init

.air/settings.json

@@ -0,0 +1,3 @@
+{
+    "editor.guides": []
+}

.dockerignore

@@ -0,0 +1,17 @@
+.git
+.github
+.idea
+.air
+.agents
+.env
+.venv
+venv
+__pycache__
+*.py[cod]
+*.sqlite3
+*.sqlite3-*
+.coverage
+.pytest_cache
+.mypy_cache
+.ruff_cache
+tests

.gitignore

@@ -0,0 +1,5 @@
+.env
+.venv/
+__pycache__/
+*.py[cod]
+assistant_data.sqlite3*

.idea/.gitignore

@@ -0,0 +1,5 @@
+# Default ignored files
+/shelf/
+/workspace.xml
+# Editor-based HTTP Client requests
+/httpRequests/

.idea/KAndrusyak_Bot.iml

@@ -0,0 +1,10 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<module type="PYTHON_MODULE" version="4">
+  <component name="NewModuleRootManager">
+    <content url="file://$MODULE_DIR$">
+      <excludeFolder url="file://$MODULE_DIR$/.venv" />
+    </content>
+    <orderEntry type="jdk" jdkName="kandrusyak_bot" jdkType="Python SDK" />
+    <orderEntry type="sourceFolder" forTests="false" />
+  </component>
+</module>

.idea/inspectionProfiles/profiles_settings.xml

@@ -0,0 +1,6 @@
+<component name="InspectionProjectProfileManager">
+  <settings>
+    <option name="USE_PROJECT_PROFILE" value="false" />
+    <version value="1.0" />
+  </settings>
+</component>

.idea/misc.xml

@@ -0,0 +1,7 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project version="4">
+  <component name="Black">
+    <option name="sdkName" value="kandrusyak_bot" />
+  </component>
+  <component name="ProjectRootManager" version="2" project-jdk-name="kandrusyak_bot" project-jdk-type="Python SDK" />
+</project>

.idea/modules.xml

@@ -0,0 +1,8 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project version="4">
+  <component name="ProjectModuleManager">
+    <modules>
+      <module fileurl="file://$PROJECT_DIR$/.idea/KAndrusyak_Bot.iml" filepath="$PROJECT_DIR$/.idea/KAndrusyak_Bot.iml" />
+    </modules>
+  </component>
+</project>

.idea/vcs.xml

@@ -0,0 +1,6 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project version="4">
+  <component name="VcsDirectoryMappings">
+    <mapping directory="" vcs="Git" />
+  </component>
+</project>

AGENTS.md

@@ -0,0 +1,70 @@
+# AGENTS.md
+
+This repository contains a Python Telegram assistant bot with local Ollama integration and SQLite-backed storage.
+
+## Working scope
+
+- Keep changes targeted to this repository.
+- Prefer small, reviewable edits over broad refactors.
+- Do not commit secrets from [.env](air-file://v3c0f26v8ert02idn5b5/C:/Users/solda/PycharmProjects/KAndrusyak_Bot/.env?type=file&root=C%3A).
+- Treat [assistant_data.sqlite3](air-file://v3c0f26v8ert02idn5b5/C:/Users/solda/PycharmProjects/KAndrusyak_Bot/assistant_data.sqlite3?type=file&root=C%3A) as local runtime data, not source.
+
+## Project structure
+
+- [main.py](air-file://v3c0f26v8ert02idn5b5/C:/Users/solda/PycharmProjects/KAndrusyak_Bot/main.py?type=file&root=C%3A) is the compatible entry point.
+- [assistant_bot](air-file://v3c0f26v8ert02idn5b5/C:/Users/solda/PycharmProjects/KAndrusyak_Bot/assistant_bot?type=file&root=C%3A) contains the application package.
+- [tests](air-file://v3c0f26v8ert02idn5b5/C:/Users/solda/PycharmProjects/KAndrusyak_Bot/tests?type=file&root=C%3A) contains unittest-based tests.
+
+Key modules:
+
+- [application.py](air-file://v3c0f26v8ert02idn5b5/C:/Users/solda/PycharmProjects/KAndrusyak_Bot/assistant_bot/application.py?type=file&root=C%3A): Telegram app construction and handler registration.
+- [handlers.py](air-file://v3c0f26v8ert02idn5b5/C:/Users/solda/PycharmProjects/KAndrusyak_Bot/assistant_bot/handlers.py?type=file&root=C%3A): bot commands and message handling.
+- [agent.py](air-file://v3c0f26v8ert02idn5b5/C:/Users/solda/PycharmProjects/KAndrusyak_Bot/assistant_bot/agent.py?type=file&root=C%3A): agent loop and tool decision handling.
+- [storage.py](air-file://v3c0f26v8ert02idn5b5/C:/Users/solda/PycharmProjects/KAndrusyak_Bot/assistant_bot/storage.py?type=file&root=C%3A): SQLite persistence.
+- [ollama.py](air-file://v3c0f26v8ert02idn5b5/C:/Users/solda/PycharmProjects/KAndrusyak_Bot/assistant_bot/ollama.py?type=file&root=C%3A): Ollama client integration.
+- [reminders.py](air-file://v3c0f26v8ert02idn5b5/C:/Users/solda/PycharmProjects/KAndrusyak_Bot/assistant_bot/reminders.py?type=file&root=C%3A): reminder parsing and scheduling logic.
+
+## Environment and dependencies
+
+- Python project with dependencies from [requirements.txt](air-file://v3c0f26v8ert02idn5b5/C:/Users/solda/PycharmProjects/KAndrusyak_Bot/requirements.txt?type=file&root=C%3A).
+- Required environment variable: `BOT_TOKEN`.
+- Supported optional variables include `OLLAMA_BASE_URL`, `OLLAMA_MODEL`, `ASSISTANT_DB`, and `ASSISTANT_TIMEZONE`.
+
+## Run and test
+
+Install dependencies:
+
+```powershell
+python -m pip install -r requirements.txt
+```
+
+Run the bot:
+
+```powershell
+python main.py
+```
+
+Alternative package entry point:
+
+```powershell
+python -m assistant_bot
+```
+
+Run tests:
+
+```powershell
+python -m unittest discover -v
+```
+
+## Change guidance
+
+- Preserve the existing stdlib `unittest` style unless there is a clear reason to introduce a new test runner.
+- Keep Telegram handlers, storage logic, and Ollama client concerns separated by module.
+- Prefer configuration through environment variables instead of hardcoded values.
+- When changing reminder parsing, storage behavior, or agent decision parsing, update or add tests in [test_core.py](air-file://v3c0f26v8ert02idn5b5/C:/Users/solda/PycharmProjects/KAndrusyak_Bot/tests/test_core.py?type=file&root=C%3A).
+- Avoid editing generated caches under `__pycache__`.
+
+## Operational cautions
+
+- Bot startup depends on a valid Telegram token and reachable Ollama endpoint when model-backed features are used.
+- SQLite files may be environment-specific; avoid relying on checked-in runtime state for tests.

Dockerfile

@@ -0,0 +1,64 @@
+# Build the local GPU image with:  docker build --target local -t kandrusyak-bot:local .
+FROM nvidia/cuda:12.3.2-cudnn9-runtime-ubuntu22.04 AS local
+
+ENV PYTHONDONTWRITEBYTECODE=1 \
+    PYTHONUNBUFFERED=1 \
+    PIP_DISABLE_PIP_VERSION_CHECK=1 \
+    ASSISTANT_MODE=local \
+    ASSISTANT_DB=/data/assistant_data.sqlite3 \
+    HF_HOME=/data/huggingface
+
+WORKDIR /app
+
+RUN apt-get update \
+    && apt-get install --no-install-recommends --yes python3 python3-pip tzdata \
+    && rm -rf /var/lib/apt/lists/*
+
+COPY requirements-common.txt requirements-local.txt ./
+RUN python3 -m pip install --no-cache-dir --requirement requirements-local.txt
+
+RUN groupadd --system bot \
+    && useradd --system --gid bot --create-home bot \
+    && mkdir /data \
+    && chown bot:bot /data
+
+COPY --chown=bot:bot assistant_bot ./assistant_bot
+COPY --chown=bot:bot main.py ./main.py
+
+USER bot
+VOLUME ["/data"]
+STOPSIGNAL SIGTERM
+CMD ["python3", "-m", "assistant_bot"]
+
+
+# Build the cloud image with:  docker build --target yandex -t kandrusyak-bot:yandex .
+# This is the default final target when --target is omitted.
+FROM python:3.12.13-slim-bookworm AS yandex
+
+ENV PYTHONDONTWRITEBYTECODE=1 \
+    PYTHONUNBUFFERED=1 \
+    PIP_DISABLE_PIP_VERSION_CHECK=1 \
+    ASSISTANT_MODE=yandex \
+    ASSISTANT_DB=/data/assistant_data.sqlite3
+
+WORKDIR /app
+
+RUN apt-get update \
+    && apt-get install --no-install-recommends --yes tzdata \
+    && rm -rf /var/lib/apt/lists/*
+
+COPY requirements-common.txt requirements-yandex.txt ./
+RUN python -m pip install --no-cache-dir --requirement requirements-yandex.txt
+
+RUN groupadd --system bot \
+    && useradd --system --gid bot --create-home bot \
+    && mkdir /data \
+    && chown bot:bot /data
+
+COPY --chown=bot:bot assistant_bot ./assistant_bot
+COPY --chown=bot:bot main.py ./main.py
+
+USER bot
+VOLUME ["/data"]
+STOPSIGNAL SIGTERM
+CMD ["python", "-m", "assistant_bot"]

README.md

@@ -0,0 +1,141 @@
+# KAndrusyak Bot
+
+Telegram-ассистент с двумя режимами AI: локальными Ollama + faster-whisper или
+облачными YandexGPT + SpeechKit через Yandex AI Studio,
+контекстом текущего диалога,
+долговременной памятью, заметками, напоминаниями и отслеживанием статусов.
+
+Вся переписка сохраняется в SQLite отдельно для каждого чата. В текущий контекст попадают
+последние реплики активной темы, причем более свежие сообщения имеют больший приоритет.
+При явной смене темы модель начинает новый контекст, не удаляя архив. К старому обсуждению
+можно вернуться обычной просьбой к боту или найти сообщения командой `/history ключевые слова`.
+Команда `/new` вручную начинает новую тему и также сохраняет предыдущую переписку.
+
+## Структура
+
+- `main.py` — совместимая точка входа;
+- `assistant_bot/application.py` — сборка Telegram-приложения;
+- `assistant_bot/config.py` — переменные окружения и настройки;
+- `assistant_bot/storage.py` — SQLite-хранилище;
+- `assistant_bot/ollama.py` — HTTP-клиент Ollama;
+- `assistant_bot/yandex_ai.py` — адаптер Yandex AI Studio SDK;
+- `assistant_bot/agent.py` — агентный цикл и выполнение внутренних tools;
+- `assistant_bot/handlers.py` — Telegram-команды и сообщения;
+- `assistant_bot/reminders.py` — разбор времени напоминаний;
+- `assistant_bot/jobs.py` — фоновые задачи;
+- `tests/` — модульные тесты ядра.
+
+## Запуск
+
+```powershell
+conda activate kandrusyak_bot
+python -m pip install -r requirements.txt
+python main.py
+```
+
+Также пакет можно запустить командой `python -m assistant_bot`.
+
+Обязательная настройка в `.env`:
+
+```dotenv
+BOT_TOKEN=...
+ASSISTANT_MODE=local
+```
+
+`ASSISTANT_MODE` принимает `local` (значение по умолчанию) или `yandex`.
+Общие дополнительные настройки: `ASSISTANT_DB`, `ASSISTANT_TIMEZONE` и
+`VOICE_MAX_DURATION_SECONDS`.
+
+### Локальный режим
+
+В режиме `local` текст обрабатывает Ollama, а голосовые сообщения распознаются
+локально через faster-whisper. Настройки:
+
+```dotenv
+ASSISTANT_MODE=local
+OLLAMA_BASE_URL=http://localhost:11434
+OLLAMA_MODEL=qwen3.5:9b
+```
+
+При первом голосовом сообщении модель faster-whisper будет загружена
+автоматически. Настройки распознавания:
+
+```dotenv
+WHISPER_MODEL=large-v3
+WHISPER_DEVICE=cuda
+WHISPER_COMPUTE_TYPE=int8
+WHISPER_LANGUAGE=ru
+VOICE_MAX_DURATION_SECONDS=120
+```
+
+По умолчанию используется полная модель `large-v3` на NVIDIA GPU с вычислениями INT8
+и beam size 5. Крупная модель сохраняет приоритет качества, а INT8 уменьшает расход
+видеопамяти. Для этого режима необходимы CUDA 12, cuBLAS для CUDA 12 и cuDNN 9.
+Чтобы определять язык автоматически, укажите `WHISPER_LANGUAGE=auto`.
+
+На Windows используется зафиксированная версия CTranslate2 4.6.0: более новые
+сборки 4.7.x могут аварийно завершаться при инициализации модели. Проект настроен
+для Conda-окружения `kandrusyak_bot`. Полный набор DLL cuDNN 9 должен находиться
+в системном `PATH` или рядом с библиотекой CTranslate2.
+
+### Режим Yandex AI Studio
+
+В режиме `yandex` бот использует официальный Yandex AI Studio SDK: YandexGPT для
+текста и SpeechKit для Telegram-аудио в формате OGG Opus. Создайте API-ключ с
+необходимыми ролями и укажите стандартные переменные авторизации SDK:
+
+```dotenv
+ASSISTANT_MODE=yandex
+YANDEX_CLOUD_FOLDER=...
+YC_API_KEY=...
+YANDEX_CLOUD_MODEL=yandexgpt/latest
+YANDEX_STT_MODEL=general
+YANDEX_STT_LANGUAGE=ru-RU
+VOICE_MAX_DURATION_SECONDS=120
+```
+
+Значения моделей можно посмотреть командой `/models`, а пользовательский выбор
+сохранить командой `/model имя-модели`. Выбор хранится отдельно для локального и
+облачного режимов. Короткое имя облачной модели автоматически преобразуется в URI
+`gpt://<YANDEX_CLOUD_FOLDER>/<YANDEX_CLOUD_MODEL>`.
+
+## Docker
+
+Один `Dockerfile` содержит два независимых target. Облачный Yandex-образ является
+target по умолчанию и не содержит CUDA, faster-whisper и CTranslate2:
+
+```bash
+docker build --target yandex -t kandrusyak-bot:yandex .
+docker run -d \
+  --name kandrusyak-bot \
+  --restart unless-stopped \
+  --env-file .env \
+  -e ASSISTANT_MODE=yandex \
+  -v kandrusyak-data:/data \
+  kandrusyak-bot:yandex
+```
+
+Локальный образ содержит CUDA 12, cuDNN 9 и faster-whisper. Для него требуется
+NVIDIA Container Toolkit:
+
+```bash
+docker build --target local -t kandrusyak-bot:local .
+docker run -d \
+  --name kandrusyak-bot \
+  --restart unless-stopped \
+  --gpus all \
+  --env-file .env \
+  -e ASSISTANT_MODE=local \
+  -v kandrusyak-data:/data \
+  kandrusyak-bot:local
+```
+
+Ollama должна быть доступна контейнеру по адресу из `OLLAMA_BASE_URL`; адрес
+`localhost` внутри контейнера указывает на сам контейнер, а не на Linux-хост.
+
+## Проверка
+
+```powershell
+conda activate kandrusyak_bot
+python -m unittest discover -v
+```

