In production2026BotServer
Telegram AI agents with model fallback
Always-on AI agents reachable from Telegram, running on a VPS and an Intel NUC, that fall back across model providers and local models.
What I built
- Fallback chain across OpenAI, Google and xAI models so the agent keeps answering when one provider fails
- Local Ollama inference evaluated on NUC, Apple Silicon and an Android phone
- Config-schema, token and CPU-inference issues solved across machines