Original website:
Don’t say it, it’s actually quite impressive mathematically, mainly because it’s fast—you get it, the translation goes crazy. I might not have adapted this non‑stream request, but you can take the original project and tweak it yourself to support non‑stream, that’s all.
Usage effect (the context seems a bit off; since this model has limited context, those who need more can add context themselves, feel free to modify the project)
Model list and settings in Cherry Studio:
Online demo URL:
https://questionai-2api.to2ai.workers.dev/
Open‑source repository:
Project technical features:
Headless browser simulation
Fingerprint forgery technology (UID generator with 18‑digit big integer seed)
Temporal synchronization attack (Fiber ID/Server Call ID timing control)
Automatic session persistence (handshake mechanism to obtain bubble_session_uid)
Perfect request header cloning (full x‑bubble header series + Chrome 142 UA)
Bypass Cloudflare verification
Native streaming SSE conversion
Anti‑detection payload construction (Element instance dehydration)
Multi‑dimensional answer extraction engine (response1_text prioritized + _api_c2_choices as fallback + deep traversal fallback)
