Local inference
Models run on the machine and captured audio and video never leave it. Unreleased products and internal training material are the normal case, not the exception.
01The problem
Value drains fastest right after the stream ends
A live session is information-dense: what was explained, which steps were demoed, what the audience asked, what was promised as follow-up. All of it exists at the end — and an hour later only a vague impression remains.
- Reviewing means scrubbing a timeline; finding one sentence costs more than re-explaining it.
- Chapters, key points and follow-ups are manual work, which in practice means they never get done.
- Uploading footage to a cloud transcription service hands unpublished material to a third party.
- The recording needs a second editing pass before it can be published, adding another day of delay.
02Capabilities
Output while you are still live
Live chapter splitting
Chapters are detected as the topic shifts and titled automatically, so a full table of contents exists by the time the stream ends.
Key points and follow-ups
Conclusions, figures and commitments are extracted from the narration into an actionable list.
Screen aligned with narration
Demo actions are aligned to the same timeline as the commentary, so review does not require watching two streams at once.
One-click recap export
Structured prose plus chapter timestamps, ready for a blog post, documentation or a second edit.
Bilingual Chinese and English
Transcription and summaries support both languages, for cross-border teams and English-language technical streams.
03Who it is for
Three main scenarios
Primary use cases
| Scenario | What matters |
|---|---|
| Technical streams | Extracting key points from live coding, chaptered replay, a publishable write-up the next day |
| Corporate training | Content stays on the internal network, attendance and key points recorded, material archived automatically |
| Remote teaching | Lecture points captured, student questions collected, revision outline generated |
04Technical notes
How it is built
StreamCopilot components
| Part | Notes |
|---|---|
| Inference | A local model service; desktop capture feeds transcription and summarisation on-device |
| Desktop app | Capture, timeline alignment and export; talks to the local inference service |
| Output | Structured prose, chapter timestamps, key points and a follow-up list |
| Boundary | No cloud transcription dependency; anything requiring network states its data flow explicitly |
05Quick facts
At a glance
Product summary
| Item | Detail |
|---|---|
| Status | Live |
| Product | StreamCopilot (StreamCopilot) |
| Standalone site | https://zbzs.zyaisoft.xyz/ |
- Built for tech streams, corporate training and remote classes
- Local models plus desktop capture
- Bilingual Chinese / English output