assistant_bot/__init__.py

@@ -0,0 +1,2 @@
+"""Personal Telegram assistant package."""
+

assistant_bot/__main__.py

@@ -0,0 +1,5 @@
+from .application import run
+
+
+if __name__ == "__main__":
+    run()

assistant_bot/agent.py

@@ -0,0 +1,516 @@
+import json
+import logging
+import re
+from datetime import datetime, timezone
+from typing import Any
+from zoneinfo import ZoneInfo
+
+from telegram import Update
+from telegram.constants import ChatAction
+from telegram.ext import ContextTypes
+
+from .ai import AIClientError
+from .config import CONVERSATION_HISTORY_LIMIT, MAX_AGENT_STEPS
+from .models import AgentDecision, ReminderParseResult
+from .prompts import ASSISTANT_SYSTEM_PROMPT
+from .reminders import parse_reminder
+from .storage import AssistantStorage
+from .telegram_utils import (
+    edit_or_reply,
+    get_ai_client,
+    get_storage,
+    get_tz,
+    parse_positive_int,
+    require_user_id,
+)
+from .time_utils import format_local_dt, to_utc_iso, utc_now
+
+
+logger = logging.getLogger(__name__)
+
+
+def build_assistant_context(storage: AssistantStorage, user_id: int, tz: ZoneInfo) -> str:
+    memories = storage.list_memories(user_id, limit=20)
+    notes = storage.list_notes(user_id, limit=10)
+    reminders = storage.list_reminders(user_id, limit=10)
+    tracked_items = storage.list_tracked_items(user_id, limit=20)
+
+    sections: list[str] = []
+    if memories:
+        sections.append(
+            "Долговременная память:\n"
+            + "\n".join(f"- #{row['id']}: {row['text']}" for row in memories)
+        )
+    if notes:
+        sections.append(
+            "Последние заметки:\n"
+            + "\n".join(f"- #{row['id']}: {row['text']}" for row in notes)
+        )
+    if reminders:
+        sections.append(
+            "Активные напоминания:\n"
+            + "\n".join(
+                f"- #{row['id']} {format_local_dt(row['remind_at'], tz)}: {row['text']}"
+                for row in reminders
+            )
+        )
+    if tracked_items:
+        sections.append(
+            "Отслеживаемые статусы:\n"
+            + "\n".join(
+                f"- #{row['id']} {row['title']}: {row['status']}"
+                for row in tracked_items
+            )
+        )
+
+    if not sections:
+        return "Долговременный контекст пока пуст."
+    return "\n\n".join(sections)
+
+
+def json_dumps(data: Any) -> str:
+    return json.dumps(data, ensure_ascii=False, indent=2)
+
+
+def build_agent_tool_prompt(tz: ZoneInfo) -> str:
+    now_local = datetime.now(tz).strftime("%Y-%m-%d %H:%M")
+    return (
+        "Ты работаешь как агент с внутренними tools. Пользователь пишет свободным текстом, "
+        "а Telegram-команды ему не нужны.\n"
+        f"Текущее локальное время: {now_local}. Таймзона: {tz.key}.\n\n"
+        "Отвечай СТРОГО одним JSON-объектом без Markdown и без текста вокруг.\n"
+        "Если нужно выполнить действие, верни tool_calls. Если действие уже выполнено "
+        "или tool не нужен, верни final.\n\n"
+        "Форматы ответа:\n"
+        "{\"tool_calls\":[{\"name\":\"create_note\",\"arguments\":{\"text\":\"...\"}}],\"reset_context\":false}\n"
+        "{\"final\":\"Короткий ответ пользователю\",\"reset_context\":false}\n\n"
+        "Доступные tools:\n"
+        "- get_current_datetime {} - узнать текущее локальное и UTC-время.\n"
+        "- remember {\"text\": string} - сохранить важный долгосрочный факт о пользователе.\n"
+        "- list_memory {\"limit\": number} - показать сохраненную память.\n"
+        "- delete_memory {\"id\": number} - удалить запись памяти.\n"
+        "- create_note {\"text\": string} - сохранить заметку.\n"
+        "- list_notes {\"limit\": number} - показать заметки.\n"
+        "- delete_note {\"id\": number} - удалить заметку.\n"
+        "- create_reminder {\"when\": string, \"text\": string} - поставить напоминание. "
+        "when можно указывать как '30m', 'через 2 часа', '18:30', '2026-07-16 18:30'. "
+        "Если пользователь говорит 'завтра/послезавтра/через неделю', сам рассчитай "
+        "дату от текущего локального времени и передай 'YYYY-MM-DD HH:MM'.\n"
+        "- list_reminders {\"limit\": number} - показать активные напоминания.\n"
+        "- cancel_reminder {\"id\": number} - отменить напоминание.\n"
+        "- create_status {\"title\": string, \"status\": string} - начать отслеживать статус.\n"
+        "- list_statuses {\"limit\": number} - показать отслеживаемые статусы.\n"
+        "- update_status {\"id\": number, \"status\": string} - обновить статус.\n"
+        "- delete_status {\"id\": number} - удалить отслеживаемый объект.\n\n"
+        "- search_conversation {\"query\": string, \"limit\": number} - найти старое обсуждение "
+        "во всей сохраненной переписке по содержательным ключевым словам.\n\n"
+        "Правила:\n"
+        "- Для просьб 'запомни', 'сохрани как факт', 'учти на будущее' используй remember.\n"
+        "- Для заметок используй create_note, для напоминаний create_reminder, "
+        "для контроля дел/заявок/ожиданий create_status или update_status.\n"
+        "- Если для действия не хватает данных, не вызывай tool, а задай уточняющий вопрос через final.\n"
+        "- Учитывай историю диалога для коротких ответов на уточняющие вопросы. "
+        "Если новый запрос явно начинает другую, не связанную с историей тему, не опирайся на старую тему "
+        "и верни reset_context=true. Для продолжения темы и сомнительных случаев верни false.\n"
+        "- Если пользователь просит найти, вспомнить или продолжить старое обсуждение, вызови "
+        "search_conversation. Передавай в query только ключевые слова темы, без общих слов. "
+        "Результаты поиска содержат соседние реплики; более свежие совпадения при прочих равных важнее.\n"
+        "- После результата tool верни final с кратким подтверждением или следующим tool_calls."
+    )
+
+
+def extract_json_object(raw_text: str) -> dict[str, Any] | None:
+    candidates = [raw_text.strip()]
+    code_block = re.search(r"```(?:json)?\s*(.*?)```", raw_text, re.IGNORECASE | re.DOTALL)
+    if code_block:
+        candidates.insert(0, code_block.group(1).strip())
+
+    first_brace = raw_text.find("{")
+    last_brace = raw_text.rfind("}")
+    if 0 <= first_brace < last_brace:
+        candidates.append(raw_text[first_brace : last_brace + 1])
+
+    for candidate in candidates:
+        if not candidate:
+            continue
+        try:
+            parsed = json.loads(candidate)
+        except json.JSONDecodeError:
+            continue
+        if isinstance(parsed, dict):
+            return parsed
+
+    return None
+
+
+def parse_agent_decision(raw_text: str) -> AgentDecision:
+    parsed = extract_json_object(raw_text)
+    if parsed is None:
+        return AgentDecision(final=raw_text.strip(), tool_calls=[], reset_context=False)
+
+    raw_calls = parsed.get("tool_calls")
+    if raw_calls is None and parsed.get("tool"):
+        raw_calls = [
+            {
+                "name": parsed.get("tool"),
+                "arguments": parsed.get("arguments", {}),
+            }
+        ]
+
+    tool_calls: list[dict[str, Any]] = []
+    if isinstance(raw_calls, list):
+        for raw_call in raw_calls:
+            if not isinstance(raw_call, dict):
+                continue
+            name = raw_call.get("name") or raw_call.get("tool")
+            arguments = raw_call.get("arguments", {})
+            if isinstance(name, str):
+                tool_calls.append(
+                    {
+                        "name": name.strip(),
+                        "arguments": arguments if isinstance(arguments, dict) else {},
+                    }
+                )
+
+    final = parsed.get("final") or parsed.get("answer")
+    reset_context = parsed.get("reset_context") is True
+    if isinstance(final, str) and final.strip():
+        return AgentDecision(
+            final=final.strip(),
+            tool_calls=tool_calls,
+            reset_context=reset_context,
+        )
+    return AgentDecision(final=None, tool_calls=tool_calls, reset_context=reset_context)
+
+
+def save_conversation_exchange(
+    storage: AssistantStorage,
+    user_id: int,
+    chat_id: int,
+    prompt: str,
+    answer: str,
+    reset_context: bool,
+) -> None:
+    if reset_context:
+        storage.start_new_conversation(user_id, chat_id)
+    storage.add_conversation_exchange(user_id, chat_id, prompt, answer)
+
+
+def coerce_positive_int(value: Any, default: int, maximum: int = 50) -> int:
+    try:
+        parsed = int(value)
+    except (TypeError, ValueError):
+        return default
+    if parsed <= 0:
+        return default
+    return min(parsed, maximum)
+
+
+def coerce_required_text(arguments: dict[str, Any], key: str) -> str:
+    value = arguments.get(key, "")
+    return str(value).strip()
+
+
+def parse_agent_reminder(when: str, text: str, tz: ZoneInfo) -> ReminderParseResult | None:
+    parsed = parse_reminder(f"{when} {text}".strip(), tz)
+    if parsed:
+        return parsed
+
+    normalized_when = when.strip().replace("T", " ")
+    try:
+        parsed_dt = datetime.fromisoformat(normalized_when)
+    except ValueError:
+        return None
+
+    if parsed_dt.tzinfo is None:
+        parsed_dt = parsed_dt.replace(tzinfo=tz)
+    return ReminderParseResult(parsed_dt.astimezone(timezone.utc), text)
+
+
+def execute_agent_tool(
+    name: str,
+    arguments: dict[str, Any],
+    storage: AssistantStorage,
+    user_id: int,
+    chat_id: int,
+    tz: ZoneInfo,
+) -> dict[str, Any]:
+    tool_name = name.strip().lower()
+
+    if tool_name == "get_current_datetime":
+        now = utc_now()
+        return {
+            "ok": True,
+            "local": now.astimezone(tz).strftime("%Y-%m-%d %H:%M:%S"),
+            "utc": to_utc_iso(now),
+            "timezone": tz.key,
+        }
+
+    if tool_name == "remember":
+        text = coerce_required_text(arguments, "text")
+        if not text:
+            return {"ok": False, "error": "text is required"}
+        memory_id = storage.add_memory(user_id, text)
+        return {"ok": True, "id": memory_id, "text": text}
+
+    if tool_name == "list_memory":
+        rows = storage.list_memories(user_id, coerce_positive_int(arguments.get("limit"), 20))
+        return {"ok": True, "items": rows}
+
+    if tool_name == "delete_memory":
+        memory_id = parse_positive_int(str(arguments.get("id", "")))
+        if memory_id is None:
+            return {"ok": False, "error": "valid id is required"}
+        return {"ok": storage.delete_memory(user_id, memory_id), "id": memory_id}
+
+    if tool_name == "create_note":
+        text = coerce_required_text(arguments, "text")
+        if not text:
+            return {"ok": False, "error": "text is required"}
+        note_id = storage.add_note(user_id, text)
+        return {"ok": True, "id": note_id, "text": text}
+
+    if tool_name == "list_notes":
+        rows = storage.list_notes(user_id, coerce_positive_int(arguments.get("limit"), 20))
+        return {"ok": True, "items": rows}
+
+    if tool_name == "delete_note":
+        note_id = parse_positive_int(str(arguments.get("id", "")))
+        if note_id is None:
+            return {"ok": False, "error": "valid id is required"}
+        return {"ok": storage.delete_note(user_id, note_id), "id": note_id}
+
+    if tool_name == "create_reminder":
+        when = coerce_required_text(arguments, "when")
+        text = coerce_required_text(arguments, "text")
+        if not when or not text:
+            return {"ok": False, "error": "when and text are required"}
+        parsed = parse_agent_reminder(when, text, tz)
+        if not parsed:
+            return {"ok": False, "error": "could not parse reminder time"}
+        if parsed.remind_at_utc <= utc_now():
+            return {"ok": False, "error": "reminder time must be in the future"}
+
+        reminder_id = storage.add_reminder(user_id, chat_id, parsed.text, parsed.remind_at_utc)
+        return {
+            "ok": True,
+            "id": reminder_id,
+            "text": parsed.text,
+            "remind_at": format_local_dt(parsed.remind_at_utc, tz),
+        }
+
+    if tool_name == "list_reminders":
+        rows = storage.list_reminders(user_id, coerce_positive_int(arguments.get("limit"), 20))
+        for row in rows:
+            row["remind_at_local"] = format_local_dt(row["remind_at"], tz)
+        return {"ok": True, "items": rows}
+
+    if tool_name == "cancel_reminder":
+        reminder_id = parse_positive_int(str(arguments.get("id", "")))
+        if reminder_id is None:
+            return {"ok": False, "error": "valid id is required"}
+        return {"ok": storage.cancel_reminder(user_id, reminder_id), "id": reminder_id}
+
+    if tool_name == "create_status":
+        title = coerce_required_text(arguments, "title")
+        status = coerce_required_text(arguments, "status") or "open"
+        if not title:
+            return {"ok": False, "error": "title is required"}
+        item_id = storage.add_tracked_item(user_id, title, status)
+        return {"ok": True, "id": item_id, "title": title, "status": status}
+
+    if tool_name == "list_statuses":
+        rows = storage.list_tracked_items(user_id, coerce_positive_int(arguments.get("limit"), 30))
+        return {"ok": True, "items": rows}
+
+    if tool_name == "update_status":
+        item_id = parse_positive_int(str(arguments.get("id", "")))
+        status = coerce_required_text(arguments, "status")
+        if item_id is None or not status:
+            return {"ok": False, "error": "valid id and status are required"}
+        return {"ok": storage.set_tracked_status(user_id, item_id, status), "id": item_id, "status": status}
+
+    if tool_name == "delete_status":
+        item_id = parse_positive_int(str(arguments.get("id", "")))
+        if item_id is None:
+            return {"ok": False, "error": "valid id is required"}
+        return {"ok": storage.delete_tracked_item(user_id, item_id), "id": item_id}
+
+    if tool_name == "search_conversation":
+        query = coerce_required_text(arguments, "query")
+        if not query:
+            return {"ok": False, "error": "query is required"}
+        limit = coerce_positive_int(arguments.get("limit"), 5, maximum=10)
+        hits = storage.search_conversation_messages(user_id, chat_id, query, limit)
+        discussions = [
+            {
+                "matched_message_id": hit["id"],
+                "score": hit["score"],
+                "messages": storage.conversation_message_window(
+                    user_id,
+                    chat_id,
+                    int(hit["id"]),
+                ),
+            }
+            for hit in hits
+        ]
+        return {"ok": True, "query": query, "discussions": discussions}
+
+    return {"ok": False, "error": f"unknown tool: {name}"}
+
+
+async def run_agent_prompt(
+    update: Update,
+    context: ContextTypes.DEFAULT_TYPE,
+    prompt: str,
+) -> None:
+    if not update.effective_message:
+        return
+
+    user_id = require_user_id(update)
+    if user_id is None:
+        await update.effective_message.reply_text("Не могу определить пользователя.")
+        return
+
+    storage = get_storage(context)
+    ai_client = get_ai_client(context)
+    tz = get_tz(context)
+    model = ai_client.normalize_model(
+        storage.get_user_model(user_id, ai_client.provider)
+    )
+    chat_id = int(update.effective_message.chat_id)
+    assistant_context = build_assistant_context(storage, user_id, tz)
+    conversation_history = storage.list_conversation_messages(
+        user_id,
+        chat_id,
+        limit=CONVERSATION_HISTORY_LIMIT,
+    )
+    messages = [
+        {"role": "system", "content": ASSISTANT_SYSTEM_PROMPT},
+        {"role": "system", "content": build_agent_tool_prompt(tz)},
+        {"role": "system", "content": assistant_context},
+        {
+            "role": "system",
+            "content": (
+                "Далее идет недавняя история текущей темы от старых реплик к новым. "
+                "Чем старше реплика, тем меньше ее приоритет; при противоречии опирайся "
+                "на более новые сообщения. Архив других тем доступен через search_conversation."
+            ),
+        },
+        *(
+            {"role": str(row["role"]), "content": str(row["content"])}
+            for row in conversation_history
+        ),
+        {"role": "user", "content": prompt},
+    ]
+
+    await context.bot.send_chat_action(
+        chat_id=chat_id,
+        action=ChatAction.TYPING,
+    )
+    placeholder = await update.effective_message.reply_text(
+        f"Думаю через {ai_client.display_name} / {model}..."
+    )
+
+    try:
+        last_tool_results: list[dict[str, Any]] = []
+        reset_context = False
+        for _step in range(MAX_AGENT_STEPS):
+            raw_answer = await ai_client.chat(model, messages, json_mode=True)
+            decision = parse_agent_decision(raw_answer)
+            reset_context = reset_context or decision.reset_context
+
+            if decision.tool_calls:
+                tool_results = []
+                for call in decision.tool_calls:
+                    result = execute_agent_tool(
+                        name=str(call.get("name", "")),
+                        arguments=call.get("arguments", {}),
+                        storage=storage,
+                        user_id=user_id,
+                        chat_id=chat_id,
+                        tz=tz,
+                    )
+                    tool_results.append(
+                        {
+                            "name": call.get("name"),
+                            "arguments": call.get("arguments", {}),
+                            "result": result,
+                        }
+                    )
+
+                last_tool_results = tool_results
+                messages.append({"role": "assistant", "content": json_dumps({"tool_calls": decision.tool_calls})})
+                messages.append(
+                    {
+                        "role": "user",
+                        "content": (
+                            "Результаты tools:\n"
+                            f"{json_dumps({'tool_results': tool_results})}\n"
+                            "Продолжи. Верни либо следующий tool_calls, либо final. "
+                            "Ответ снова строго JSON."
+                        ),
+                    }
+                )
+                continue
+
+            if decision.final:
+                await edit_or_reply(placeholder, decision.final)
+                save_conversation_exchange(
+                    storage,
+                    user_id,
+                    chat_id,
+                    prompt,
+                    decision.final,
+                    reset_context,
+                )
+                return
+
+            messages.append({"role": "assistant", "content": raw_answer})
+
+        final_answer = await ai_client.chat(
+            model,
+            [
+                *messages,
+                {
+                    "role": "user",
+                    "content": (
+                        "Лимит tool-шагов исчерпан. Больше не вызывай tools. "
+                        f"Последние результаты tools: {json_dumps(last_tool_results)}. "
+                        "Сформулируй короткий финальный ответ пользователю обычным текстом."
+                    ),
+                },
+            ],
+        )
+        await edit_or_reply(placeholder, final_answer)
+        save_conversation_exchange(
+            storage,
+            user_id,
+            chat_id,
+            prompt,
+            final_answer,
+            reset_context,
+        )
+    except AIClientError as exc:
+        if ai_client.provider == "local":
+            hint = f"Проверь, что Ollama запущена и модель установлена: ollama pull {model}"
+        else:
+            hint = (
+                "Проверь YANDEX_CLOUD_FOLDER, YC_API_KEY "
+                "и доступ к выбранной модели."
+            )
+        await placeholder.edit_text(
+            f"Не получилось вызвать {ai_client.display_name}.\n{exc}\n\n{hint}"
+        )
+        return
+    except Exception:
+        logger.exception("Failed to run AI prompt")
+        await placeholder.edit_text("Произошла внутренняя ошибка при запросе к модели.")
+
+
+async def run_ai_prompt(
+    update: Update,
+    context: ContextTypes.DEFAULT_TYPE,
+    prompt: str,
+) -> None:
+    await run_agent_prompt(update, context, prompt)

