Google Drive assistant bot

• License
• How to contribute
• Code of conduct

Semantic search and Q&A Telegram bot for Google Drive — powered by Qdrant and an optional OpenAI-compatible LLM.

---

Why this project

If your Google Drive is full of documents and you keep asking questions like:

• "What did I write last January?"
• "Is there any document mentioning X?"
• "Does my work Google Drive actually contain anything about this?"

This bot turns Google Drive into a searchable knowledge base — and answers directly in Telegram.

---

What it does

• Recursively indexes documents from Google Drive, from:
  • specific folders, or
  • everything accessible to the account
• Splits content into chunks, generates embeddings, and stores them in Qdrant
• Answers questions via /ask using semantic search
• Optionally uses an OpenAI-compatible LLM to generate final answers
• Accepts manual notes from Telegram via /ingest

LLM usage is optional. Without it, the bot returns the most relevant text fragments.

---

Supported file types

• Google Docs
• Google Sheets
• Google Slides
• Text-based files (configuration files, source code, plain text)
• PDF documents
• Microsoft Office formats (DOC, DOCX, XLS, XLSX, PPT, PPTX)

---

Components

• ingest — background service that syncs cloud files to Qdrant
• bot — Telegram interface for asking questions and adding notes
• qdrant — vector database for embeddings and search
• llm (optional) — OpenAI-compatible model for answer generation

---

Architecture

The system consists of three main components: ingestion, retrieval, and answering. Files are indexed in the background and queried via a Telegram bot.

System flows

Ingestion

flowchart LR
    U[User]
    CS[Cloud storage]
    ING[Ingester]
    BOT[Telegram bot]
    RAG[RAG store]
    QD[("Qdrant<br>(vector DB)")]

    U -->|/ingest notes| BOT
    CS -->|files & updates| ING
    ING -->|content + metadata| RAG
    BOT -->|content + metadata| RAG
    RAG -->|upsert vectors| QD

Loading

Retrieval & answering

flowchart LR
    U[User]
    BOT[Telegram bot]
    RAG[RAG store]
    QD[("Qdrant<br>(vector DB)")]
    LLM["LLM<br>(optional)"]

    U -->|/ask question| BOT
    BOT -->|question| RAG
    RAG -->|query vectors| QD
    QD -->|search hits| RAG
    RAG -->|relevant chunks| BOT
    BOT -->|context| LLM
    LLM -->|final answer| BOT

Loading

---

Quick start

1. Copy environment config: cp .env.example .env
2. Configure Google Drive access:
   • Create a Google service account
   • Download its JSON key
   • Place it at secrets/google_sa.json
   • Share target Google Drive folders with the service account email
   • Enable Google Drive API, Google Docs API, Google Sheets API, and Google Slides API
3. Set required environment variables:
   • TELEGRAM_BOT_TOKEN
   • Either STORAGE_GOOGLE_DRIVE_FOLDER_IDS (JSON array) or STORAGE_GOOGLE_DRIVE_ALL_ACCESSIBLE=true
4. Start services: make start

Google APIs and scopes

The bot uses these APIs during ingestion:

• Google Drive API: list files, export Google-native docs, download binaries
• Google Docs API: extract Google Docs text
• Google Sheets API: extract Google Sheets tables
• Google Slides API: extract Google Slides text

Service account scopes used by the app:

• https://www.googleapis.com/auth/drive.readonly
• https://www.googleapis.com/auth/documents.readonly
• https://www.googleapis.com/auth/spreadsheets.readonly
• https://www.googleapis.com/auth/presentations.readonly

---

Access control (optional)

By default, the bot is publicly accessible. You can restrict access using the following options:

• Set TELEGRAM_ALLOWED_USER_IDS — a JSON array of Telegram user IDs allowed to interact with the bot in private chats
• Set TELEGRAM_ALLOWED_GROUP_IDS — a JSON array of Telegram group or supergroup IDs where the bot is allowed to respond

Behavior:

• If both lists are empty (default), the bot responds to everyone
• If user IDs are set, only those users can use the bot in private chats
• If group IDs are set, the bot responds only in those groups (any member can use it there)
• If access is not allowed, the bot silently ignores the message

To find your Telegram user or group ID, you can use bots like @userinfobot or @getidsbot.

---

Telegram commands

• /start — show help
• /ask <question> — search the knowledge base and answer
• /ingest <text> — manually add a note to the knowledge base

---

Configuration

All settings are defined via .env.

