No key, no account, no download. Belgo first uses his own AI, which runs gpt-oss on this site's Cloudflare account, and the header shows "Own AI" when it answers. If that is not available he falls back to the free Pollinations text service, where your messages are then sent.
Runs on your device's processor with wllama, so it works in any modern browser. Bigger models answer better but are slower and need more memory. Each one downloads once. If a model fails to load, Belgo tries the next smaller one by himself.
Credits
Belgo was inspired by Capri. Thank you to its makers for the idea of an assistant with a body.
Belgo's own AI runs gpt-oss by OpenAI and FLUX by Black Forest Labs on Cloudflare Workers AI. Apps are written by Qwen 3.8 (Alibaba) or GLM-4.7 Flash (Z.ai), or Kimi K2.7 Code (Moonshot AI) on a paid Cloudflare plan.
Text answers come from Cerebras (gpt-oss, then Qwen 3.8) when its key is set. Photos are read by Qwen 3.8 or Gemma 4 on Workers AI.
DOOM runs on the wasm port by Cornelius Diekmann (GPL source, id Software shareware data). Quake II runs on Qwasm2 by GMH-Code, with id Software's demo.
Backup answers come from Pollinations; the GitHub option uses GitHub Models.
On-device answers run on wllama by Xuan Son Nguyen, built on llama.cpp.
Pictures are painted by Z-Image-Turbo (Tongyi-MAI) and FLUX.2 klein (Black Forest Labs) on their public Hugging Face Spaces.
