Esin

An assistant that doesn't speak unasked.

Esin ("inspiration"; the root of "esinti", a light breeze) is a panel that reads the page, summarises it and answers your questions by looking at it. Open it with ⌘E.

How it works

You type a question in the panel. The tab you have open is attached by default; you see it as a chip under the panel and can remove it with ×. You can attach up to five tabs together. Page text is read when you send the question; the first 20,000 characters of each page go. You see what was sent (which site, how many characters) next to the answer.

The first time, Esin shows a short note on how it works, and nothing is sent until you say "Got it". Unless you ask, your page does not go to Esin.

Where a request goes

YouQuestion + text of attached pages
VentoOnly when you send
Proxyesin.cansoykanyilmaz.com; the key lives here
LLMTRModel gateway
GreenPTFrance

There is no API key in the browser. Requests go through Vento's own small proxy server, and the key lives only there. That way nobody can extract a key embedded in the browser and drain the quota. The proxy does not pass along whatever the browser sends; it rebuilds the fields it knows, caps message count and length, and applies a per-minute and a daily limit per IP. The proxy does not record prompt or response content; it writes only the time, model, status code, message count and duration.

Models and data

Models the Esin proxy accepts
Model	Status	Data
`greenpt/gpt-oss-120b-eu`	Default	GreenPT, France (Scaleway). No training on inputs. No prompt/response logging at LLMTR or GreenPT.
`qwen/qwen3.6-35b-a3b` `qwen/qwen3.6-flash`	Selectable: Customize ▸ Esin model	The provider may keep logs. That is why they are not the default; when you pick one, the difference is written next to the choice.

According to what LLMTR told us, they keep no prompt or response records on their side (only model, token count and cost). GreenPT models require a data-transfer approval in the LLMTR account; this is managed through Vento's account, and you are not asked to create an account or approve anything.

What is the difference? With the default model we can say page text is stored nowhere. With the Qwen models we cannot: the provider may keep logs. That is why they are not the default.

Daily allowance

There is a paid budget behind Esin's model provider, and it is shared. So Esin has a daily allowance: 30 questions per person per day. It renews at 00:00 Turkey time. The panel shows "Left today: N / 30" underneath; when it runs out, Vento tells you when it renews.

There is also a daily total ceiling for all users; if that is reached, Esin answers nobody for the rest of the day and Vento says so separately. We will review the numbers as usage grows; what is written here is the current value.

To count your allowance the proxy makes a salted digest of the IP address a request came from and keeps it only for that day's counter. The raw IP address is not recorded, and the digest is deleted the next day. A request that never reached the provider does not count against you.

LLMTR

LLMTR is an AI model gateway built by Knowhy.co (llmtr.com). In its early days a small browser needs a dependable default provider. They were open to talking at a point when we had almost no users, agreed to let Esin's default model run through them, and set up a test budget. Thank you.

This is not a sponsor badge; it is a technical fact you should know: what you write to Esin passes through their gateway first. If the provider changes, we change this page.

Open points

If you choose a Qwen model the provider may keep logs; Vento says so at the choice, but the logs are in the provider's hands.
Using your own API key or an on-device model is not in this version.
Esin only reads text; it does not click or fill things in on a page.
Esin's daily allowance is limited (see "Daily allowance" above); once used up, Esin does not answer until it renews.