assistant_bot/ai.py

@@ -0,0 +1,24 @@
+from typing import Protocol
+
+
+class AIClientError(RuntimeError):
+    """Raised when the configured AI provider cannot complete a request."""
+
+
+class AIClient(Protocol):
+    provider: str
+    display_name: str
+
+    async def list_models(self) -> list[str]:
+        ...
+
+    def normalize_model(self, model: str) -> str:
+        ...
+
+    async def chat(
+        self,
+        model: str,
+        messages: list[dict[str, str]],
+        json_mode: bool = False,
+    ) -> str:
+        ...

assistant_bot/application.py

@@ -0,0 +1,112 @@
+import logging
+
+from telegram import Update
+from telegram.ext import (
+    Application,
+    CommandHandler,
+    InlineQueryHandler,
+    MessageHandler,
+    filters,
+)
+
+from .ai import AIClient
+from .config import (
+    get_assistant_mode,
+    get_bot_token,
+    get_db_path,
+    get_local_timezone,
+    get_ollama_base_url,
+    get_voice_max_duration_seconds,
+    get_whisper_compute_type,
+    get_whisper_device,
+    get_whisper_language,
+    get_whisper_model,
+    get_yandex_cloud_folder,
+    get_yandex_stt_language,
+    get_yandex_stt_model,
+)
+from .handlers import (
+    ask_command,
+    help_command,
+    history_command,
+    inline_query,
+    model_command,
+    models_command,
+    new_conversation_command,
+    private_text,
+    private_voice,
+    start,
+)
+from .jobs import post_init
+from .ollama import OllamaClient
+from .storage import AssistantStorage
+from .speech import SpeechRecognizer, SpeechTranscriber
+from .yandex_ai import YandexAIClient, YandexSpeechRecognizer
+
+
+def configure_logging() -> None:
+    logging.basicConfig(
+        format="%(asctime)s - %(name)s - %(levelname)s - %(message)s",
+        level=logging.INFO,
+    )
+    logging.getLogger("httpx").setLevel(logging.WARNING)
+
+
+def create_application() -> Application:
+    storage = AssistantStorage(get_db_path())
+    mode = get_assistant_mode()
+    application = (
+        Application.builder()
+        .token(get_bot_token())
+        .post_init(post_init)
+        .build()
+    )
+    application.bot_data["storage"] = storage
+    ai_client: AIClient
+    speech_recognizer: SpeechTranscriber
+    if mode == "local":
+        ai_client = OllamaClient(get_ollama_base_url())
+        speech_recognizer = SpeechRecognizer(
+            model_name=get_whisper_model(),
+            device=get_whisper_device(),
+            compute_type=get_whisper_compute_type(),
+            language=get_whisper_language(),
+        )
+    else:
+        folder_id = get_yandex_cloud_folder()
+        ai_client = YandexAIClient(folder_id=folder_id)
+        speech_recognizer = YandexSpeechRecognizer(
+            folder_id=folder_id,
+            language=get_yandex_stt_language(),
+            model=get_yandex_stt_model(),
+        )
+
+    application.bot_data["ai_client"] = ai_client
+    application.bot_data["assistant_mode"] = mode
+    application.bot_data["timezone"] = get_local_timezone()
+    application.bot_data["speech_recognizer"] = speech_recognizer
+    application.bot_data["voice_max_duration"] = get_voice_max_duration_seconds()
+
+    application.add_handler(CommandHandler("start", start))
+    application.add_handler(CommandHandler("help", help_command))
+    application.add_handler(CommandHandler("ask", ask_command))
+    application.add_handler(CommandHandler("models", models_command))
+    application.add_handler(CommandHandler("model", model_command))
+    application.add_handler(CommandHandler("new", new_conversation_command))
+    application.add_handler(CommandHandler("history", history_command))
+    application.add_handler(InlineQueryHandler(inline_query))
+    application.add_handler(
+        MessageHandler(filters.VOICE & filters.ChatType.PRIVATE, private_voice)
+    )
+    application.add_handler(
+        MessageHandler(
+            filters.TEXT & filters.ChatType.PRIVATE & ~filters.COMMAND,
+            private_text,
+        )
+    )
+    return application
+
+
+def run() -> None:
+    configure_logging()
+    create_application().run_polling(allowed_updates=Update.ALL_TYPES)

assistant_bot/config.py

@@ -0,0 +1,213 @@
+import logging
+import os
+from pathlib import Path
+from zoneinfo import ZoneInfo, ZoneInfoNotFoundError
+
+
+logger = logging.getLogger(__name__)
+
+PROJECT_ROOT = Path(__file__).resolve().parent.parent
+ENV_FILE = PROJECT_ROOT / ".env"
+
+TOKEN_ENV_NAME = "BOT_TOKEN"
+DB_ENV_NAME = "ASSISTANT_DB"
+MODE_ENV_NAME = "ASSISTANT_MODE"
+OLLAMA_BASE_URL_ENV_NAME = "OLLAMA_BASE_URL"
+OLLAMA_MODEL_ENV_NAME = "OLLAMA_MODEL"
+YANDEX_CLOUD_FOLDER_ENV_NAME = "YANDEX_CLOUD_FOLDER"
+YANDEX_CLOUD_MODEL_ENV_NAME = "YANDEX_CLOUD_MODEL"
+YANDEX_STT_MODEL_ENV_NAME = "YANDEX_STT_MODEL"
+YANDEX_STT_LANGUAGE_ENV_NAME = "YANDEX_STT_LANGUAGE"
+TIMEZONE_ENV_NAME = "ASSISTANT_TIMEZONE"
+WHISPER_MODEL_ENV_NAME = "WHISPER_MODEL"
+WHISPER_DEVICE_ENV_NAME = "WHISPER_DEVICE"
+WHISPER_COMPUTE_TYPE_ENV_NAME = "WHISPER_COMPUTE_TYPE"
+WHISPER_LANGUAGE_ENV_NAME = "WHISPER_LANGUAGE"
+VOICE_MAX_DURATION_ENV_NAME = "VOICE_MAX_DURATION_SECONDS"
+
+DEFAULT_DB_FILE = PROJECT_ROOT / "assistant_data.sqlite3"
+DEFAULT_MODE = "local"
+DEFAULT_OLLAMA_BASE_URL = "http://localhost:11434"
+DEFAULT_OLLAMA_MODEL = "qwen3.5:9b"
+DEFAULT_YANDEX_CLOUD_MODEL = "yandexgpt/latest"
+DEFAULT_YANDEX_STT_MODEL = "general"
+DEFAULT_YANDEX_STT_LANGUAGE = "ru-RU"
+DEFAULT_TIMEZONE = "Europe/Moscow"
+DEFAULT_WHISPER_MODEL = "large-v3"
+DEFAULT_WHISPER_DEVICE = "cuda"
+DEFAULT_WHISPER_COMPUTE_TYPE = "int8"
+DEFAULT_WHISPER_LANGUAGE = "ru"
+DEFAULT_VOICE_MAX_DURATION_SECONDS = 120
+
+REMINDER_POLL_SECONDS = 30
+MAX_TELEGRAM_MESSAGE_LENGTH = 3900
+MAX_AGENT_STEPS = 5
+CONVERSATION_HISTORY_LIMIT = 12
+
+HTML_FORMATS = (
+    ("Bold", "b"),
+    ("Italic", "i"),
+)
+
+
+def load_env_file(path: Path = ENV_FILE) -> None:
+    if not path.exists():
+        return
+
+    for raw_line in path.read_text(encoding="utf-8").splitlines():
+        line = raw_line.strip()
+        if not line or line.startswith("#") or "=" not in line:
+            continue
+
+        key, value = line.split("=", 1)
+        key = key.strip()
+        value = value.strip()
+
+        if not key or key in os.environ:
+            continue
+
+        if len(value) >= 2 and value[0] == value[-1] and value[0] in {'"', "'"}:
+            value = value[1:-1]
+
+        os.environ[key] = value
+
+
+def get_bot_token() -> str:
+    load_env_file()
+    token = os.getenv(TOKEN_ENV_NAME)
+    if not token:
+        raise RuntimeError(f"Set {TOKEN_ENV_NAME} in environment or .env file.")
+    return token
+
+
+def get_db_path() -> Path:
+    load_env_file()
+    raw_path = os.getenv(DB_ENV_NAME)
+    if not raw_path:
+        return DEFAULT_DB_FILE
+
+    path = Path(raw_path).expanduser()
+    if path.is_absolute():
+        return path
+    return PROJECT_ROOT / path
+
+
+def get_assistant_mode() -> str:
+    load_env_file()
+    mode = os.getenv(MODE_ENV_NAME, DEFAULT_MODE).strip().lower()
+    if mode not in {"local", "yandex"}:
+        raise RuntimeError(
+            f"{MODE_ENV_NAME} must be either 'local' or 'yandex', got {mode!r}."
+        )
+    return mode
+
+
+def get_ollama_base_url() -> str:
+    load_env_file()
+    return os.getenv(OLLAMA_BASE_URL_ENV_NAME, DEFAULT_OLLAMA_BASE_URL).rstrip("/")
+
+
+def get_default_ollama_model() -> str:
+    load_env_file()
+    return os.getenv(OLLAMA_MODEL_ENV_NAME, DEFAULT_OLLAMA_MODEL)
+
+
+def get_yandex_cloud_folder() -> str:
+    load_env_file()
+    folder = os.getenv(YANDEX_CLOUD_FOLDER_ENV_NAME, "").strip()
+    if not folder:
+        raise RuntimeError(
+            f"Set {YANDEX_CLOUD_FOLDER_ENV_NAME} in environment or .env file."
+        )
+    return folder.strip("/")
+
+
+def get_yandex_cloud_model() -> str:
+    load_env_file()
+    model = os.getenv(
+        YANDEX_CLOUD_MODEL_ENV_NAME,
+        DEFAULT_YANDEX_CLOUD_MODEL,
+    ).strip()
+    return model.strip("/") or DEFAULT_YANDEX_CLOUD_MODEL
+
+
+def get_default_yandex_model() -> str:
+    return f"gpt://{get_yandex_cloud_folder()}/{get_yandex_cloud_model()}"
+
+
+def get_default_model(provider: str) -> str:
+    if provider == "local":
+        return get_default_ollama_model()
+    if provider == "yandex":
+        return get_default_yandex_model()
+    raise ValueError(f"Unknown AI provider: {provider}")
+
+
+def get_yandex_stt_model() -> str:
+    load_env_file()
+    return os.getenv(YANDEX_STT_MODEL_ENV_NAME, DEFAULT_YANDEX_STT_MODEL).strip()
+
+
+def get_yandex_stt_language() -> str:
+    load_env_file()
+    return os.getenv(
+        YANDEX_STT_LANGUAGE_ENV_NAME,
+        DEFAULT_YANDEX_STT_LANGUAGE,
+    ).strip()
+
+
+def get_local_timezone() -> ZoneInfo:
+    load_env_file()
+    timezone_name = os.getenv(TIMEZONE_ENV_NAME, DEFAULT_TIMEZONE)
+    try:
+        return ZoneInfo(timezone_name)
+    except ZoneInfoNotFoundError:
+        logger.warning("Unknown timezone %s, falling back to UTC", timezone_name)
+        return ZoneInfo("UTC")
+
+
+def get_whisper_model() -> str:
+    load_env_file()
+    return os.getenv(WHISPER_MODEL_ENV_NAME, DEFAULT_WHISPER_MODEL).strip()
+
+
+def get_whisper_device() -> str:
+    load_env_file()
+    return os.getenv(WHISPER_DEVICE_ENV_NAME, DEFAULT_WHISPER_DEVICE).strip()
+
+
+def get_whisper_compute_type() -> str:
+    load_env_file()
+    return os.getenv(
+        WHISPER_COMPUTE_TYPE_ENV_NAME,
+        DEFAULT_WHISPER_COMPUTE_TYPE,
+    ).strip()
+
+
+def get_whisper_language() -> str | None:
+    load_env_file()
+    language = os.getenv(
+        WHISPER_LANGUAGE_ENV_NAME,
+        DEFAULT_WHISPER_LANGUAGE,
+    ).strip()
+    return None if language.lower() == "auto" else language or None
+
+
+def get_voice_max_duration_seconds() -> int:
+    load_env_file()
+    raw_value = os.getenv(
+        VOICE_MAX_DURATION_ENV_NAME,
+        str(DEFAULT_VOICE_MAX_DURATION_SECONDS),
+    )
+    try:
+        value = int(raw_value)
+    except ValueError:
+        value = 0
+    if value <= 0:
+        logger.warning(
+            "%s must be a positive integer, using %s",
+            VOICE_MAX_DURATION_ENV_NAME,
+            DEFAULT_VOICE_MAX_DURATION_SECONDS,
+        )
+        return DEFAULT_VOICE_MAX_DURATION_SECONDS
+    return value

