Office Doc Assistant

A Telegram bot that answers questions about Excel, CSV, and PDF documents using Claude (Anthropic).

Upload a file → ask anything → get a grounded answer based on the document content.

Use case. Sales sheets, financial reports, contracts, scraped data exports — anything you'd otherwise scroll through manually. Ideal for ops/finance teams who need quick lookups without standing up a full BI stack.

Features

• Multi-format input — .xlsx, .xls, .csv, .pdf
• Grounded answers — Claude is instructed to answer strictly from document content (or say so when info is missing)
• Prompt caching — repeated questions about the same document within ~5 minutes are billed at ~10% of the first call's price
• Per-chat memory — each Telegram chat keeps its current document; /clear to forget
• Containerized — single docker run deploy

Architecture

user ──Telegram──▶ bot.py ──▶ parser.py ──▶ plain text
                       │                        │
                       └──▶ llm.py ──▶ Claude API (cached)
                                          │
                       ◀── answer ────────┘

File	Responsibility
src/bot.py	Telegram update loop, document/text handlers, in-memory chat store
src/parser.py	Decode .xlsx / .csv / .pdf bytes into plain text
src/llm.py	Claude wrapper with prompt caching of system + document

Stack

Layer	Library
LLM	anthropic (Claude Opus 4.7 by default)
Telegram	python-telegram-bot
XLSX	openpyxl
CSV	pandas
PDF	pypdf

Quickstart

1. Get credentials

• Telegram bot token — message @BotFather, /newbot, save the token
• Anthropic API key — console.anthropic.com

2. Run with Docker

git clone https://github.com/DzenInCode/office-doc-assistant
cd office-doc-assistant
cp .env.example .env
# edit .env — fill TELEGRAM_BOT_TOKEN and ANTHROPIC_API_KEY

docker build -t office-doc-assistant .
docker run --env-file .env office-doc-assistant

3. Run locally (no Docker)

python -m venv .venv
source .venv/bin/activate     # on Windows: .venv\Scripts\activate
pip install -e .

export TELEGRAM_BOT_TOKEN=...
export ANTHROPIC_API_KEY=...
office-doc-bot

Then in Telegram: open your bot → /start → send a file → ask a question.

Configuration

All set via env vars (or .env with Docker):

Variable	Default	Description
TELEGRAM_BOT_TOKEN	(required)	Bot token from @BotFather
ANTHROPIC_API_KEY	(required)	Anthropic API key
ANTHROPIC_MODEL	claude-opus-4-7	Override to claude-sonnet-4-6 for ~5× cheaper
ANTHROPIC_MAX_TOKENS	4096	Max output tokens per answer
MAX_FILE_SIZE_MB	20	Per Telegram bot API limit
LOG_LEVEL	INFO	Standard Python logging level

Commands

• /start, /help — show help
• /clear — forget the currently stored document for this chat
• Send a file — store it for this chat (overwrites previous)
• Send a text message — ask a question about the stored file

Limitations

• In-memory storage — documents are lost on restart. For production, swap _documents in src/bot.py for Redis or a DB.
• One document per chat — sending a new file overwrites the previous one.
• Telegram bot limits — max file size 20 MB on the free Bot API. Premium bots can do 50 MB.
• PDF text extraction — image-only / scanned PDFs need OCR (not included).

Tests

pip install -e ".[dev]"
pytest

License

MIT — see LICENSE.