recap — structured data from any URL

recap Point it at any URL, PDF, video, or local file.
Define the fields you want. Get clean, structured data out.

CLI HTTP API Web UI Go 1.25+ MIT License

recap — zsh — 120×36

~/research via go 1.25

how it works

[PDF]

$ recap run
"paper.pdf"
--template research-paper

{
  "title": "Attention Is All You Need",
  "authors": ["Vaswani et al."],
  "methodology": "transformer...",
  "key_findings": [...]
}

[VIDEO]

$ recap run
  "youtube.com/watch?v=..."
  --template youtube-video
  --format markdown

## The Intelligence Explosion
**channel:** Lex Fridman
**duration:** 187 min
**takeaways:** [3 items]

[LOCAL]

$ recap run
  ~/docs/meeting-notes.md
  --fields
    "action_items:list,
     owner:text,due:date"

{
  "action_items": [3],
  "owner": "humza",
  "due": "2025-04-01"
}

field syntax

--fields → hover each token to learn the type

title , author:text , published:date , tags:list , rating:number , is_opinion:boolean , summary:paragraph , sentiment:enum[positive,negative,neutral]

text plain string

list array of strings

number numeric

boolean true / false

date ISO 8601

paragraph long text

enum[...] constrained values

three interfaces · one output

⌨ CLI for scripting & automation

# pipe it, schedule it, automate it

$ recap run "$URL" \
  --template news-article \
  --format json \
  --output out.json

# pipe to jq
$ recap run "$URL" | jq '.tags'

⚡ HTTP API for integrations

# start the server, hit the endpoint

$ recap serve
→ listening on :8080

$ curl -X POST \
  localhost:8080/api/v1/extract
  -d '{"url":"...",
    "fields":[...]}'

◻ Web UI for visual workflows

# the same thing, in a browser

$ make dev
→ go server on :8080
→ next.js ui on :3000

open http://localhost:3000

visual field builder · history · templates

# identical output. your choice of interface.

supported content types

WEB

Web Pages

Any URL. Article text extracted with Readability, noise stripped, content passed to the model.

PDF

PDF Files

Detected by .pdf extension or Content-Type header. Text extracted from all pages automatically.

LOCAL

Local Files

Pass a path with no scheme — ./doc.txt, ~/notes.md — read directly from disk.

VIDEO

Video & Audio

.mp4 .mov .webm .mp3 .wav .m4a — audio extracted with ffmpeg, transcribed via Whisper, then extracted.

supported models

$ recap config models configure once · use forever

PROVIDER MODEL ID

── Anthropic ──────────────────────────

Anthropic Claude Sonnet 4 default claude-sonnet-4-20250514

Claude Opus 4 claude-opus-4-20250514

Claude Haiku 4 claude-haiku-4-20250514

── OpenAI ──────────────────────────────

OpenAI GPT-4o gpt-4o

O3 o3

O4 Mini o4-mini

── Google ──────────────────────────────

Google Gemini 2.5 Pro gemini-2.5-pro

Gemini 2.5 Flash gemini-2.5-flash

get started

clone & build

$ git clone https://github.com/humzakhan/recap.git
$ cd recap && make build
# binary at ./bin/recap

configure your api key

# pick at least one provider
$ recap config set api_keys.anthropic sk-ant-... # anthropic
$ recap config set api_keys.openai sk-... # openai
$ recap config set api_keys.google AIza... # google

run your first extraction

$ ./bin/recap run "https://example.com/article" \
--fields "title,author,summary:paragraph,tags:list"