assistant_bot/handlers.py

@@ -0,0 +1,498 @@
+import asyncio
+import logging
+import os
+import tempfile
+from pathlib import Path
+from typing import Any
+
+from telegram import Update
+from telegram.constants import ChatAction
+from telegram.error import TelegramError
+from telegram.ext import ContextTypes
+
+from .ai import AIClientError
+from .agent import run_agent_prompt
+from .reminders import parse_reminder
+from .telegram_utils import (
+    build_inline_results,
+    command_text,
+    get_ai_client,
+    get_speech_recognizer,
+    get_storage,
+    get_tz,
+    get_voice_max_duration,
+    parse_positive_int,
+    reply_long,
+    require_user_id,
+)
+from .speech import SpeechRecognitionError
+from .time_utils import format_local_dt
+
+
+logger = logging.getLogger(__name__)
+
+
+def format_numbered_rows(
+    rows: list[dict[str, Any]],
+    formatter,
+    empty_text: str,
+) -> str:
+    if not rows:
+        return empty_text
+    return "\n".join(formatter(row) for row in rows)
+
+
+async def start(update: Update, _context: ContextTypes.DEFAULT_TYPE) -> None:
+    if update.message:
+        await update.message.reply_text(
+            "Я готов как персональный ассистент.\n"
+            "Пиши обычным текстом или отправляй голосовые сообщения: "
+            "'запомни, что...', 'напомни завтра в 10:00...', "
+            "'сохрани заметку...', 'покажи мои статусы'.\n"
+            "Я помню последние реплики диалога; команда /new очищает текущий контекст.\n"
+            "Я сам решу, когда нужно вызвать внутренний tool."
+        )
+
+
+async def help_command(update: Update, _context: ContextTypes.DEFAULT_TYPE) -> None:
+    if update.message:
+        await update.message.reply_text(
+            "Основной режим - свободный диалог текстом или голосовыми сообщениями.\n\n"
+            "Примеры:\n"
+            "Запомни, что я предпочитаю короткие ответы.\n"
+            "Сохрани заметку: идея для проекта.\n"
+            "Напомни через 30 минут проверить сборку.\n"
+            "Отслеживай паспорт, статус: жду ответа.\n"
+            "Покажи активные напоминания.\n\n"
+            "Вся переписка сохраняется. /history тема ищет старое обсуждение, "
+            "/new начинает новый контекст без удаления архива.\n\n"
+            "Служебные команды оставлены для настройки и отладки: /models, /model, /ask.\n"
+            "Режим выбирается через ASSISTANT_MODE=local или ASSISTANT_MODE=yandex."
+        )
+
+
+async def new_conversation_command(update: Update, context: ContextTypes.DEFAULT_TYPE) -> None:
+    if not update.message:
+        return
+    user_id = require_user_id(update)
+    if user_id is None:
+        await update.message.reply_text("Не могу определить пользователя.")
+        return
+    get_storage(context).start_new_conversation(user_id, int(update.message.chat_id))
+    await update.message.reply_text("Начинаем новую тему. Предыдущая переписка сохранена в архиве.")
+
+
+async def history_command(update: Update, context: ContextTypes.DEFAULT_TYPE) -> None:
+    if not update.message:
+        return
+    user_id = require_user_id(update)
+    if user_id is None:
+        await update.message.reply_text("Не могу определить пользователя.")
+        return
+
+    query = command_text(context)
+    if not query:
+        await update.message.reply_text("Укажи тему или ключевые слова: /history отпуск")
+        return
+
+    rows = get_storage(context).search_conversation_messages(
+        user_id,
+        int(update.message.chat_id),
+        query,
+        limit=10,
+    )
+    if not rows:
+        await update.message.reply_text("В сохраненной переписке ничего не найдено.")
+        return
+
+    role_names = {"user": "Вы", "assistant": "Бот"}
+    tz = get_tz(context)
+    text = "Найденные сообщения:\n\n" + "\n\n".join(
+        f"{role_names.get(str(row['role']), row['role'])} · "
+        f"{format_local_dt(row['created_at'], tz)}\n{row['content']}"
+        for row in rows
+    )
+    await reply_long(update.message, text)
+
+
+async def ask_command(update: Update, context: ContextTypes.DEFAULT_TYPE) -> None:
+    prompt = command_text(context)
+    if not prompt:
+        if update.message:
+            await update.message.reply_text("Напиши текст после /ask или просто отправь сообщение в личный чат.")
+        return
+    await run_agent_prompt(update, context, prompt)
+
+
+async def private_text(update: Update, context: ContextTypes.DEFAULT_TYPE) -> None:
+    if not update.message or not update.message.text:
+        return
+    await run_agent_prompt(update, context, update.message.text.strip())
+
+
+async def private_voice(update: Update, context: ContextTypes.DEFAULT_TYPE) -> None:
+    if not update.message or not update.message.voice:
+        return
+
+    voice = update.message.voice
+    max_duration = get_voice_max_duration(context)
+    if voice.duration > max_duration:
+        await update.message.reply_text(
+            f"Голосовое сообщение слишком длинное. Максимум: {max_duration} сек."
+        )
+        return
+
+    await context.bot.send_chat_action(
+        chat_id=int(update.message.chat_id),
+        action=ChatAction.TYPING,
+    )
+    status_message = await update.message.reply_text("Распознаю голосовое сообщение...")
+    temp_path: Path | None = None
+
+    try:
+        telegram_file = await context.bot.get_file(voice.file_id)
+        file_descriptor, raw_path = tempfile.mkstemp(suffix=".ogg")
+        os.close(file_descriptor)
+        temp_path = Path(raw_path)
+        await telegram_file.download_to_drive(custom_path=temp_path)
+        transcript = await asyncio.to_thread(
+            get_speech_recognizer(context).transcribe,
+            temp_path,
+        )
+    except (SpeechRecognitionError, TelegramError):
+        logger.exception("Failed to transcribe Telegram voice message")
+        await status_message.edit_text(
+            "Не получилось распознать голосовое сообщение. Попробуй еще раз позже."
+        )
+        return
+    except Exception:
+        logger.exception("Unexpected error while processing Telegram voice message")
+        await status_message.edit_text(
+            "Произошла внутренняя ошибка при обработке голосового сообщения."
+        )
+        return
+    finally:
+        if temp_path is not None:
+            try:
+                temp_path.unlink(missing_ok=True)
+            except OSError:
+                logger.warning("Could not delete temporary voice file %s", temp_path)
+
+    if not transcript:
+        await status_message.edit_text("Не удалось расслышать речь в сообщении.")
+        return
+
+    preview = transcript if len(transcript) <= 500 else f"{transcript[:497]}..."
+    await status_message.edit_text(f"Распознано: {preview}")
+    await run_agent_prompt(update, context, transcript)
+
+
+async def models_command(update: Update, context: ContextTypes.DEFAULT_TYPE) -> None:
+    if not update.message:
+        return
+
+    ai_client = get_ai_client(context)
+    try:
+        models = await ai_client.list_models()
+    except AIClientError as exc:
+        await update.message.reply_text(
+            f"Не получилось получить список моделей {ai_client.display_name}.\n{exc}"
+        )
+        return
+
+    if not models:
+        await update.message.reply_text(
+            f"{ai_client.display_name} доступна, но моделей не найдено."
+        )
+        return
+
+    await update.message.reply_text(
+        f"Модели {ai_client.display_name}:\n"
+        + "\n".join(f"- {model}" for model in models)
+    )
+
+
+async def model_command(update: Update, context: ContextTypes.DEFAULT_TYPE) -> None:
+    if not update.message:
+        return
+
+    user_id = require_user_id(update)
+    if user_id is None:
+        await update.message.reply_text("Не могу определить пользователя.")
+        return
+
+    storage = get_storage(context)
+    ai_client = get_ai_client(context)
+    model = command_text(context)
+    if not model:
+        await update.message.reply_text(
+            f"Текущая модель {ai_client.display_name}: "
+            f"{ai_client.normalize_model(storage.get_user_model(user_id, ai_client.provider))}"
+        )
+        return
+
+    model = ai_client.normalize_model(model)
+    storage.set_user_model(user_id, model, ai_client.provider)
+    await update.message.reply_text(
+        f"Модель {ai_client.display_name} сохранена: {model}"
+    )
+
+
+async def remember_command(update: Update, context: ContextTypes.DEFAULT_TYPE) -> None:
+    if not update.message:
+        return
+
+    user_id = require_user_id(update)
+    text = command_text(context)
+    if user_id is None:
+        await update.message.reply_text("Не могу определить пользователя.")
+        return
+    if not text:
+        await update.message.reply_text("Напиши факт после команды: /remember я предпочитаю короткие ответы")
+        return
+
+    memory_id = get_storage(context).add_memory(user_id, text)
+    await update.message.reply_text(f"Запомнил. ID памяти: {memory_id}")
+
+
+async def memory_command(update: Update, context: ContextTypes.DEFAULT_TYPE) -> None:
+    if not update.message:
+        return
+
+    user_id = require_user_id(update)
+    if user_id is None:
+        await update.message.reply_text("Не могу определить пользователя.")
+        return
+
+    rows = get_storage(context).list_memories(user_id)
+    text = format_numbered_rows(
+        rows,
+        lambda row: f"#{row['id']} - {row['text']}",
+        "Долговременная память пока пустая.",
+    )
+    await reply_long(update.message, text)
+
+
+async def forget_memory_command(update: Update, context: ContextTypes.DEFAULT_TYPE) -> None:
+    if not update.message:
+        return
+
+    user_id = require_user_id(update)
+    memory_id = parse_positive_int(command_text(context))
+    if user_id is None:
+        await update.message.reply_text("Не могу определить пользователя.")
+        return
+    if memory_id is None:
+        await update.message.reply_text("Укажи ID: /forget_memory 3")
+        return
+
+    deleted = get_storage(context).delete_memory(user_id, memory_id)
+    await update.message.reply_text("Удалил." if deleted else "Не нашел такую запись памяти.")
+
+
+async def note_command(update: Update, context: ContextTypes.DEFAULT_TYPE) -> None:
+    if not update.message:
+        return
+
+    user_id = require_user_id(update)
+    text = command_text(context)
+    if user_id is None:
+        await update.message.reply_text("Не могу определить пользователя.")
+        return
+    if not text:
+        await update.message.reply_text("Напиши текст заметки: /note идея для проекта")
+        return
+
+    note_id = get_storage(context).add_note(user_id, text)
+    await update.message.reply_text(f"Заметка сохранена. ID: {note_id}")
+
+
+async def notes_command(update: Update, context: ContextTypes.DEFAULT_TYPE) -> None:
+    if not update.message:
+        return
+
+    user_id = require_user_id(update)
+    if user_id is None:
+        await update.message.reply_text("Не могу определить пользователя.")
+        return
+
+    rows = get_storage(context).list_notes(user_id)
+    text = format_numbered_rows(
+        rows,
+        lambda row: f"#{row['id']} - {row['text']}",
+        "Заметок пока нет.",
+    )
+    await reply_long(update.message, text)
+
+
+async def forget_note_command(update: Update, context: ContextTypes.DEFAULT_TYPE) -> None:
+    if not update.message:
+        return
+
+    user_id = require_user_id(update)
+    note_id = parse_positive_int(command_text(context))
+    if user_id is None:
+        await update.message.reply_text("Не могу определить пользователя.")
+        return
+    if note_id is None:
+        await update.message.reply_text("Укажи ID: /forget 3")
+        return
+
+    deleted = get_storage(context).delete_note(user_id, note_id)
+    await update.message.reply_text("Удалил." if deleted else "Не нашел такую заметку.")
+
+
+async def remind_command(update: Update, context: ContextTypes.DEFAULT_TYPE) -> None:
+    if not update.message or not update.effective_chat:
+        return
+
+    user_id = require_user_id(update)
+    if user_id is None:
+        await update.message.reply_text("Не могу определить пользователя.")
+        return
+
+    parsed = parse_reminder(command_text(context), get_tz(context))
+    if not parsed:
+        await update.message.reply_text(
+            "Формат напоминания:\n"
+            "/remind 30m позвонить\n"
+            "/remind через 2 часа проверить статус\n"
+            "/remind 18:30 отправить отчет\n"
+            "/remind 2026-07-16 18:30 отправить отчет"
+        )
+        return
+
+    reminder_id = get_storage(context).add_reminder(
+        user_id=user_id,
+        chat_id=int(update.effective_chat.id),
+        text=parsed.text,
+        remind_at_utc=parsed.remind_at_utc,
+    )
+    await update.message.reply_text(
+        f"Напоминание #{reminder_id} поставлено на "
+        f"{format_local_dt(parsed.remind_at_utc, get_tz(context))}."
+    )
+
+
+async def reminders_command(update: Update, context: ContextTypes.DEFAULT_TYPE) -> None:
+    if not update.message:
+        return
+
+    user_id = require_user_id(update)
+    if user_id is None:
+        await update.message.reply_text("Не могу определить пользователя.")
+        return
+
+    tz = get_tz(context)
+    rows = get_storage(context).list_reminders(user_id)
+    text = format_numbered_rows(
+        rows,
+        lambda row: f"#{row['id']} - {format_local_dt(row['remind_at'], tz)} - {row['text']}",
+        "Активных напоминаний нет.",
+    )
+    await reply_long(update.message, text)
+
+
+async def cancel_reminder_command(update: Update, context: ContextTypes.DEFAULT_TYPE) -> None:
+    if not update.message:
+        return
+
+    user_id = require_user_id(update)
+    reminder_id = parse_positive_int(command_text(context))
+    if user_id is None:
+        await update.message.reply_text("Не могу определить пользователя.")
+        return
+    if reminder_id is None:
+        await update.message.reply_text("Укажи ID: /cancelreminder 3")
+        return
+
+    cancelled = get_storage(context).cancel_reminder(user_id, reminder_id)
+    await update.message.reply_text("Отменил." if cancelled else "Не нашел активное напоминание.")
+
+
+def split_tracking_text(raw: str) -> tuple[str, str]:
+    if "|" not in raw:
+        return raw.strip(), "open"
+    title, status = raw.split("|", 1)
+    return title.strip(), status.strip() or "open"
+
+
+async def track_command(update: Update, context: ContextTypes.DEFAULT_TYPE) -> None:
+    if not update.message:
+        return
+
+    user_id = require_user_id(update)
+    raw_text = command_text(context)
+    if user_id is None:
+        await update.message.reply_text("Не могу определить пользователя.")
+        return
+    if not raw_text:
+        await update.message.reply_text("Формат: /track паспорт | жду ответа")
+        return
+
+    title, status = split_tracking_text(raw_text)
+    if not title:
+        await update.message.reply_text("Название не должно быть пустым.")
+        return
+
+    item_id = get_storage(context).add_tracked_item(user_id, title, status)
+    await update.message.reply_text(f"Добавил отслеживание #{item_id}: {title} - {status}")
+
+
+async def status_command(update: Update, context: ContextTypes.DEFAULT_TYPE) -> None:
+    if not update.message:
+        return
+
+    user_id = require_user_id(update)
+    if user_id is None:
+        await update.message.reply_text("Не могу определить пользователя.")
+        return
+
+    raw_text = command_text(context)
+    storage = get_storage(context)
+    if not raw_text:
+        rows = storage.list_tracked_items(user_id)
+        text = format_numbered_rows(
+            rows,
+            lambda row: f"#{row['id']} - {row['title']}: {row['status']}",
+            "Нет отслеживаемых статусов.",
+        )
+        await reply_long(update.message, text)
+        return
+
+    parts = raw_text.split(maxsplit=1)
+    item_id = parse_positive_int(parts[0])
+    new_status = parts[1].strip() if len(parts) > 1 else ""
+    if item_id is None or not new_status:
+        await update.message.reply_text("Формат: /status 3 новый статус")
+        return
+
+    updated = storage.set_tracked_status(user_id, item_id, new_status)
+    await update.message.reply_text("Статус обновлен." if updated else "Не нашел такой объект.")
+
+
+async def untrack_command(update: Update, context: ContextTypes.DEFAULT_TYPE) -> None:
+    if not update.message:
+        return
+
+    user_id = require_user_id(update)
+    item_id = parse_positive_int(command_text(context))
+    if user_id is None:
+        await update.message.reply_text("Не могу определить пользователя.")
+        return
+    if item_id is None:
+        await update.message.reply_text("Укажи ID: /untrack 3")
+        return
+
+    deleted = get_storage(context).delete_tracked_item(user_id, item_id)
+    await update.message.reply_text("Удалил." if deleted else "Не нашел такой объект.")
+
+
+async def inline_query(update: Update, _context: ContextTypes.DEFAULT_TYPE) -> None:
+    if not update.inline_query or not update.inline_query.query:
+        return
+
+    try:
+        await update.inline_query.answer(build_inline_results(update.inline_query.query))
+    except Exception:
+        logger.exception("Failed to answer inline query")

