What we measured, and how.
These results come from the benchmark suite in the Memora Browser repository. Every number on this page can be reproduced with one command.
37%
Fewer tokens on average than the raw HTML of the four test pages
~4 ms
Average time to parse a page and build its action map and Markdown
~84 MB
Total memory with 10 test pages open (resident set size)
Compared with Chrome.
Both browsers loaded the same 200 local test pages one after another in a single tab, on the same laptop. Memory was read with ten tabs open.
11x faster
Time to load 200 pages
11x less memory
Memory with 10 tabs open
200 local test pages, Windows 11 laptop, median of three runs, October 2026. Timed after each browser had started. Chrome is Playwright’s headless shell.
Fewer tokens per page.
Estimated tokens in the raw HTML compared with Memora’s action map and Markdown for the same page. The savings on your pages depend on how much markup they carry.
| Test page | Raw tokens | Memora tokens | Reduction | Actions | Total time |
|---|---|---|---|---|---|
| Login portal | 836 | 502 | 40% | 16 | 9.79 ms |
| Product grid | 1,494 | 941 | 37% | 32 | 3.85 ms |
| Financial data table | 1,155 | 697 | 39.7% | 20 | 4.25 ms |
| Settings form | 910 | 609 | 33.1% | 18 | 2.25 ms |
Memory that grows slowly.
Memora Browser runs as one Node.js process. Opening more pages adds very little on top of the engine itself.
| Measurement | Resident memory | What it means |
|---|---|---|
| Engine at start | 79.6 MB | Resident memory of the Node.js process after loading the engine |
| With 10 pages open | 84.4 MB | Same process with ten complex test pages loaded |
| Each extra page | ~0.5 MB | Average increase per additional page |
Page fork timing.
Twenty forks of one test page with 200 links and 200 buttons. Fork time grows with the size of the page.
4.9 ms
Median
2.5 ms
Fastest
28.4 ms
Slowest
Run it yourself.
From the root of the memora-browser repository:
$ npm run benchmark$ npm run benchmark:chrome- Machine: Windows 11 laptop, Node.js 22, 30 October 2026
- Chrome comparison: 200 local pages in one tab, timed after start-up, median of three runs
- Tokens are estimated from text length, the same way for raw HTML and Memora output
- Memory is resident set size (RSS) reported by Node.js
- Each test page is parsed, its scripts run, then the action map and Markdown are built
Test it on your pages.
Point Memora Browser at the sites your agents use and compare the output yourself.