Installation info on my other site

But the short answer is:

brew install llm

# Paste your OpenAI API key into this
llm keys set openai

llm install llm-gemini
llm keys set gemini

llm install llm-anthropic
llm keys set anthropic

Links

• llm
• llm-gemini
• llm-anthropic

HTML + Text

wget https://thefocus.ai/blog -O blog.html

cat blog.html | \
llm -m gemini-2.5-pro-exp-03-25 \
    --schema-multi 'title, date str: mm:ss,description,topics' \
    parse | \
tee blog.json

Or with claude

wget https://turingpost.com/ -O turingpost.html

cat turingpost.html | \
llm -m claude-3.7-sonnet \
    --schema-multi 'title, link, date str: yyyy/mm/dd,description,topics' \
    parse | tee turingpost2.json

Make sense of some real estate data

wget --user-agent="Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/122.0.0.0 Safari/537.36" "https://www.redfin.com/NY/Binghamton/3-Brookside-Rd-13903/home/115601261"

cat 115601261 | llm -m gemini-2.5-pro-exp-03-25 "extracts the price history of this property and who the broker was"

Or if you want something more structured

cat 115601261 | \
llm -m gemini-2.5-pro-exp-03-25 \
	--schema-multi 'date: status, broker: broker name of was in the transaction, description: listed/price change/offered/sold etc, price: the price' \
	"figure out the price history for when this property was bought and sold"

Audio

Here’s one to get a transcript of an audio file, for example of a pod cast

wget -O audio.mp3 "https://podtrac.com/pts/redirect.mp3/tracking.swap.fm/track/SxlTEPDY7xDg35RXkASs/traffic.omny.fm/d/clips/e73c998e-6e60-432f-8610-ae210140c5b1/afbd76b8-eff2-442a-b938-b28e0126edad/4c60a33c-1249-4bd6-ae78-b29100f1413d/audio.mp3?utm_source=Podcast&in_playlist=d08826cd-f888-4cd3-b700-b28e0126edbb"

llm -m gemini-2.5-pro-exp-03-25 \
	-a audio.mp3 \
	--schema-multi 'timestamp str: mm:ss,ad bool, speaker, text, tone: the conversation tone' \
	transcript | tee transcript.json

Video

Lets pull down something from youtube

  brew install yt-dlp

Grab something, the smallest something and turn it into an mp4 if it isn’t:

# These flags try to get the smallest filesize
yt-dlp -S +size,+br,+res,+fps \
	-o video.webm \
	"https://www.youtube.com/watch?v=ndvRsQx3xBw"

# Convert it to mp4 if its in some unsupported format
ffmpeg -i video.webm.mkv video.mp4

And then lets see what happens:

llm -m gemini-2.5-pro-exp-03-25 \
	-a video.mp4 \
    --schema-multi 'timestamp: in the format mm:ss,
     text: what was said,
     mood: what was the tone of voice,
     sponsor_section bool: is this section an add or not,
     ' \
    transcript | \
tee video.json

which gives us

sed '1,2d;$d' video.json | jq '.items | length'

47 items. Or we can look at the moods:

sed '1,2d;$d' video.json | jq '.items[].mood' | sort | uniq -c | sort -nr
  18 "Analytical"
  15 "Explanatory"
   7 "Informative"
   2 "Conclusive"
   2 "Comparative"
   1 "Reflective"
   1 "Problem-solving"
   1 "Confident"

Schema

This command helps see how schemas are defined

llm schemas dsl

And you can do multi, so:

llm schemas dsl --multi 'timestamp: in the format mm:ss,
     text: what was said,
     mood: what was the tone of voice,
     sponsor_section bool: is this section an add or not,
     '