Web search · sources under every answer

Search the web without sending it your conversation.

Web search is off by default, in the chat and in Agent mode: your messages never leave Europe. If you switch it on, the queries the model writes, and the addresses of the pages it reads, go to Linkup, a French company whose registered office is in Fresnes. Evanesca has not verified where its servers run; its own privacy policy applies. Before it answers, the model can then write a search, read the pages that come back and list them under its answer — never sending your conversation. Each search costs €0.0076 and each page read €0.0015, plus the tokens the model spends reading them.

Checked on 13 Oct 2026

Opening soon: what follows describes the service as it will work.

By default: Offin the chat and in Agent mode
Per search: €0.0076€0.0015 per page read, VAT included
Results per search: 8each cut at 15,000 characters
In Agent mode: 40 per taskweb calls, 300 per 24 hours per account

How it works

You switch it onWeb search is a switch, off by default — in the chat and in Agent mode. Once it is on, the model may search; it judges whether a question needs it. Off, nothing goes to Linkup.
The model writes a queryIt sends Linkup a search in its own words (web_search), or asks for a single page (web_fetch) when it already has an address — one you gave it, for example. Each call is billed: €0.0076 a search, €0.0015 a page.
Linkup sends pages backUp to 8 results, each with its title, its address and the text of the page, cut at 15,000 characters.
The model reads, and may search againUp to 4 rounds of searches for one message; after them, the model has to answer with what it found.
The answer comes with its sourcesThe pages that came back are listed under the answer, with their links, whether or not the model cites them in its text. Each search also shows the exact query that was sent.

What goes to Linkup, and what doesn't

Two tools, and only two kinds of thing sent. The rest of the conversation goes only where your message goes anyway: to the model you chose.

Sent to Linkup

A query, some addresses

The query the model writes, in its own words.
The address of each page it wants to read.

Never sent

Everything else

Your conversation. Your message is not forwarded as such — but the query may repeat its words (see below).
Your attachments, documents and images.
Your name, your email address or your IP address: the requests come from Evanesca's server, under its own account.

Who Linkup is

Company: Linkup Technologies, a French simplified joint-stock company (SAS)
Registered office: Fresnes, France, as its privacy policy gives it
Where its servers run: Not verified by Evanesca; see its privacy policy.
What it receives: The query the model writes, and the address of each page to read — sent by Evanesca's server.
What it keeps: What its own privacy policy and terms say: Evanesca does not control it.
What websites see: Pages are fetched by Linkup, not by your browser: the websites it reads never see your address.

The limits

They keep the model's window — and your bill — within known bounds: without them, a page of 200,000 characters could overflow the window, and would be billed in full.

Results per search: 8 at most.
Text per result or page: 15,000 characters at most; the rest of a longer page is cut.
Rounds per message: 4 at most in the chat, each with one or more searches; then the model has to answer.
In Agent mode: Once web search is on: 40 web calls per task — searches and page reads together — and 300 per 24 hours per account; each call brings back 30,000 characters at most, and counts toward the task's cap.

What it costs

Each search costs €0.0076 and each page read €0.0015, VAT included. On top comes what the model reads — the results, the pages: it enters its context and is billed as input tokens, at the model's input price, like text you paste. Each round of searches is also one more call to the model, and while search is on, the description of the two tools adds a few hundred input tokens to each message.

The most one search can bring back: 8 results × 15,000 characters = 120,000 characters, about 30,000 tokens at 4 characters per token. That search, read once, costs:

ONE FULL SEARCH, READ ONCE · VAT INCLUDED
Model	Where	Input / M tokens	Reading	With the search
Gemma 4 26B	Germany (AKI.IO)	€0.168	€0.0050	€0.0126
DeepSeek V4.1 Flash	Germany (AKI.IO)	€1.008	€0.0302	€0.0378
GLM-5.2	Paris, France	€3.024	€0.0907	€0.0983

The cheapest model for input, a middle one and the dearest, among those that can search. Most pages are shorter than the limit, so the reading usually costs less. A result read in one round is read again in the following rounds of the same answer. Every price

Which models can search

Searching needs a model that calls tools — measured with a real call. Of the 17 models Evanesca serves — all run in Europe — 17 can. In Auto mode, with search on, Evanesca picks one that can.

Mistral Large 4EU region of Mistral AI
DeepSeek V4.1 FlashGermany (AKI.IO)
GLM-5.3Germany (AKI.IO)
Qwen3.8 27BParis, France
DeepSeek V4 FlashParis, France
Apertus v1.5 70BSwitzerland (Infomaniak)
GLM-5.2Paris, France
Qwen3.6 35BParis, France
Gemma 4 26BGermany (AKI.IO)
Mistral Small 4EU region of Mistral AI
Qwen3.5 397BParis, France
GPT-OSS 120BParis, France
Qwen3 Coder 30BParis, France
Qwen3 235BParis, France
Mistral Small 3.2Paris, France
Llama 3.3 70BParis, France
Pixtral 12BParis, France

How answers cite their sources

Under the answer, a “Sources” list gives each page that came back — its title and its link — whether or not the model cites it in its text.
In the conversation, each search shows the exact query the model sent, and the start of what came back.
A listed source is a page the model received, not proof that it read it right: open the ones your decision depends on.
In Agent mode, each search and each page read shows among the task's steps.

Frequently asked questions

Does Linkup see my conversation?

No. Linkup receives the query the model writes — which may repeat words of your message — and the address of the pages to read. Your conversation, your attachments and your account are not sent to it: the requests come from Evanesca's server, under its own account.

What does web search cost?

€0.0076 per search and €0.0015 per page read, VAT included, plus the tokens the model spends reading the results, at its price. Each round of searches is also one more call to the model. The exact cost shows under the answer.

Who is Linkup?

Linkup Technologies, a French company whose registered office is in Fresnes; Evanesca has not verified where its servers run. What it keeps follows its own privacy policy, which Evanesca does not control.

Can it read a page I give it?

Yes, once web search is on. Put the address in your message: the model can read that page with web_fetch, up to 15,000 characters, for €0.0015.

Why didn't it search?

Web search is off by default: switch it on. Once it is on, the model judges whether a question needs the web; ask it to search, or give it the page to read.

Is web search on by default?

No: it is off by default, in the chat and in Agent mode. Off, the model answers from what it learned and what you send it — nothing goes to Linkup.

Does it work in Agent mode?

Yes, once you switch it on — the agent follows the same switch — with its own limits: 40 web calls per task (searches and page reads together), 300 per 24 hours per account, and 30,000 characters per call. Each call is billed and counts toward the task's cap.