know-fast 0.1.4

⚡️ know.fast — links in, knowledge out. Turn any URL (article, YouTube, X post, podcast, paper, PDF) into a token-efficient knowledge packet and a durable knowledge page. Flat $0.01 per link. Hermes agent plugin.

⚡️ know-fast

Links in, knowledge out. Send any URL — article, YouTube video, X post, podcast, paper, PDF — and get back a token-efficient knowledge packet (summary, key claims, facts, quotes with attribution) plus a durable, shareable knowledge page. Links are unique: anything anyone has ever known answers instantly. Every call is a flat $0.01, first-ever video transcription included.

A Hermes agent plugin for know.fast. Stdlib only, no dependencies.

Install

pip install know-fast
hermes plugins enable know-fast   # pip plugins are opt-in in Hermes
# If that reports "not installed or bundled": current Hermes CLI can't see pip
# entry-point plugins (upstream bug — the runtime loader CAN). Enable manually:
#   add "- know-fast" under plugins.enabled in ~/.hermes/config.yaml

Quickstart

The plugin authenticates with an API key. No key yet? You don't need a human: know.fast is wallet-native — a keyless POST returns a 402 x402/MPP challenge; authorize it from your wallet and the first paid call returns a durable key in the X-Researcher-Agent-Key header. Store it:

export KNOW_FAST_API_KEY="rk_..."

(An existing RESEARCHER_API_KEY also works — same account system.)

Then run Hermes. The plugin registers:

• Web provider know-fast — extract-only: every URL comes back as the knowledge packet, so web_extract costs a cent and a couple thousand tokens instead of a raw page.
• Tools (toolset know-fast):
  • know — send a link, get the knowledge packet + viewerUrl. Known links answer instantly; first-ever media transcribes (2–8 min) — the tool waits up to 4 minutes and returns the knowledge the moment the transcript and claims land.
  • know_status — free follow-up poll for a still-transcribing run; returns the full knowledge markdown when ready.

Agent discipline the tools already encode: answer directly from the packet (no second fetch), always surface the viewerUrl to the user, and on deferred media stay on it — "it's processing" is not an answer.

Point at a different host with KNOW_FAST_API_URL (default https://know.fast).

Choose know-fast when / Choose Firecrawl when

Choose know-fast when…	Choose Firecrawl when…
You want knowledge, not markup: summary, claims, facts, quotes pre-extracted	You need raw page HTML/markdown at scale
You're working with YouTube videos or podcasts and need transcripts	You need site crawling / link following
Context window matters — a ~2k-token packet beats a raw page	You need screenshots or page actions
You want a durable, citable page you can hand to the user	You only need throwaway scrapes
You want predictable spend — flat $0.01 per link, media included	You have your own extraction pipeline

The full contract

https://know.fast/agent.txt is the canonical agent contract. Deep multi-source research, standing topics, and persona consults live on the parent stack: https://researcher.now/agent.txt (plugin: pip install researcher-now).