assistant_bot/jobs.py

@@ -0,0 +1,42 @@
+import asyncio
+import logging
+from zoneinfo import ZoneInfo
+
+from telegram.ext import Application
+
+from .config import REMINDER_POLL_SECONDS
+from .storage import AssistantStorage
+from .time_utils import format_local_dt, utc_now
+
+
+logger = logging.getLogger(__name__)
+
+
+async def reminder_loop(application: Application) -> None:
+    storage: AssistantStorage = application.bot_data["storage"]
+    tz: ZoneInfo = application.bot_data["timezone"]
+
+    while True:
+        try:
+            due_reminders = storage.due_reminders(utc_now())
+            for reminder in due_reminders:
+                await application.bot.send_message(
+                    chat_id=reminder["chat_id"],
+                    text=(
+                        f"Напоминание #{reminder['id']} "
+                        f"({format_local_dt(reminder['remind_at'], tz)}):\n"
+                        f"{reminder['text']}"
+                    ),
+                )
+                storage.mark_reminder_sent(int(reminder["id"]))
+        except asyncio.CancelledError:
+            raise
+        except Exception:
+            logger.exception("Reminder loop failed")
+
+        await asyncio.sleep(REMINDER_POLL_SECONDS)
+
+
+async def post_init(application: Application) -> None:
+    application.create_task(reminder_loop(application))
+

assistant_bot/models.py

@@ -0,0 +1,16 @@
+from dataclasses import dataclass
+from datetime import datetime
+from typing import Any
+
+
+@dataclass(frozen=True)
+class ReminderParseResult:
+    remind_at_utc: datetime
+    text: str
+
+
+@dataclass(frozen=True)
+class AgentDecision:
+    final: str | None
+    tool_calls: list[dict[str, Any]]
+    reset_context: bool = False

assistant_bot/ollama.py

@@ -0,0 +1,99 @@
+import asyncio
+import json
+import urllib.error
+import urllib.request
+from typing import Any
+
+from .ai import AIClientError
+
+
+class OllamaError(AIClientError):
+    pass
+
+
+class OllamaClient:
+    provider = "local"
+    display_name = "Ollama"
+
+    def __init__(self, base_url: str) -> None:
+        self.base_url = base_url.rstrip("/")
+
+    def normalize_model(self, model: str) -> str:
+        return model.strip()
+
+    async def list_models(self) -> list[str]:
+        return await asyncio.to_thread(self._list_models)
+
+    async def chat(
+        self,
+        model: str,
+        messages: list[dict[str, str]],
+        json_mode: bool = False,
+    ) -> str:
+        return await asyncio.to_thread(self._chat, model, messages, json_mode)
+
+    def _list_models(self) -> list[str]:
+        data = self._request_json("GET", "/api/tags", None, timeout=15)
+        models = data.get("models", [])
+        return sorted(model["name"] for model in models if model.get("name"))
+
+    def _chat(
+        self,
+        model: str,
+        messages: list[dict[str, str]],
+        json_mode: bool,
+    ) -> str:
+        payload = {
+            "model": model,
+            "messages": messages,
+            "stream": False,
+        }
+        if json_mode:
+            payload["format"] = "json"
+
+        data = self._request_json("POST", "/api/chat", payload, timeout=180)
+        content = data.get("message", {}).get("content")
+        if isinstance(content, str) and content.strip():
+            return content.strip()
+
+        fallback = data.get("response")
+        if isinstance(fallback, str) and fallback.strip():
+            return fallback.strip()
+
+        raise OllamaError("Ollama returned an empty response.")
+
+    def _request_json(
+        self,
+        method: str,
+        path: str,
+        payload: dict[str, Any] | None,
+        timeout: int,
+    ) -> dict[str, Any]:
+        url = f"{self.base_url}{path}"
+        data = json.dumps(payload).encode("utf-8") if payload is not None else None
+        request = urllib.request.Request(
+            url,
+            data=data,
+            method=method,
+            headers={"Content-Type": "application/json"},
+        )
+
+        try:
+            with urllib.request.urlopen(request, timeout=timeout) as response:
+                raw = response.read().decode("utf-8")
+        except urllib.error.HTTPError as exc:
+            body = exc.read().decode("utf-8", errors="replace")[:400]
+            raise OllamaError(f"Ollama HTTP {exc.code}: {body}") from exc
+        except urllib.error.URLError as exc:
+            raise OllamaError(f"Ollama is unavailable at {self.base_url}: {exc.reason}") from exc
+        except TimeoutError as exc:
+            raise OllamaError("Ollama request timed out.") from exc
+
+        try:
+            parsed = json.loads(raw)
+        except json.JSONDecodeError as exc:
+            raise OllamaError("Ollama returned invalid JSON.") from exc
+
+        if not isinstance(parsed, dict):
+            raise OllamaError("Ollama returned an unexpected response.")
+        return parsed

assistant_bot/prompts.py

@@ -0,0 +1,7 @@
+ASSISTANT_SYSTEM_PROMPT = (
+    "Ты персональный AI ассистент в Telegram. Отвечай кратко, по делу и на русском, "
+    "если пользователь не попросил другой язык. Используй долговременный контекст "
+    "только как вспомогательную информацию, не выдумывай факты и явно говори, "
+    "когда данных недостаточно."
+)
+

assistant_bot/reminders.py

@@ -0,0 +1,64 @@
+import re
+from datetime import datetime, timedelta, timezone
+from zoneinfo import ZoneInfo
+
+from .models import ReminderParseResult
+from .time_utils import utc_now
+
+
+def unit_to_timedelta(amount: int, unit: str) -> timedelta | None:
+    normalized = unit.lower().strip(".")
+    if normalized in {"s", "sec", "secs", "second", "seconds", "с", "сек", "секунд"}:
+        return timedelta(seconds=amount)
+    if normalized in {"m", "min", "mins", "minute", "minutes", "м"} or normalized.startswith("мин"):
+        return timedelta(minutes=amount)
+    if normalized in {"h", "hr", "hrs", "hour", "hours", "ч"} or normalized.startswith("час"):
+        return timedelta(hours=amount)
+    if normalized in {"d", "day", "days", "д"} or normalized.startswith(("дн", "ден")):
+        return timedelta(days=amount)
+    return None
+
+
+def parse_reminder(raw_text: str, tz: ZoneInfo) -> ReminderParseResult | None:
+    text = raw_text.strip()
+    if not text:
+        return None
+
+    relative_match = re.match(r"^(?:через\s+)?(\d+)\s*([a-zа-я.]+)\s+(.+)$", text, re.IGNORECASE)
+    if relative_match:
+        amount = int(relative_match.group(1))
+        delta = unit_to_timedelta(amount, relative_match.group(2))
+        reminder_text = relative_match.group(3).strip()
+        if delta and reminder_text:
+            return ReminderParseResult(utc_now() + delta, reminder_text)
+
+    now_local = datetime.now(tz)
+    absolute_patterns = (
+        (r"^(\d{4}-\d{2}-\d{2})\s+(\d{1,2}:\d{2})\s+(.+)$", "%Y-%m-%d %H:%M"),
+        (r"^(\d{1,2}\.\d{1,2}\.\d{4})\s+(\d{1,2}:\d{2})\s+(.+)$", "%d.%m.%Y %H:%M"),
+    )
+
+    for pattern, date_format in absolute_patterns:
+        match = re.match(pattern, text)
+        if not match:
+            continue
+
+        naive = datetime.strptime(f"{match.group(1)} {match.group(2)}", date_format)
+        remind_at = naive.replace(tzinfo=tz)
+        reminder_text = match.group(3).strip()
+        if reminder_text:
+            return ReminderParseResult(remind_at.astimezone(timezone.utc), reminder_text)
+
+    time_match = re.match(r"^(\d{1,2}:\d{2})\s+(.+)$", text)
+    if time_match:
+        hour, minute = (int(part) for part in time_match.group(1).split(":", 1))
+        if 0 <= hour <= 23 and 0 <= minute <= 59:
+            remind_at = now_local.replace(hour=hour, minute=minute, second=0, microsecond=0)
+            if remind_at <= now_local:
+                remind_at += timedelta(days=1)
+            reminder_text = time_match.group(2).strip()
+            if reminder_text:
+                return ReminderParseResult(remind_at.astimezone(timezone.utc), reminder_text)
+
+    return None
+

