
-
-        return text, text
-

services/llm_factory.py

@@ -0,0 +1,27 @@
+# services/llm_factory.py
+from services.llm_providers.ionet_provider import IoNetProvider
+from services.llm_providers.gemini_provider import GeminiProvider
+from services.llm_providers.gpt4free_provider import Gpt4FreeProvider
+from services.llm_providers.base_provider import BaseLLMProvider
+from services.llm_providers.custom_provider import CustomProvider
+
+def get_llm_provider(provider_name: str, api_key: str | None = None, base_url: str | None = None) -> BaseLLMProvider:
+    """
+    Фабричная функция для получения экземпляра провайдера LLM.
+    """
+    if provider_name == 'io.net':
+        if not api_key:
+            raise ValueError("API ключ обязателен для io.net")
+        return IoNetProvider(api_key)
+    elif provider_name == 'Gemini':
+        if not api_key:
+            raise ValueError("API ключ обязателен для Gemini")
+        return GeminiProvider(api_key)
+    elif provider_name == 'gpt4free':
+        return Gpt4FreeProvider()
+    elif provider_name == 'Custom':
+        if not base_url:
+            raise ValueError("Base URL обязателен для Custom провайдера")
+        return CustomProvider(api_key, base_url)  # base_url будет установлен позже
+    else:
+        raise ValueError(f"Неизвестный провайдер: {provider_name}")

services/llm_providers/base_provider.py

@@ -0,0 +1,18 @@
+from abc import ABC, abstractmethod
+
+class BaseLLMProvider(ABC):
+    """
+    Абстрактный базовый класс для всех провайдеров LLM.
+    Каждый провайдер должен реализовать метод generate.
+    """
+    def __init__(self, api_key: str | None = None):
+        self.api_key = api_key
+
+    @abstractmethod
+    def generate(self, model: str, system_prompt: str, user_prompt: str, temp: float):
+        """
+        Основной метод для генерации текста.
+
+        Должен возвращать кортеж из двух строк: (чистый_текст, markdown_текст)
+        """
+        pass

services/llm_providers/custom_provider.py

@@ -0,0 +1,44 @@
+import requests
+from .base_provider import BaseLLMProvider
+
+class CustomProvider(BaseLLMProvider):
+    def __init__(self, api_key: str | None = None, base_url: str | None = None):
+        super().__init__(api_key)
+        self.base_url = base_url or 'http://127.0.0.1:1234/v1/'
+        
+        # ГАРАНТИРУЕМ наличие слеша в конце URL
+        if not self.base_url.endswith('/'):
+            self.base_url += '/'
+
+    def generate(self, model: str, system_prompt: str, user_prompt: str, temp: float):
+        url = f"{self.base_url}chat/completions"
+        
+        # Некоторые Custom провайдеры (как vLLM или Ollama) могут требовать API Key, даже если он фиктивный
+        headers = {"Content-Type": "application/json"}
+        if self.api_key:
+             headers["Authorization"] = f"Bearer {self.api_key}"
+
+        data = {
+            "model": model,
+            "messages": [
+                {'role': 'system', 'content': system_prompt},
+                {'role': 'user', 'content': user_prompt},
+            ],
+            "temperature": temp,
+            "stream": False
+        }
+        
+        try:
+            response = requests.post(url, headers=headers, json=data, timeout=300)
+            response.raise_for_status()
+            result = response.json()
+            
+            # Обработка разных форматов ответа (на всякий случай)
+            if 'choices' in result and len(result['choices']) > 0:
+                text = str(result['choices'][0]['message']['content'])
+                return text, text
+            else:
+                return f"Неожиданный ответ от сервера: {result}", str(result)
+            
+        except requests.exceptions.RequestException as e:
+            return f"Ошибка при запросе к CustomProvider API: {e}", f"Ошибка: {e}"

