01
TL;DR and Executive Summary
A short summary up front, and who the video is for.
Open-source project
transcript-report
Turn a YouTube transcript into an honest, practical briefing, on your own GPU.
- Python 3.10+
- No dependencies
- Runs locally
- MIT licence
- v0.2.0
The example report is based on an AI-generated sample transcript, not a real video: view the transcript (TXT), opens in a new tab. The report and transcript both open in a new tab.
The report
What’s in a Report
02
Actionable Points
Ranked by impact and effort, with a phased implementation plan.
03
Accuracy Check
The main claims, marked ✅ / ⚠️ / ❌ / ❓, and which ones to check yourself.
04
Critical Analysis
Bias and sponsorships, weak arguments, what’s missing, and the speaker’s credibility.
05
Glossary and Resources
The terms used in the video, and the resources it mentions.
06
Content Density and AI Likelihood
A content density score (how much is filler) and an AI-generated likelihood score.
TL;DR: A 24-minute SaaS growth video delivers five solid early-stage tips in roughly 5 minutes, buried under ~80% filler, sponsor reads, and recycled loops.
Why it’s different
Private, Any Length, Nothing to Install
01
Private
The transcript, the model and the report stay on your computer. It only talks to LM Studio or Ollama.
02
Any Length
Long videos are split into sections, summarised and combined. If a long run stops, --resume carries on from the last saved section.
03
Nothing to Install but Python
It uses only the standard library. PDFs are made by printing with Edge or Chrome, so no PDF library is needed.
How it works
From Transcript to Report
Transcript
Clean up timestamps, captions
Fits the context?
Yes
: go straight to writing the report.
No
:
Split into sections
Notes for each section saved as each one is done
Merge the notes until they fit
Write the report
Markdown or PDF
Requirements
What You Need
- LM Studio or Ollama (both free)
- A GPU with about 8 GB of VRAM or more (tested on NVIDIA)
- Python 3.10 or newer
- Microsoft Edge or Google Chrome, only for PDF reports
Note: developed and used on Windows 11. macOS and Linux are supported by the code but not yet tried on real videos.
Quick start
5 Steps to Your First Report
-
Install LM Studio and open it once. Download a model, then run
lms lsto see its key.lms ls -
Install transcript-report:
python -m pip install git+<REPO_URL> -
Create settings and folders, then put your model’s key and
"ctx"insettings.json.python -m transcript_report --init -
Check the setup:
python -m transcript_report --check -
Run it:
python -m transcript_report "My video.txt" --title "Video title - Channel"
Using Ollama, other options and troubleshooting: see the README
Models
Choosing a Model
| VRAM | Model size and quantisation | ctx to try |
|---|---|---|
| 8 GB | 7–9B at Q4_K_M | 16384 |
| 12 GB | 8–9B at Q6_K | 32768–65536 |
| 16 GB | 12–14B at Q4_K_M–Q5_K_M | 32768 |
| 24 GB | 14B at Q8_0, or 24–32B at Q4_K_M | 32768–65536 |
Limitations
What It Can’t Do
- The accuracy check is offline: the model only knows what it learnt in training, so it marks recent claims “can’t verify” and lists the claims to check yourself.
- A report is only as good as the model and the transcript. Auto-captions have no punctuation and contain misheard words.
- The prompts are written for English-language tech and business videos.