Name	Required	Default	Description
TELEGRAM_BOT_TOKEN	yes	–	Telegram bot token
TELEGRAM_ALLOWED_USER_IDS	no	[]	Allowed user IDs for private messages (JSON array)
TELEGRAM_ALLOWED_GROUP_IDS	no	[]	Allowed group IDs for group chats (JSON array)
STORAGE_BACKEND	no	google_drive	Storage provider identifier
STORAGE_GOOGLE_DRIVE_SERVICE_ACCOUNT_JSON	no	/run/secrets/google_sa	Service account credentials
STORAGE_GOOGLE_DRIVE_FOLDER_IDS	conditional	[]	Google Drive folder IDs (JSON array)
STORAGE_GOOGLE_DRIVE_ALL_ACCESSIBLE	conditional	false	Index all accessible files
STORAGE_GOOGLE_DRIVE_MAX_ROWS_PER_SHEET	no	10000	Max rows per sheet
STORAGE_GOOGLE_DRIVE_BACKOFF_RETRIES	no	8	Retry count for Google API
STORAGE_GOOGLE_DRIVE_BACKOFF_BASE_DELAY_SECONDS	no	1.0	Backoff base delay
STORAGE_GOOGLE_DRIVE_BACKOFF_MAX_DELAY_SECONDS	no	30.0	Backoff max delay
STORAGE_GOOGLE_DRIVE_API_RPS	no	8.0	Google API rate limit (tokens/sec)
STORAGE_GOOGLE_DRIVE_API_BURST	no	16.0	Google API burst size (tokens)
FILE_TYPE_GDOCS_ENABLED	no	true	Enable Google Docs ingestion
FILE_TYPE_GSHEETS_ENABLED	no	true	Enable Google Sheets ingestion
FILE_TYPE_GSLIDES_ENABLED	no	true	Enable Google Slides ingestion
FILE_TYPE_TEXT_BASED_ENABLED	no	true	Enable text-based files ingestion
FILE_TYPE_PDF_ENABLED	no	true	Enable PDF ingestion
FILE_TYPE_DOCX_ENABLED	no	true	Enable DOCX ingestion
FILE_TYPE_DOC_ENABLED	no	true	Enable DOC ingestion
FILE_TYPE_XLSX_ENABLED	no	true	Enable XLSX ingestion
FILE_TYPE_XLS_ENABLED	no	true	Enable XLS ingestion
FILE_TYPE_PPTX_ENABLED	no	true	Enable PPTX ingestion
FILE_TYPE_PPT_ENABLED	no	true	Enable PPT ingestion
TEXT_MAX_FILE_SIZE_MB	no	10	Max text-based file size (MB)
PDF_MAX_PAGES	no	100	Max extracted PDF pages
PDF_MAX_FILE_SIZE_MB	no	50	Max PDF file size (MB)
PDF_EXTRACTION_ENGINE	no	pypdf	pypdf or pdfplumber (both installed)
OFFICE_MAX_FILE_SIZE_MB	no	50	Max Office file size (MB)
EXCEL_MAX_SHEETS	no	50	Max sheets for XLS/XLSX extraction
QDRANT_URL	no	http://qdrant:6333	Qdrant endpoint
QDRANT_COLLECTION	no	docs	Collection name
EMBED_MODEL	no	sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2	fastembed model
HF_TOKEN	no	–	HuggingFace token for model downloads
TOP_K	no	6	Retrieved chunks per query
MAX_CONTEXT_CHARS	no	6000	Context size limit
LLM_BASE_URL	no	https://api.openai.com/v1	OpenAI-compatible API base
LLM_API_KEY	no	–	LLM API key
LLM_MODEL	no	gpt-4o-mini	LLM model name
LLM_SYSTEM_PROMPT	no	Built-in prompt	LLM system prompt
INGEST_MODE	no	loop	once or loop
INGEST_POLL_SECONDS	no	600	Poll interval
INGEST_WORKERS	no	6	Parallel workers
INGEST_PROGRESS_FILES	no	25	Log every N files
INGEST_PROGRESS_SECONDS	no	30	Log progress at least every N sec
INGEST_SHUTDOWN_GRACE_SECONDS	no	20	Graceful shutdown delay
HEALTH_HOST	no	localhost	Health server bind host
BOT_HEALTH_PORT	no	8080	Bot health endpoint
INGEST_HEALTH_PORT	no	8081	Ingest health endpoint
LOG_LEVEL	no	INFO	Logging level
LOG_PLAIN_TEXT	no	false	Use plain text logs instead of JSON

Set either STORAGE_GOOGLE_DRIVE_FOLDER_IDS (JSON array) or STORAGE_GOOGLE_DRIVE_ALL_ACCESSIBLE=true when using the Google Drive provider.

The service account JSON key must exist at the configured STORAGE_GOOGLE_DRIVE_SERVICE_ACCOUNT_JSON path.