assistant_bot/speech.py

@@ -0,0 +1,85 @@
+import sys
+import threading
+import warnings
+from pathlib import Path
+from typing import Any, Callable, Protocol
+
+
+class SpeechRecognitionError(RuntimeError):
+    """Raised when a voice message cannot be transcribed."""
+
+
+class SpeechTranscriber(Protocol):
+    def transcribe(self, audio_path: str | Path) -> str:
+        ...
+
+
+def load_whisper_model_class() -> Any:
+    # CTranslate2 imports torch for optional model converters. Speech inference
+    # does not need it, so do not let an unrelated torch installation prevent
+    # faster-whisper from loading.
+    missing = object()
+    loaded_torch = sys.modules.get("torch", missing)
+    if loaded_torch is missing:
+        sys.modules["torch"] = None
+    try:
+        with warnings.catch_warnings():
+            warnings.filterwarnings(
+                "ignore",
+                message="pkg_resources is deprecated as an API.*",
+                category=UserWarning,
+            )
+            from faster_whisper import WhisperModel
+    finally:
+        if loaded_torch is missing:
+            sys.modules.pop("torch", None)
+    return WhisperModel
+
+
+class SpeechRecognizer:
+    def __init__(
+        self,
+        model_name: str,
+        device: str,
+        compute_type: str,
+        language: str | None,
+        model_factory: Callable[..., Any] | None = None,
+    ) -> None:
+        self.model_name = model_name
+        self.device = device
+        self.compute_type = compute_type
+        self.language = language
+        self._model_factory = model_factory
+        self._model: Any | None = None
+        self._lock = threading.Lock()
+
+    def _create_model(self) -> Any:
+        factory = self._model_factory
+        if factory is None:
+            factory = load_whisper_model_class()
+        return factory(
+            self.model_name,
+            device=self.device,
+            compute_type=self.compute_type,
+        )
+
+    def transcribe(self, audio_path: str | Path) -> str:
+        try:
+            with self._lock:
+                if self._model is None:
+                    self._model = self._create_model()
+                segments, _info = self._model.transcribe(
+                    str(audio_path),
+                    language=self.language,
+                    beam_size=5,
+                    vad_filter=True,
+                )
+                parts = [
+                    segment.text.strip()
+                    for segment in segments
+                    if segment.text.strip()
+                ]
+        except Exception as exc:
+            raise SpeechRecognitionError("Voice transcription failed") from exc
+
+        return " ".join(parts)

assistant_bot/storage.py

@@ -0,0 +1,490 @@
+import re
+import sqlite3
+from contextlib import contextmanager
+from datetime import datetime
+from pathlib import Path
+from typing import Any
+
+from .config import get_default_model
+from .time_utils import to_utc_iso, utc_now
+
+
+class AssistantStorage:
+    def __init__(self, path: Path) -> None:
+        self.path = path
+        self.path.parent.mkdir(parents=True, exist_ok=True)
+        self._init_db()
+
+    def _connect(self) -> sqlite3.Connection:
+        connection = sqlite3.connect(self.path)
+        connection.row_factory = sqlite3.Row
+        return connection
+
+    @contextmanager
+    def _connection(self):
+        connection = self._connect()
+        try:
+            yield connection
+            connection.commit()
+        except Exception:
+            connection.rollback()
+            raise
+        finally:
+            connection.close()
+
+    def _init_db(self) -> None:
+        with self._connection() as connection:
+            connection.execute("PRAGMA journal_mode=WAL")
+            connection.executescript(
+                """
+                CREATE TABLE IF NOT EXISTS user_settings (
+                    user_id INTEGER PRIMARY KEY,
+                    ollama_model TEXT,
+                    yandex_model TEXT,
+                    updated_at TEXT NOT NULL
+                );
+
+                CREATE TABLE IF NOT EXISTS memories (
+                    id INTEGER PRIMARY KEY AUTOINCREMENT,
+                    user_id INTEGER NOT NULL,
+                    text TEXT NOT NULL,
+                    created_at TEXT NOT NULL
+                );
+
+                CREATE TABLE IF NOT EXISTS notes (
+                    id INTEGER PRIMARY KEY AUTOINCREMENT,
+                    user_id INTEGER NOT NULL,
+                    text TEXT NOT NULL,
+                    created_at TEXT NOT NULL
+                );
+
+                CREATE TABLE IF NOT EXISTS reminders (
+                    id INTEGER PRIMARY KEY AUTOINCREMENT,
+                    user_id INTEGER NOT NULL,
+                    chat_id INTEGER NOT NULL,
+                    text TEXT NOT NULL,
+                    remind_at TEXT NOT NULL,
+                    status TEXT NOT NULL DEFAULT 'pending',
+                    created_at TEXT NOT NULL,
+                    sent_at TEXT
+                );
+
+                CREATE TABLE IF NOT EXISTS tracked_items (
+                    id INTEGER PRIMARY KEY AUTOINCREMENT,
+                    user_id INTEGER NOT NULL,
+                    title TEXT NOT NULL,
+                    status TEXT NOT NULL,
+                    created_at TEXT NOT NULL,
+                    updated_at TEXT NOT NULL
+                );
+
+                CREATE TABLE IF NOT EXISTS conversation_messages (
+                    id INTEGER PRIMARY KEY AUTOINCREMENT,
+                    user_id INTEGER NOT NULL,
+                    chat_id INTEGER NOT NULL,
+                    role TEXT NOT NULL CHECK(role IN ('user', 'assistant')),
+                    content TEXT NOT NULL,
+                    created_at TEXT NOT NULL
+                );
+
+                CREATE TABLE IF NOT EXISTS conversation_contexts (
+                    user_id INTEGER NOT NULL,
+                    chat_id INTEGER NOT NULL,
+                    started_after_id INTEGER NOT NULL DEFAULT 0,
+                    updated_at TEXT NOT NULL,
+                    PRIMARY KEY (user_id, chat_id)
+                );
+
+                CREATE INDEX IF NOT EXISTS idx_memories_user_created
+                    ON memories(user_id, created_at DESC);
+                CREATE INDEX IF NOT EXISTS idx_notes_user_created
+                    ON notes(user_id, created_at DESC);
+                CREATE INDEX IF NOT EXISTS idx_reminders_due
+                    ON reminders(status, remind_at);
+                CREATE INDEX IF NOT EXISTS idx_tracked_user_updated
+                    ON tracked_items(user_id, updated_at DESC);
+                CREATE INDEX IF NOT EXISTS idx_conversation_chat_id
+                    ON conversation_messages(user_id, chat_id, id DESC);
+                """
+            )
+            columns = {
+                str(row["name"])
+                for row in connection.execute("PRAGMA table_info(user_settings)")
+            }
+            if "yandex_model" not in columns:
+                connection.execute(
+                    "ALTER TABLE user_settings ADD COLUMN yandex_model TEXT"
+                )
+
+    def add_conversation_exchange(
+        self,
+        user_id: int,
+        chat_id: int,
+        user_content: str,
+        assistant_content: str,
+    ) -> None:
+        now = to_utc_iso(utc_now())
+        with self._connection() as connection:
+            connection.executemany(
+                """
+                INSERT INTO conversation_messages(user_id, chat_id, role, content, created_at)
+                VALUES (?, ?, ?, ?, ?)
+                """,
+                (
+                    (user_id, chat_id, "user", user_content, now),
+                    (user_id, chat_id, "assistant", assistant_content, now),
+                ),
+            )
+
+    def list_conversation_messages(
+        self,
+        user_id: int,
+        chat_id: int,
+        limit: int = 12,
+    ) -> list[dict[str, Any]]:
+        with self._connection() as connection:
+            context_row = connection.execute(
+                """
+                SELECT started_after_id
+                FROM conversation_contexts
+                WHERE user_id = ? AND chat_id = ?
+                """,
+                (user_id, chat_id),
+            ).fetchone()
+            started_after_id = int(context_row["started_after_id"]) if context_row else 0
+            rows = connection.execute(
+                """
+                SELECT id, role, content, created_at
+                FROM conversation_messages
+                WHERE user_id = ? AND chat_id = ? AND id > ?
+                ORDER BY id DESC
+                LIMIT ?
+                """,
+                (user_id, chat_id, started_after_id, limit),
+            ).fetchall()
+        return [dict(row) for row in reversed(rows)]
+
+    def start_new_conversation(self, user_id: int, chat_id: int) -> None:
+        now = to_utc_iso(utc_now())
+        with self._connection() as connection:
+            row = connection.execute(
+                """
+                SELECT COALESCE(MAX(id), 0) AS last_id
+                FROM conversation_messages
+                WHERE user_id = ? AND chat_id = ?
+                """,
+                (user_id, chat_id),
+            ).fetchone()
+            last_id = int(row["last_id"])
+            connection.execute(
+                """
+                INSERT INTO conversation_contexts(user_id, chat_id, started_after_id, updated_at)
+                VALUES (?, ?, ?, ?)
+                ON CONFLICT(user_id, chat_id) DO UPDATE SET
+                    started_after_id = excluded.started_after_id,
+                    updated_at = excluded.updated_at
+                """,
+                (user_id, chat_id, last_id, now),
+            )
+
+    def search_conversation_messages(
+        self,
+        user_id: int,
+        chat_id: int,
+        query: str,
+        limit: int = 10,
+    ) -> list[dict[str, Any]]:
+        normalized_query = query.casefold().strip()
+        terms = list(
+            dict.fromkeys(
+                part for part in re.findall(r"\w+", normalized_query) if len(part) > 1
+            )
+        )
+        if not terms:
+            return []
+
+        with self._connection() as connection:
+            rows = connection.execute(
+                """
+                SELECT id, role, content, created_at
+                FROM conversation_messages
+                WHERE user_id = ? AND chat_id = ?
+                ORDER BY id DESC
+                """,
+                (user_id, chat_id),
+            ).fetchall()
+
+        if not rows:
+            return []
+
+        newest_id = int(rows[0]["id"])
+        matches: list[tuple[float, dict[str, Any]]] = []
+        for row in rows:
+            content = str(row["content"]).casefold()
+            matched_terms = sum(term in content for term in terms)
+            if not matched_terms:
+                continue
+
+            coverage = matched_terms / len(terms)
+            phrase_bonus = 1.0 if len(normalized_query) > 2 and normalized_query in content else 0.0
+            age = max(0, newest_id - int(row["id"]))
+            recency_weight = 1.0 / (1.0 + age / 50.0)
+            score = (coverage + phrase_bonus) * (0.5 + 0.5 * recency_weight)
+            item = dict(row)
+            item["score"] = round(score, 4)
+            matches.append((score, item))
+
+        matches.sort(key=lambda item: (item[0], int(item[1]["id"])), reverse=True)
+        return [item for _score, item in matches[: max(1, limit)]]
+
+    def conversation_message_window(
+        self,
+        user_id: int,
+        chat_id: int,
+        message_id: int,
+        radius: int = 2,
+    ) -> list[dict[str, Any]]:
+        with self._connection() as connection:
+            before = connection.execute(
+                """
+                SELECT id, role, content, created_at
+                FROM conversation_messages
+                WHERE user_id = ? AND chat_id = ? AND id <= ?
+                ORDER BY id DESC
+                LIMIT ?
+                """,
+                (user_id, chat_id, message_id, radius + 1),
+            ).fetchall()
+            after = connection.execute(
+                """
+                SELECT id, role, content, created_at
+                FROM conversation_messages
+                WHERE user_id = ? AND chat_id = ? AND id > ?
+                ORDER BY id ASC
+                LIMIT ?
+                """,
+                (user_id, chat_id, message_id, radius),
+            ).fetchall()
+        rows = [*reversed(before), *after]
+        return [dict(row) for row in rows]
+
+    def get_user_model(self, user_id: int, provider: str = "local") -> str:
+        column = self._model_column(provider)
+        with self._connection() as connection:
+            row = connection.execute(
+                f"SELECT {column} AS model FROM user_settings WHERE user_id = ?",
+                (user_id,),
+            ).fetchone()
+        if row and row["model"]:
+            return str(row["model"])
+        return get_default_model(provider)
+
+    def set_user_model(
+        self,
+        user_id: int,
+        model: str,
+        provider: str = "local",
+    ) -> None:
+        column = self._model_column(provider)
+        now = to_utc_iso(utc_now())
+        with self._connection() as connection:
+            connection.execute(
+                f"""
+                INSERT INTO user_settings(user_id, {column}, updated_at)
+                VALUES (?, ?, ?)
+                ON CONFLICT(user_id) DO UPDATE SET
+                    {column} = excluded.{column},
+                    updated_at = excluded.updated_at
+                """,
+                (user_id, model, now),
+            )
+
+    @staticmethod
+    def _model_column(provider: str) -> str:
+        try:
+            return {
+                "local": "ollama_model",
+                "yandex": "yandex_model",
+            }[provider]
+        except KeyError as exc:
+            raise ValueError(f"Unknown AI provider: {provider}") from exc
+
+    def add_memory(self, user_id: int, text: str) -> int:
+        with self._connection() as connection:
+            cursor = connection.execute(
+                "INSERT INTO memories(user_id, text, created_at) VALUES (?, ?, ?)",
+                (user_id, text, to_utc_iso(utc_now())),
+            )
+            return int(cursor.lastrowid)
+
+    def list_memories(self, user_id: int, limit: int = 20) -> list[dict[str, Any]]:
+        with self._connection() as connection:
+            rows = connection.execute(
+                """
+                SELECT id, text, created_at
+                FROM memories
+                WHERE user_id = ?
+                ORDER BY created_at DESC, id DESC
+                LIMIT ?
+                """,
+                (user_id, limit),
+            ).fetchall()
+        return [dict(row) for row in rows]
+
+    def delete_memory(self, user_id: int, memory_id: int) -> bool:
+        with self._connection() as connection:
+            cursor = connection.execute(
+                "DELETE FROM memories WHERE user_id = ? AND id = ?",
+                (user_id, memory_id),
+            )
+            return cursor.rowcount > 0
+
+    def add_note(self, user_id: int, text: str) -> int:
+        with self._connection() as connection:
+            cursor = connection.execute(
+                "INSERT INTO notes(user_id, text, created_at) VALUES (?, ?, ?)",
+                (user_id, text, to_utc_iso(utc_now())),
+            )
+            return int(cursor.lastrowid)
+
+    def list_notes(self, user_id: int, limit: int = 20) -> list[dict[str, Any]]:
+        with self._connection() as connection:
+            rows = connection.execute(
+                """
+                SELECT id, text, created_at
+                FROM notes
+                WHERE user_id = ?
+                ORDER BY created_at DESC, id DESC
+                LIMIT ?
+                """,
+                (user_id, limit),
+            ).fetchall()
+        return [dict(row) for row in rows]
+
+    def delete_note(self, user_id: int, note_id: int) -> bool:
+        with self._connection() as connection:
+            cursor = connection.execute(
+                "DELETE FROM notes WHERE user_id = ? AND id = ?",
+                (user_id, note_id),
+            )
+            return cursor.rowcount > 0
+
+    def add_reminder(
+        self,
+        user_id: int,
+        chat_id: int,
+        text: str,
+        remind_at_utc: datetime,
+    ) -> int:
+        with self._connection() as connection:
+            cursor = connection.execute(
+                """
+                INSERT INTO reminders(user_id, chat_id, text, remind_at, created_at)
+                VALUES (?, ?, ?, ?, ?)
+                """,
+                (
+                    user_id,
+                    chat_id,
+                    text,
+                    to_utc_iso(remind_at_utc),
+                    to_utc_iso(utc_now()),
+                ),
+            )
+            return int(cursor.lastrowid)
+
+    def list_reminders(self, user_id: int, limit: int = 20) -> list[dict[str, Any]]:
+        with self._connection() as connection:
+            rows = connection.execute(
+                """
+                SELECT id, text, remind_at, status, sent_at
+                FROM reminders
+                WHERE user_id = ? AND status = 'pending'
+                ORDER BY remind_at ASC, id ASC
+                LIMIT ?
+                """,
+                (user_id, limit),
+            ).fetchall()
+        return [dict(row) for row in rows]
+
+    def cancel_reminder(self, user_id: int, reminder_id: int) -> bool:
+        with self._connection() as connection:
+            cursor = connection.execute(
+                """
+                UPDATE reminders
+                SET status = 'cancelled'
+                WHERE user_id = ? AND id = ? AND status = 'pending'
+                """,
+                (user_id, reminder_id),
+            )
+            return cursor.rowcount > 0
+
+    def due_reminders(self, now_utc: datetime, limit: int = 20) -> list[dict[str, Any]]:
+        with self._connection() as connection:
+            rows = connection.execute(
+                """
+                SELECT id, user_id, chat_id, text, remind_at
+                FROM reminders
+                WHERE status = 'pending' AND remind_at <= ?
+                ORDER BY remind_at ASC, id ASC
+                LIMIT ?
+                """,
+                (to_utc_iso(now_utc), limit),
+            ).fetchall()
+        return [dict(row) for row in rows]
+
+    def mark_reminder_sent(self, reminder_id: int) -> None:
+        with self._connection() as connection:
+            connection.execute(
+                """
+                UPDATE reminders
+                SET status = 'sent', sent_at = ?
+                WHERE id = ? AND status = 'pending'
+                """,
+                (to_utc_iso(utc_now()), reminder_id),
+            )
+
+    def add_tracked_item(self, user_id: int, title: str, status: str) -> int:
+        now = to_utc_iso(utc_now())
+        with self._connection() as connection:
+            cursor = connection.execute(
+                """
+                INSERT INTO tracked_items(user_id, title, status, created_at, updated_at)
+                VALUES (?, ?, ?, ?, ?)
+                """,
+                (user_id, title, status, now, now),
+            )
+            return int(cursor.lastrowid)
+
+    def list_tracked_items(self, user_id: int, limit: int = 30) -> list[dict[str, Any]]:
+        with self._connection() as connection:
+            rows = connection.execute(
+                """
+                SELECT id, title, status, created_at, updated_at
+                FROM tracked_items
+                WHERE user_id = ?
+                ORDER BY updated_at DESC, id DESC
+                LIMIT ?
+                """,
+                (user_id, limit),
+            ).fetchall()
+        return [dict(row) for row in rows]
+
+    def set_tracked_status(self, user_id: int, item_id: int, status: str) -> bool:
+        with self._connection() as connection:
+            cursor = connection.execute(
+                """
+                UPDATE tracked_items
+                SET status = ?, updated_at = ?
+                WHERE user_id = ? AND id = ?
+                """,
+                (status, to_utc_iso(utc_now()), user_id, item_id),
+            )
+            return cursor.rowcount > 0
+
+    def delete_tracked_item(self, user_id: int, item_id: int) -> bool:
+        with self._connection() as connection:
+            cursor = connection.execute(
+                "DELETE FROM tracked_items WHERE user_id = ? AND id = ?",
+                (user_id, item_id),
+            )
+            return cursor.rowcount > 0

