// the find
yihong0618/tg_bot_collections
collections of yihong0618's telegram bot
A grab-bag Telegram bot that exposes a dozen or so AI/media APIs (Gemini, Claude, GPT, Yi, Cohere, MiniMax, Groq's Llama, Together's Qwen, Dify) plus some unrelated toys like poster generation, map rendering, TTS and Kling image/video gen, all behind one bot process. It's for someone who wants a personal multi-model Telegram assistant and doesn't mind running yihong0618's own side project rather than something built for general adoption.
Provider integration is genuinely modular — each LLM gets its own file under handlers/ with a consistent `provider: message` command convention, so bolting on a new model is a small, isolated change. Config is entirely env-var driven with sane per-provider overrides (e.g. ANTHROPIC_BASE_URL, MINIMAX_API_TYPE) for routing through third-party or regional endpoints. Docker build/run is a one-liner per provider, which lowers the bar for just trying it.
No test suite in the repo despite a CI workflow — CI.yaml almost certainly just runs black/lint, not any functional check, so regressions in a dozen handler files rely entirely on manual testing. Telegraph token handling defaults to writing a bearer token to a plaintext token_key.json on disk with no mention of permissions or rotation. There's no shared conversation/context abstraction visible — each handler looks like it talks to its provider independently, so context window handling and history are presumably duplicated or missing per-handler rather than centralized. It's a personal collection first (optimized for yihong0618's own channel), not a library, so expect README-as-documentation and no versioned releases or changelog to track breaking changes across the many provider integrations.