magpie Install Local store Source Support Run it locally

Remember any page,
without handing it over.

A model on your own GPU reads the page and writes a few sentences. Only those sentences reach your memory layer.

Install for Chrome Source

Not on the Chrome Web Store yet. Two-minute install below.

a web page on your machine your memory
Taken from the page, condensed here, carried to memory.
Works with

Your memory layer, whichever it is.

Anona Memory POST /v1/record

Lists your spaces so you pick one.

Sign up at memory.anonalabs.com ↗ Mem0 POST /v3/memories/add/

Lists the users it already holds.

Sign up at mem0.ai ↗ Supermemory POST /v3/documents

Re-remembering updates, never duplicates.

Sign up at supermemory.ai ↗

Adding another is one file. No telemetry, and no magpie server. Your key goes straight from the browser to whichever you chose.

Or none of them: keep it on this machine. No account, no key, no network.

What is sent

The same article, two ways.

Summary345 characters
The chapter compares storage engines built on log-structured merge trees against those built on B-trees. LSM-trees write sequentially and compact in the background, giving higher write throughput; B-trees update pages in place, giving more predictable read latency.

All of it. That is the whole thing that gets sent.

Page text93,288 characters

270× the summary, one square each.

Summary is the default. Page text is opt-in, and providers bill on what they extract.

Or no service at all

Keep it on this machine.

magpie-local is a small program that holds your captures in a SQLite file on your own computer, searches them, and lets Claude read them over MCP. There is no account to make and nothing leaves the machine.

Install it

npm install -g @anona-labs/magpie-local

magpie-local install starts it now, and with every login after.

Let Claude read it

claude mcp add --scope user magpie -- magpie-local mcp

Three read-only tools: search what you saved, open one page, list what is recent.

Point magpie at it

In the popup choose This machine and paste the token it printed.

Look at it

magpie-local ui

Everything you saved, in a page on 127.0.0.1: search it, read it, forget it.

magpie reads the page your GPU summarises it in the browser POST magpie-local chunks the text embeds it on your CPU 127.0.0.1, never the network memories.db full text and summary text index and vectors one file, yours Claude over MCP read-only the popup searches the same file
Nothing in that row leaves the machine: not the page, not the summary, not the question you search it with. The only thing that ever touches the network is the model, downloaded once.
What it keeps

The page, its summary and the full text. A memory layer is billed on what it extracts, so magpie sends it a summary; a file on your own disk has no bill, so the source is kept and indexed.

How it searches

Keyword and meaning together: SQLite’s own text index and a small embedding model running on your CPU, fused into one ranking. Every result says which of the two found it.

One machine, no sync, no sharing. It finds what you saved; it does not reason across it. That is what a memory layer is for, and you can change your mind later without losing anything. Full documentation: every command, how search works, and what to do when it will not start.

Install

Four steps.

Build it

git clone https://github.com/anonalabs/magpie

npm install && npm run vendor:wasm && npm run build

Load it

chrome://extensions → Developer mode → Load unpacked → pick dist/

Add a key

Open the popup, choose your memory layer, paste an API key.

Press Alt Shift M

The first capture downloads the model once, then it is cached.

Models

Pick one that fits your GPU.

ModelVideo memoryContextFor
Qwen2.5 0.5B945 MB4,096integrated graphics
Qwen 2.5 1.5B1,630 MB4,096a good default
Llama 3.2 3B2,264 MB4,096best summaries

Also reads PDFs and Google Docs, Sheets and Slides, and keeps a highlighted passage exactly as you selected it.