assistant_bot/telegram_utils.py

@@ -0,0 +1,95 @@
+from html import escape
+from uuid import uuid4
+from zoneinfo import ZoneInfo
+
+from telegram import InlineQueryResultArticle, InputTextMessageContent, Message, Update
+from telegram.constants import ParseMode
+from telegram.ext import ContextTypes
+
+from .ai import AIClient
+from .config import HTML_FORMATS, MAX_TELEGRAM_MESSAGE_LENGTH
+from .speech import SpeechTranscriber
+from .storage import AssistantStorage
+
+
+def make_article(
+    title: str,
+    text: str,
+    parse_mode: str | None = None,
+) -> InlineQueryResultArticle:
+    return InlineQueryResultArticle(
+        id=f"inline_{uuid4()}",
+        title=title,
+        input_message_content=InputTextMessageContent(text, parse_mode=parse_mode),
+    )
+
+
+def build_inline_results(query: str) -> list[InlineQueryResultArticle]:
+    escaped_query = escape(query)
+
+    return [
+        make_article("Caps", query.upper()),
+        *[
+            make_article(title, f"<{tag}>{escaped_query}</{tag}>", ParseMode.HTML)
+            for title, tag in HTML_FORMATS
+        ],
+    ]
+
+
+def get_storage(context: ContextTypes.DEFAULT_TYPE) -> AssistantStorage:
+    return context.application.bot_data["storage"]
+
+
+def get_ai_client(context: ContextTypes.DEFAULT_TYPE) -> AIClient:
+    return context.application.bot_data["ai_client"]
+
+
+def get_tz(context: ContextTypes.DEFAULT_TYPE) -> ZoneInfo:
+    return context.application.bot_data["timezone"]
+
+
+def get_speech_recognizer(context: ContextTypes.DEFAULT_TYPE) -> SpeechTranscriber:
+    return context.application.bot_data["speech_recognizer"]
+
+
+def get_voice_max_duration(context: ContextTypes.DEFAULT_TYPE) -> int:
+    return int(context.application.bot_data["voice_max_duration"])
+
+
+def command_text(context: ContextTypes.DEFAULT_TYPE) -> str:
+    return " ".join(context.args).strip()
+
+
+def require_user_id(update: Update) -> int | None:
+    if update.effective_user:
+        return int(update.effective_user.id)
+    return None
+
+
+async def reply_long(message: Message, text: str) -> None:
+    chunks = [
+        text[index : index + MAX_TELEGRAM_MESSAGE_LENGTH]
+        for index in range(0, len(text), MAX_TELEGRAM_MESSAGE_LENGTH)
+    ] or [""]
+
+    for chunk in chunks:
+        await message.reply_text(chunk)
+
+
+async def edit_or_reply(message: Message, text: str) -> None:
+    chunks = [
+        text[index : index + MAX_TELEGRAM_MESSAGE_LENGTH]
+        for index in range(0, len(text), MAX_TELEGRAM_MESSAGE_LENGTH)
+    ] or [""]
+
+    await message.edit_text(chunks[0])
+    for chunk in chunks[1:]:
+        await message.reply_text(chunk)
+
+
+def parse_positive_int(value: str) -> int | None:
+    try:
+        parsed = int(value)
+    except ValueError:
+        return None
+    return parsed if parsed > 0 else None

assistant_bot/time_utils.py

@@ -0,0 +1,23 @@
+from datetime import datetime, timezone
+from zoneinfo import ZoneInfo
+
+
+def utc_now() -> datetime:
+    return datetime.now(timezone.utc)
+
+
+def to_utc_iso(value: datetime) -> str:
+    return value.astimezone(timezone.utc).isoformat(timespec="seconds")
+
+
+def parse_utc_iso(value: str) -> datetime:
+    parsed = datetime.fromisoformat(value)
+    if parsed.tzinfo is None:
+        return parsed.replace(tzinfo=timezone.utc)
+    return parsed.astimezone(timezone.utc)
+
+
+def format_local_dt(value: str | datetime, tz: ZoneInfo) -> str:
+    parsed = parse_utc_iso(value) if isinstance(value, str) else value
+    return parsed.astimezone(tz).strftime("%Y-%m-%d %H:%M")
+

assistant_bot/yandex_ai.py

@@ -0,0 +1,115 @@
+from pathlib import Path
+from typing import Any
+
+from .ai import AIClientError
+from .speech import SpeechRecognitionError
+
+
+class YandexAIError(AIClientError):
+    pass
+
+
+def create_async_sdk(folder_id: str) -> Any:
+    try:
+        from yandex_ai_studio_sdk import AsyncAIStudio
+    except ImportError as exc:
+        raise RuntimeError(
+            "Yandex AI Studio SDK is not installed. "
+            "Run: python -m pip install -r requirements.txt"
+        ) from exc
+    return AsyncAIStudio(folder_id=folder_id)
+
+
+def create_sync_sdk(folder_id: str) -> Any:
+    try:
+        from yandex_ai_studio_sdk import AIStudio
+    except ImportError as exc:
+        raise RuntimeError(
+            "Yandex AI Studio SDK is not installed. "
+            "Run: python -m pip install -r requirements.txt"
+        ) from exc
+    return AIStudio(folder_id=folder_id)
+
+
+class YandexAIClient:
+    provider = "yandex"
+    display_name = "Yandex AI Studio"
+
+    def __init__(self, folder_id: str, sdk: Any | None = None) -> None:
+        self.folder_id = folder_id.strip("/")
+        self._sdk = sdk if sdk is not None else create_async_sdk(self.folder_id)
+
+    def normalize_model(self, model: str) -> str:
+        normalized = model.strip()
+        if normalized.startswith("gpt://"):
+            return normalized
+        return f"gpt://{self.folder_id}/{normalized.lstrip('/')}"
+
+    async def list_models(self) -> list[str]:
+        try:
+            models = await self._sdk.chat.completions.list()
+        except Exception as exc:
+            raise YandexAIError(f"Не удалось получить список моделей: {exc}") from exc
+
+        model_names = {
+            str(uri)
+            for model in models
+            if (uri := getattr(model, "uri", None))
+        }
+        return sorted(model_names)
+
+    async def chat(
+        self,
+        model: str,
+        messages: list[dict[str, str]],
+        json_mode: bool = False,
+    ) -> str:
+        sdk_messages = [
+            {
+                "role": message.get("role", "user"),
+                "text": message.get("content", ""),
+            }
+            for message in messages
+        ]
+
+        try:
+            completion = self._sdk.models.completions(self.normalize_model(model))
+            if json_mode:
+                completion = completion.configure(response_format="json")
+            result = await completion.run(sdk_messages, timeout=180)
+            content = getattr(result[0], "text", None)
+        except Exception as exc:
+            raise YandexAIError(f"Запрос к модели завершился ошибкой: {exc}") from exc
+
+        if isinstance(content, str) and content.strip():
+            return content.strip()
+        raise YandexAIError("Yandex AI Studio вернула пустой ответ.")
+
+
+class YandexSpeechRecognizer:
+    def __init__(
+        self,
+        folder_id: str,
+        language: str,
+        model: str,
+        sdk: Any | None = None,
+    ) -> None:
+        self.folder_id = folder_id.strip("/")
+        self.language = language
+        self.model = model
+        self._sdk = sdk if sdk is not None else create_sync_sdk(self.folder_id)
+
+    def transcribe(self, audio_path: str | Path) -> str:
+        try:
+            audio = Path(audio_path).read_bytes()
+            recognizer = self._sdk.speechkit.speech_to_text(
+                audio_format=self._sdk.speechkit.AudioFormat.OGG_OPUS,
+                language_codes=self.language,
+                model=self.model,
+            )
+            result = recognizer.run(audio, timeout=180)
+            text = getattr(result, "text", None)
+        except Exception as exc:
+            raise SpeechRecognitionError("Yandex SpeechKit transcription failed") from exc
+
+        return text.strip() if isinstance(text, str) else ""

main.py

@@ -0,0 +1,5 @@
+from assistant_bot.application import run
+
+
+if __name__ == "__main__":
+    run()

requirements-common.txt

@@ -0,0 +1 @@
+python-telegram-bot==22.8

requirements-local.txt

@@ -0,0 +1,4 @@
+-r requirements-common.txt
+faster-whisper==1.2.1
+ctranslate2==4.6.0
+setuptools==80.10.2

requirements-yandex.txt

@@ -0,0 +1,2 @@
+-r requirements-common.txt
+yandex-ai-studio-sdk==0.22.1

requirements.txt

@@ -0,0 +1,3 @@
+-r requirements-local.txt
+-r requirements-yandex.txt
+tzdata>=2024.1; platform_system == "Windows"

tests/__init__.py

@@ -0,0 +1 @@
+"""Project tests."""

tests/test_config.py

@@ -0,0 +1,71 @@
+import os
+import unittest
+from unittest.mock import patch
+
+from assistant_bot import config
+
+
+class WhisperConfigTests(unittest.TestCase):
+    def test_gpu_int8_defaults(self) -> None:
+        variable_names = (
+            config.WHISPER_MODEL_ENV_NAME,
+            config.WHISPER_DEVICE_ENV_NAME,
+            config.WHISPER_COMPUTE_TYPE_ENV_NAME,
+        )
+        with patch("assistant_bot.config.load_env_file"), patch.dict(
+            os.environ,
+            {},
+            clear=False,
+        ):
+            for variable_name in variable_names:
+                os.environ.pop(variable_name, None)
+
+            self.assertEqual(config.get_whisper_model(), "large-v3")
+            self.assertEqual(config.get_whisper_device(), "cuda")
+            self.assertEqual(config.get_whisper_compute_type(), "int8")
+
+
+class AssistantModeConfigTests(unittest.TestCase):
+    def test_local_mode_is_default(self) -> None:
+        with patch("assistant_bot.config.load_env_file"), patch.dict(
+            os.environ,
+            {},
+            clear=False,
+        ):
+            os.environ.pop(config.MODE_ENV_NAME, None)
+            self.assertEqual(config.get_assistant_mode(), "local")
+
+    def test_yandex_mode_is_supported(self) -> None:
+        with patch("assistant_bot.config.load_env_file"), patch.dict(
+            os.environ,
+            {config.MODE_ENV_NAME: "YANDEX"},
+            clear=False,
+        ):
+            self.assertEqual(config.get_assistant_mode(), "yandex")
+
+    def test_unknown_mode_is_rejected(self) -> None:
+        with patch("assistant_bot.config.load_env_file"), patch.dict(
+            os.environ,
+            {config.MODE_ENV_NAME: "cloud"},
+            clear=False,
+        ):
+            with self.assertRaises(RuntimeError):
+                config.get_assistant_mode()
+
+    def test_yandex_model_is_full_gpt_uri(self) -> None:
+        with patch("assistant_bot.config.load_env_file"), patch.dict(
+            os.environ,
+            {
+                config.YANDEX_CLOUD_FOLDER_ENV_NAME: "folder-id",
+                config.YANDEX_CLOUD_MODEL_ENV_NAME: "yandexgpt/latest",
+            },
+            clear=False,
+        ):
+            self.assertEqual(
+                config.get_default_yandex_model(),
+                "gpt://folder-id/yandexgpt/latest",
+            )
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_core.py

