telegram

Telegram Bot

Connect Adam to Telegram. Send text or photos to the bot and get AI responses. Supports tools, memory, vision, and all Adam features.

Telegram Setup

1. Create a Bot

1. Open Telegram and search for @BotFather
2. Send /newbot
3. Choose a name (e.g. "My Adam Bot")
4. Choose a username (must end in bot, e.g. my_adam_bot)
5. BotFather will give you a token like 7123456789:AAH... — copy it

2. Configure

Create a .env file in this directory (or the project root):

TELEGRAM_BOT_TOKEN=7123456789:AAHxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
ANTHROPIC_API_KEY=sk-ant-...

Any LLM provider works — set ANTHROPIC_API_KEY, OPENAI_API_KEY, or GEMINI_API_KEY. For local models, use --local instead. Add OPENAI_API_KEY for voice message support (Whisper STT).

3. Build & Run

make
./telegram

Or with a local model:

./telegram --local ../../models/gemma-3-4b-it-Q4_K_S.gguf

With vision:

./telegram --local ../../models/gemma-3-4b-it-Q4_K_S.gguf --mmproj ../../models/mmproj-gemma-3-4b-it-F16.gguf

4. Chat

Open your bot in Telegram (search for the username you chose) and press Start. Send any message.

Bot Commands

Command	Description
/start	Welcome message with instructions
/tools	Enable calculator and web search
/memory	Enable persistent memory (survives restart)
/clear	Clear conversation history
/status	Show current configuration

Features

• Text chat — multi-turn conversation with full history
• Voice messages — send voice notes, transcribed via Whisper STT then processed
• Image input — send photos and the bot describes/analyzes them
• Tool calling — calculator, web search (enable with /tools)
• Memory — persistent knowledge across conversations (enable with /memory)
• All providers — Anthropic, OpenAI, Gemini, or local llama.cpp

How It Works

The bot uses Telegram's long polling (getUpdates with 30s timeout) — no webhook or public IP needed. Each message is processed sequentially through adam_run().

For photos, the bot downloads the highest-resolution version via getFile, attaches it to the conversation via adam_history_attach(), and sends the result. Voice messages are downloaded as OGG/Opus files and transcribed via the OpenAI Whisper API (adam_stt_transcribe), then the transcribed text is processed through adam_run() like a normal message.