hsbot

A telegram bot for generating LLM responses, manipulating images and transcribing audio.

Prerequisites

• Telegram bot API token
• OpenRouter API key
• fal.ai API key
• ImageMagick binary installed

Copy config.sample.toml to config.toml and set your keys/options.

Handlers

• /chat: Keeping conversation context for a duration defined in the config, this handler uses OpenRouter to generate chat responses. Use #keyword in a message to target a specific model. Also works with replying to images, when using a model that supports vision.

{Keyword: "claude", Identifier: "anthropic/claude-3.5-sonnet:beta"}
{Keyword: "gpt", Identifier: "openai/gpt-4.1"}
{Keyword: "gemini", Identifier: "google/gemini-2.5-pro-preview"}
{Keyword: "grok", Identifier: "x-ai/grok-3-beta"}
{Keyword: "deepseek", Identifier: "deepseek/deepseek-chat-v3-0324"}
{Keyword: "unslop", Identifier: "thedrummer/unslopnemo-12b"}

• /image: Generating images from a prompt, set to use Flux as default.
• /edit: Edit images via prompt
• /scale: Liquid rescale images with a power factor
• /transcribe: Transcribe audio files and voice messages

Development

The base architecture is hexagonal. For business logic and its interfaces, extend the ports side on internal/core. Implementations that talks to something else than the business logic should be created as an adapter in internal/adapters.

Commands are stored and fetched dynamically, use the CommandRegistry to register new commands. After that, you can create the handler in main.go.