@@ -0,0 +1,148 @@
+import sqlite3
+import tempfile
+import unittest
+from contextlib import closing
+from datetime import datetime, timezone
+from pathlib import Path
+from zoneinfo import ZoneInfo
+
+from assistant_bot.agent import execute_agent_tool, parse_agent_decision
+from assistant_bot.reminders import parse_reminder, unit_to_timedelta
+from assistant_bot.storage import AssistantStorage
+
+
+class ReminderParserTests(unittest.TestCase):
+    def test_supported_relative_unit(self) -> None:
+        self.assertEqual(unit_to_timedelta(2, "часа").total_seconds(), 7200)
+
+    def test_absolute_datetime(self) -> None:
+        result = parse_reminder(
+            "2030-01-02 10:30 проверить отчет",
+            ZoneInfo("UTC"),
+        )
+
+        self.assertIsNotNone(result)
+        assert result is not None
+        self.assertEqual(
+            result.remind_at_utc,
+            datetime(2030, 1, 2, 10, 30, tzinfo=timezone.utc),
+        )
+        self.assertEqual(result.text, "проверить отчет")
+
+
+class AgentDecisionTests(unittest.TestCase):
+    def test_tool_call_json(self) -> None:
+        decision = parse_agent_decision(
+            '{"tool_calls":[{"name":"create_note","arguments":{"text":"идея"}}]}'
+        )
+
+        self.assertIsNone(decision.final)
+        self.assertEqual(decision.tool_calls[0]["name"], "create_note")
+        self.assertEqual(decision.tool_calls[0]["arguments"]["text"], "идея")
+
+    def test_context_reset_flag(self) -> None:
+        decision = parse_agent_decision(
+            '{"final":"Перейдем к новой теме","reset_context":true}'
+        )
+
+        self.assertEqual(decision.final, "Перейдем к новой теме")
+        self.assertTrue(decision.reset_context)
+
+
+class StorageTests(unittest.TestCase):
+    def test_legacy_user_settings_gets_yandex_model_column(self) -> None:
+        with tempfile.TemporaryDirectory() as directory:
+            database_path = Path(directory) / "assistant.sqlite3"
+            with closing(sqlite3.connect(database_path)) as connection:
+                connection.execute(
+                    """
+                    CREATE TABLE user_settings (
+                        user_id INTEGER PRIMARY KEY,
+                        ollama_model TEXT,
+                        updated_at TEXT NOT NULL
+                    )
+                    """
+                )
+                connection.commit()
+
+            storage = AssistantStorage(database_path)
+            storage.set_user_model(42, "yandexgpt", "yandex")
+
+            self.assertEqual(storage.get_user_model(42, "yandex"), "yandexgpt")
+
+    def test_provider_models_are_stored_separately(self) -> None:
+        with tempfile.TemporaryDirectory() as directory:
+            storage = AssistantStorage(Path(directory) / "assistant.sqlite3")
+
+            storage.set_user_model(42, "qwen3.5:9b", "local")
+            storage.set_user_model(42, "yandexgpt", "yandex")
+
+            self.assertEqual(storage.get_user_model(42, "local"), "qwen3.5:9b")
+            self.assertEqual(storage.get_user_model(42, "yandex"), "yandexgpt")
+
+    def test_memory_crud(self) -> None:
+        with tempfile.TemporaryDirectory() as directory:
+            storage = AssistantStorage(Path(directory) / "assistant.sqlite3")
+            memory_id = storage.add_memory(42, "короткие ответы")
+
+            self.assertEqual(storage.list_memories(42)[0]["id"], memory_id)
+            self.assertTrue(storage.delete_memory(42, memory_id))
+            self.assertEqual(storage.list_memories(42), [])
+
+    def test_new_context_preserves_searchable_archive(self) -> None:
+        with tempfile.TemporaryDirectory() as directory:
+            storage = AssistantStorage(Path(directory) / "assistant.sqlite3")
+            storage.add_conversation_exchange(
+                42,
+                100,
+                "Обсудим проект Альфа",
+                "Какой аспект проекта интересует?",
+            )
+            storage.start_new_conversation(42, 100)
+            storage.add_conversation_exchange(
+                42,
+                100,
+                "Снова обсуждаем проект Альфа",
+                "Продолжаем обсуждение проекта.",
+            )
+            storage.add_conversation_exchange(42, 200, "Другой чат", "Другой ответ")
+
+            messages = storage.list_conversation_messages(42, 100)
+            matches = storage.search_conversation_messages(42, 100, "проект Альфа")
+
+            self.assertEqual(
+                [(row["role"], row["content"]) for row in messages],
+                [
+                    ("user", "Снова обсуждаем проект Альфа"),
+                    ("assistant", "Продолжаем обсуждение проекта."),
+                ],
+            )
+            self.assertEqual(matches[0]["content"], "Снова обсуждаем проект Альфа")
+            self.assertIn("Обсудим проект Альфа", [row["content"] for row in matches])
+            self.assertEqual(len(storage.list_conversation_messages(42, 200)), 2)
+
+            archive_window = storage.conversation_message_window(
+                42,
+                100,
+                int(matches[-1]["id"]),
+            )
+            self.assertIn(
+                "Какой аспект проекта интересует?",
+                [row["content"] for row in archive_window],
+            )
+
+            tool_result = execute_agent_tool(
+                "search_conversation",
+                {"query": "проект Альфа", "limit": 2},
+                storage,
+                user_id=42,
+                chat_id=100,
+                tz=ZoneInfo("UTC"),
+            )
+            self.assertTrue(tool_result["ok"])
+            self.assertTrue(tool_result["discussions"])
+            self.assertTrue(tool_result["discussions"][0]["messages"])
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_speech.py

@@ -0,0 +1,68 @@
+import tempfile
+import unittest
+from pathlib import Path
+
+from assistant_bot.speech import SpeechRecognitionError, SpeechRecognizer
+
+
+class FakeSegment:
+    def __init__(self, text: str) -> None:
+        self.text = text
+
+
+class SpeechRecognizerTests(unittest.TestCase):
+    def test_transcribe_joins_non_empty_segments(self) -> None:
+        factory_calls = []
+        transcribe_calls = []
+
+        class FakeModel:
+            def transcribe(self, audio_path, **kwargs):
+                transcribe_calls.append((audio_path, kwargs))
+                return iter(
+                    [FakeSegment(" Привет "), FakeSegment(""), FakeSegment("мир")]
+                ), None
+
+        def model_factory(*args, **kwargs):
+            factory_calls.append((args, kwargs))
+            return FakeModel()
+
+        recognizer = SpeechRecognizer(
+            model_name="small",
+            device="cpu",
+            compute_type="int8",
+            language="ru",
+            model_factory=model_factory,
+        )
+
+        with tempfile.TemporaryDirectory() as directory:
+            audio_path = Path(directory) / "voice.ogg"
+            result = recognizer.transcribe(audio_path)
+
+        self.assertEqual(result, "Привет мир")
+        self.assertEqual(
+            factory_calls[0],
+            (("small",), {"device": "cpu", "compute_type": "int8"}),
+        )
+        self.assertEqual(
+            transcribe_calls[0][1],
+            {"language": "ru", "beam_size": 5, "vad_filter": True},
+        )
+
+    def test_transcribe_wraps_model_errors(self) -> None:
+        def failing_factory(*_args, **_kwargs):
+            raise RuntimeError("model is unavailable")
+
+        recognizer = SpeechRecognizer(
+            model_name="small",
+            device="cpu",
+            compute_type="int8",
+            language=None,
+            model_factory=failing_factory,
+        )
+
+        with self.assertRaises(SpeechRecognitionError):
+            recognizer.transcribe("voice.ogg")
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_voice_handler.py

@@ -0,0 +1,75 @@
+import unittest
+from pathlib import Path
+from types import SimpleNamespace
+from unittest.mock import AsyncMock, MagicMock, patch
+
+from assistant_bot.handlers import private_voice
+
+
+class VoiceHandlerTests(unittest.IsolatedAsyncioTestCase):
+    @staticmethod
+    def make_update_and_context(duration: int, transcript: str = "Напомни позвонить"):
+        status_message = SimpleNamespace(edit_text=AsyncMock())
+        message = SimpleNamespace(
+            chat_id=100,
+            voice=SimpleNamespace(duration=duration, file_id="voice-file-id"),
+            reply_text=AsyncMock(return_value=status_message),
+        )
+        update = SimpleNamespace(message=message)
+        telegram_file = SimpleNamespace(download_to_drive=AsyncMock())
+        bot = SimpleNamespace(
+            get_file=AsyncMock(return_value=telegram_file),
+            send_chat_action=AsyncMock(),
+        )
+        recognizer = MagicMock()
+        recognizer.transcribe.return_value = transcript
+        context = SimpleNamespace(
+            bot=bot,
+            application=SimpleNamespace(
+                bot_data={
+                    "speech_recognizer": recognizer,
+                    "voice_max_duration": 120,
+                }
+            ),
+        )
+        return update, context, status_message, recognizer, telegram_file
+
+    async def test_rejects_voice_message_over_duration_limit(self) -> None:
+        update, context, _status, recognizer, _telegram_file = (
+            self.make_update_and_context(duration=121)
+        )
+
+        await private_voice(update, context)
+
+        update.message.reply_text.assert_awaited_once_with(
+            "Голосовое сообщение слишком длинное. Максимум: 120 сек."
+        )
+        recognizer.transcribe.assert_not_called()
+
+    async def test_transcribes_voice_and_passes_text_to_agent(self) -> None:
+        update, context, status, recognizer, telegram_file = (
+            self.make_update_and_context(duration=10)
+        )
+
+        with patch(
+            "assistant_bot.handlers.run_agent_prompt",
+            new_callable=AsyncMock,
+        ) as run_agent:
+            await private_voice(update, context)
+
+        recognizer.transcribe.assert_called_once()
+        temporary_path = Path(recognizer.transcribe.call_args.args[0])
+        self.assertFalse(temporary_path.exists())
+        telegram_file.download_to_drive.assert_awaited_once_with(
+            custom_path=temporary_path
+        )
+        status.edit_text.assert_awaited_once_with("Распознано: Напомни позвонить")
+        run_agent.assert_awaited_once_with(
+            update,
+            context,
+            "Напомни позвонить",
+        )
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_yandex_ai.py

@@ -0,0 +1,132 @@
+import unittest
+from pathlib import Path
+from types import SimpleNamespace
+from unittest.mock import patch
+
+from assistant_bot.yandex_ai import YandexAIClient, YandexSpeechRecognizer
+
+
+class FakeCompletion:
+    def __init__(self) -> None:
+        self.configuration = None
+        self.messages = None
+        self.timeout = None
+
+    def configure(self, **kwargs):
+        self.configuration = kwargs
+        return self
+
+    async def run(self, messages, timeout):
+        self.messages = messages
+        self.timeout = timeout
+        return [SimpleNamespace(text="  готово  ")]
+
+
+class YandexAIClientTests(unittest.IsolatedAsyncioTestCase):
+    async def test_chat_uses_sdk_messages_and_json_mode(self) -> None:
+        completion = FakeCompletion()
+        requested_models = []
+
+        def completions(model):
+            requested_models.append(model)
+            return completion
+
+        sdk = SimpleNamespace(
+            models=SimpleNamespace(completions=completions),
+            chat=SimpleNamespace(
+                completions=SimpleNamespace(
+                    list=self._list_models,
+                )
+            ),
+        )
+        client = YandexAIClient(folder_id="folder-id", sdk=sdk)
+
+        answer = await client.chat(
+            "yandexgpt",
+            [
+                {"role": "system", "content": "Отвечай кратко"},
+                {"role": "user", "content": "Привет"},
+            ],
+            json_mode=True,
+        )
+
+        self.assertEqual(answer, "готово")
+        self.assertEqual(requested_models, ["gpt://folder-id/yandexgpt"])
+        self.assertEqual(completion.configuration, {"response_format": "json"})
+        self.assertEqual(
+            completion.messages,
+            [
+                {"role": "system", "text": "Отвечай кратко"},
+                {"role": "user", "text": "Привет"},
+            ],
+        )
+        self.assertEqual(completion.timeout, 180)
+
+    async def test_list_models_returns_sorted_uris(self) -> None:
+        sdk = SimpleNamespace(
+            chat=SimpleNamespace(
+                completions=SimpleNamespace(
+                    list=self._list_models,
+                )
+            )
+        )
+        client = YandexAIClient(folder_id="folder-id", sdk=sdk)
+
+        self.assertEqual(
+            await client.list_models(),
+            ["gpt://folder/alice-ai/latest", "gpt://folder/yandexgpt/latest"],
+        )
+
+    @staticmethod
+    async def _list_models():
+        return [
+            SimpleNamespace(uri="gpt://folder/yandexgpt/latest"),
+            SimpleNamespace(uri="gpt://folder/alice-ai/latest"),
+        ]
+
+
+class YandexSpeechRecognizerTests(unittest.TestCase):
+    def test_transcribe_uses_speechkit_ogg_opus(self) -> None:
+        calls = []
+
+        class FakeRecognizer:
+            def run(self, audio, timeout):
+                calls.append((audio, timeout))
+                return SimpleNamespace(text="  Привет, мир  ")
+
+        format_value = object()
+
+        def speech_to_text(**kwargs):
+            calls.append(kwargs)
+            return FakeRecognizer()
+
+        sdk = SimpleNamespace(
+            speechkit=SimpleNamespace(
+                AudioFormat=SimpleNamespace(OGG_OPUS=format_value),
+                speech_to_text=speech_to_text,
+            )
+        )
+        recognizer = YandexSpeechRecognizer(
+            folder_id="folder-id",
+            language="ru-RU",
+            model="general",
+            sdk=sdk,
+        )
+
+        with patch.object(Path, "read_bytes", return_value=b"ogg-data"):
+            result = recognizer.transcribe("voice.ogg")
+
+        self.assertEqual(result, "Привет, мир")
+        self.assertEqual(
+            calls[0],
+            {
+                "audio_format": format_value,
+                "language_codes": "ru-RU",
+                "model": "general",
+            },
+        )
+        self.assertEqual(calls[1], (b"ogg-data", 180))
+
+
+if __name__ == "__main__":
+    unittest.main()