Why local-first
Two reasons, and neither of them is a trend.
Privacy, by architecture not by policy
A cloud scraper has to receive the page in order to parse it. Magpie’s engine runs on your machine, so the page comes from the site to you, and goes on to a model only when you ask, on your own key. We cannot hand over what we never had — and you don’t have to take a privacy policy’s word for it, you can watch the network.
your machine → target site directpage data → this disk · your llmmagpie servers 0 hops
Priced like a tool, not infrastructure
Credits exist because the seller pays per page. We don’t: your machine fetches, and your own key pays the model. So the price is $49 a month whether you scrape ten pages or ten million — and the free tier is free because it costs us nothing to serve.
$49.00 / month · flatcredits none overage nonepages unmetered
Who builds it
Small enough to answer your email.
I build Magpie on my own, as Motherlode Lab: the code, the releases and the support inbox are all me.
built in go · wails · chrome devtools protocol · sqliteships for windows · macos · linuxshipped v0.1.31 · oct 11, 2026
[email protected]
What we optimise for
Three things we will not trade.
Local by default
Every engine runs on your machine first. Cloud isn’t an upgrade path we’re saving for later; it’s the thing we chose not to build.
default localcloud option none
Priced like a tool, not infrastructure
One flat $49 a month, and a free tier gated by capability instead of volume. A page cap would be credits under a different name.
$49.00 / month · flatfree tier capability-gated
Honest about competitors
Firecrawl and ScrapingBee are good at something we’re not: running with no machine switched on. If you need scheduled scraping and nobody logged in, use them — we’d rather say so than sell you around it.
cloud api → scheduled · no desktopmagpie → your machine · your keys
Still shipping.
Full changelogv0.1.31
LatestForgetting a research run deletes every copy of the pages it read, and Settings gains plans and licence keys.
v0.1.30
New installs keep their data in the OS app-data folder; browser rendering uses your own Chrome or Edge.
v0.1.29
A failed extraction still returns the page, a cancelled scrape stops at once, and an aborted crawl keeps its counts.