Live Recording AI Workbench
Turn one game livestream into content your team can publish, review and reuse. Recording starts with the stream, AI keeps scoring highlights, and finished clips arrive while the broadcast is still running. It is not software we license out — it is the production tool behind our delivery, and your team gets the output.
102 streams covered · 63.7 hours recorded · 3,639 finished assets delivered · zero misspelled game titles (data through 2026-08-31).
Faster, safer, more accurate — with numbers you can audit
- Faster: 3.2 minutes from stream start to the first finished asset (median across 18 streams in August 2026); delivery happens during the broadcast.
- Safer: 99.5% source-file integrity; across the last 7 streams all 388 source files were archived complete.
- More accurate: 5 database snapshots x 10 text fields x 8 competitor game titles — 400 checks, zero wrong game names.
What your team receives after a stream
Finished MP4 clips, the evidence behind each highlight call, the full recording, in-stream segment previews, a timestamped event log, chapter markers, a CapCut-ready package, the full transcript, chat sentiment analysis, and one delivery ZIP.
One pipeline from stream start to delivery
Reliable recording, live understanding, highlight scoring, delivery and review. Your editors stop hunting through footage and start picking from a shortlist: recording first, AI shortlisting, human final call.
Coverage so far
49 tournament broadcasts and 53 creator streams across six titles, including Delta Force, Honor of Kings: Chess, PUBG, Call of Duty, Dungeon & Fighter and Ace Force.
Content security
Recording, chat capture, speech-to-text, search indexing and clipping all run on local machines. Cloud semantic analysis receives text only; frame-level analysis uses a small sample of frames and can be switched off for an entire stream.
Value white paper v1.7
A 12-page edition with annotated interface walkthroughs, per-stream data by title, the exact method behind the brand-name error checks, and our content-security boundaries. Currently available in Chinese; an English version is available on request. Download the PDF.