Streams as it writes
Sources land in about 100 ms, the first words just under a second, so nobody watches a spinner for ten seconds. Stop any answer mid-flight.
One AI assistant, MDriven colours, no mode picker and no forms to fill in. Answers stream in from the Brain API with real citations. Open it from the bubble in the bottom-right corner to try it out.
Sources land in about 100 ms, the first words just under a second, so nobody watches a spinner for ten seconds. Stop any answer mid-flight.
Inline [1] markers become links to the exact wiki page or video the claim came from, listed as chips under every answer.
Every token is HTML-escaped before formatting, links are restricted to http/https, and only the public-tier key ever reaches the page.
Full-bleed on phones, chrome-trimmed on short laptops and landscape, roomier on large displays — resize this page and watch it adapt.
One tag, anywhere on the page. No dependencies, no iframe, no CSS bleed into the host site.
<!-- MDriven Assistant --> <script src="https://chat.mdriven.net/mdriven-loader.js" defer></script>
Taken directly from the tokens published on mdriven.net.
The widget calls the Brain API directly from the browser on the public tier.
POST https://brain.demotwo.site/ask/stream Authorization: Bearer fTKeUoHu… // public tier, safe in page source body: {"q": "…", "k": 6} event: sources ~100 ms -> citation chips event: delta ~0.9-1.7 s -> appended verbatim event: done 4.5-11 s (cached repeats: ~3 ms)
Public tier caps k at 8, disables deep, and allows 30 questions per
5 minutes per IP — a 429 is shown as a plain “please wait a minute” message.
The API keeps no history, so short follow-ups carry the previous question forward automatically.