services/llm_providers/gemini_provider.py

@@ -0,0 +1,74 @@
+# services/llm_providers/gemini_provider.py
+
+import requests
+from .base_provider import BaseLLMProvider
+
+class GeminiProvider(BaseLLMProvider):
+    """
+    Провайдер для Google Gemini, использующий прямые REST API вызовы 
+    через библиотеку requests для надежной работы с SOCKS-прокси.
+    """
+    def __init__(self, api_key: str):
+        super().__init__(api_key)
+        self.base_url = "https://generativelanguage.googleapis.com/v1beta/models/"
+
+    def generate(self, model: str, system_prompt: str, user_prompt: str, temp: float):
+        """
+        Генерирует текст с помощью модели Gemini, отправляя запрос через прокси.
+        """
+        # 1. Формируем URL для запроса
+        api_url = f"{self.base_url}{model}:generateContent?key={self.api_key}"
+
+        # 2. Задаем настройки прокси из вашего примера
+        # socks5h:// означает, что DNS-запросы также будут идти через прокси
+        proxies = {
+            'http': 'socks5://192.168.1.6:2080',
+            'https': 'socks5h://192.168.1.6:2080'
+        }
+
+        # 3. Собираем тело запроса (payload) в формате, который ожидает Gemini API
+        data = {
+            "system_instruction": {
+                "parts": {"text": system_prompt}
+            },
+            "contents": [{
+                "parts": [{"text": user_prompt}]
+            }],
+            "generationConfig": {
+                "temperature": temp
+            }
+        }
+        
+        try:
+            # 4. Отправляем POST-запрос с данными и настройками прокси
+            response = requests.post(api_url, json=data, proxies=proxies, timeout=400)
+            
+            # Проверяем, не вернул ли сервер ошибку (например, 4xx или 5xx)
+            response.raise_for_status()
+
+            # 5. Парсим JSON-ответ и извлекаем сгенерированный текст
+            response_json = response.json()
+            
+            # Добавим проверку на случай, если контент был заблокирован
+            if "candidates" not in response_json or not response_json["candidates"]:
+                block_reason = response_json.get("promptFeedback", {}).get("blockReason", "неизвестная причина")
+                error_message = f"Контент заблокирован. Причина: {block_reason}"
+                return error_message, error_message
+
+            text = response_json["candidates"][0]["content"]["parts"][0]["text"]
+            return text, text
+
+        except requests.exceptions.ProxyError as e:
+            error_message = f"Ошибка подключения к прокси. Убедитесь, что Nekobox запущен и слушает порт 2080. Ошибка: {e}"
+            print(error_message)
+            return error_message, error_message
+        except requests.exceptions.RequestException as e:
+            # Ловим все остальные ошибки requests (таймаут, проблемы с сетью и т.д.)
+            error_message = f"Произошла ошибка при обращении к API Gemini: {e}"
+            print(error_message)
+            return error_message, error_message
+        except (KeyError, IndexError) as e:
+            # Ловим ошибки, если структура JSON-ответа неожиданная
+            error_message = f"Не удалось разобрать ответ от API Gemini. Структура ответа изменилась. Ошибка: {e}"
+            print(error_message)
+            return error_message, error_message

services/llm_providers/gpt4free_provider.py

@@ -0,0 +1,47 @@
+# services/llm_providers/gpt4free_provider.py
+from g4f.client import Client
+from .base_provider import BaseLLMProvider
+
+class Gpt4FreeProvider(BaseLLMProvider):
+    """
+    Провайдер для работы с моделью GPT через библиотеку gpt4free.
+    
+    Этот класс реализует интерфейс BaseLLMProvider и предоставляет возможность
+    взаимодействия с различными LLM через сервис gpt4free, который не требует
+    API ключа для работы.
+    """
+    # gpt4free не требует API ключа
+    def __init__(self, api_key: str | None = None):
+        super().__init__(api_key)
+        self.client = Client()
+
+
+    def generate(self, model: str, system_prompt: str, user_prompt: str, temp: float):
+        """
+        Генерирует ответ от модели GPT с использованием gpt4free.
+        
+        Args:
+            model (str): Название модели для генерации ответа
+            system_prompt (str): Системное сообщение для контекста
+            user_prompt (str): Пользовательский запрос
+            temp (float): Температура генерации ( controls randomness of responses)
+            
+        Returns:
+            tuple: Кортеж из двух одинаковых строк - сгенерированного ответа и его копии
+        """
+        # temp в g4f может работать не для всех внутренних провайдеров
+        try:
+            response = self.client.chat.completions.create(
+            model=model,  # Пример модели, может варьироваться в зависимости от доступности провайдеров
+            messages=[
+                {"role": "system", "content": system_prompt},
+                {"role": "user", "content": user_prompt}
+                ],
+                temperature=temp
+            )
+            text = response.choices[0].message.content
+            return text, text
+        except Exception as e:
+            error_message = f"Ошибка при работе с gpt4free: {e}"
+            print(error_message)
+            return error_message, error_message

services/llm_providers/ionet_provider.py

@@ -0,0 +1,34 @@
+import requests
+from .base_provider import BaseLLMProvider
+
+class IoNetProvider(BaseLLMProvider):
+    def __init__(self, api_key: str):
+        super().__init__(api_key)
+        self.base_url = 'https://api.intelligence.io.solutions/api/v1'
+
+    def generate(self, model: str, system_prompt: str, user_prompt: str, temp: float):
+        url = f"{self.base_url}/chat/completions"
+        headers = {
+            "Content-Type": "application/json",
+            "Authorization": f"Bearer {self.api_key}"
+        }
+        
+        data = {
+            "model": model,
+            "messages": [
+                {'role': 'system', 'content': system_prompt},
+                {'role': 'user', 'content': user_prompt},
+            ],
+            "temperature": temp
+        }
+        
+        try:
+            response = requests.post(url, headers=headers, json=data)
+            response.raise_for_status()
+            
+            result = response.json()
+            text = str(result['choices'][0]['message']['content'])
+            return text, text  # Возвращаем как чистый текст, так и Markdown
+            
+        except requests.exceptions.RequestException as e:
+            raise Exception(f"Ошибка при запросе к IO.net API: {e}")