// the take
A Telegram bot that proxies LLM APIs (primarily OpenAI-compatible endpoints) with multi-model support, plugins (search, ArXiv, code interpreter), and group chat management. Targets individuals or small teams who want a self-hosted AI assistant in Telegram without building one from scratch.
Multi-provider support via uni-api integration is genuinely useful — one bot config covers dozens of backends without forking the code. Model grouping via CUSTOM_MODELS syntax (semicolons/colons) is a clean config-driven approach that doesn't require code changes. Conversation isolation per-window is implemented correctly and the whitelist/blacklist/admin tier logic covers real group chat abuse scenarios. Docker deployment is straightforward with sane volume mounting for config persistence.
Core API logic lives in a git submodule (aient) that's a separate repo — debugging issues means context-switching between two codebases and the submodule pin can silently lag. The configuration surface is enormous (30+ env vars) with interdependencies that aren't validated at startup, so misconfiguration fails silently at runtime. No tests worth mentioning — the test directory has a handful of formatting scripts, nothing that covers bot logic or plugin behavior. Config persistence is flat files in user_configs/ rather than a proper store, which will cause concurrency headaches if multiple users hit the bot simultaneously under load.
We read the README, the directory tree and a key source file in June 2026. Nobody pays to be listed. How we judge →