Show HN: Personal Telegram Bot with Mani AI Models Like ChatGPT,Claude,Flux etc. (github.com)

A community-built personal Telegram bot that stitches together a huge swath of AI tooling into a single assistant was posted on Show HN. It integrates hosted LLMs (OpenAI GPT-4/5, O3/O4-mini, DALL·E, Whisper/TTS), Anthropic’s Claude 3/3.5/Haiku/Sonnet/Opus series, local models via Ollama (Qwen, Gemma, open-source GPTs), and hundreds of models on Replicate (Stable Diffusion, Flux Pro, Kandinsky, Llama, etc.). The platform supports function-calling/tools, image analysis, speech-to-text and TTS, multi-modal conversations, Google Veo 3.1 video generation with context-aware audio, video downloads from hosting sites, and 17 specialized personas. It also includes a meta-language scheduler for automated tasks and a granular permissions system (per-user flags like is_admin, can_use_tools, can_use_ollama_llm_models). The repo is drop-in runnable: clone, copy config.json.example → config.json, register a Telegram bot with @BotFather and save TELEGRAM_TOKEN, add ALLOWED_USER_NAMES, supply OpenAI/Anthropic/Replicate keys, install/configure Ollama for local models, configure USER_PERMISSIONS, install deps and run run.sh. Significance: it offers a unified, multi-modal interface to mix cloud and local models for experimentation, tooling, and automation—useful for rapid prototyping or private deployments—but demands careful API key management, resource planning (GPU for local models), and security review when exposing function-calling and third-party model endpoints.