AI search with answers and source links — for your website

Managed AI search for your website. A visitor asks a question in plain language — the widget replies with a coherent answer with links to sources on your site, in the visitor's language. Under the hood: Google Vertex AI Search and Gemini.

Regular search returns a list of links. Here the visitor gets a ready answer with citations and links to the original page. If the answer isn't on your site, the service says so honestly instead of making things up.

In the language of the question. A visitor can ask in their own language about content in another — the answer comes back in the question's language, with links to the source pages.

Yes — documentation, scientific papers, standards, engineering and patent material. Answers always carry links so the source can be verified.

One line: paste a <script> tag with your key into your site. By default the widget is a floating "Ask AI" button (bubble mode); for an inline block, add a <div id="ai-search"></div> to the page. Everything else — mode, theme, texts — is configured in the dashboard without touching your site's code.

Three: a floating "Ask AI" button, an inline block on the page, and attaching to your existing search field. You pick one of the three in the dashboard and style it.

No. The script is lightweight, dependency-free, and loads asynchronously. Requests to the AI go to our proxy only when a visitor actually asks something.

No. We index public pages with our own crawler — no domain verification required. Just enter your site's address.

After payment and initial indexing: the service crawls the site, builds the Vertex index and a topic map. Progress is visible in the dashboard in real time. After that, the index is kept fresh automatically.

For a regular site — no: our crawler AISearchBot comes in on its own and indexes everything. Setup is only needed if your site sits behind protection (Cloudflare, a WAF, a corporate firewall, "Bot Fight Mode", and the like) — it may block bots, and then the site risks not being indexed. Not sure — just pass the details from the next item to your IT specialist or your hosting support.

Allow crawling in robots.txt: a line User-agent: AISearchBot, then Allow: /. The bot always arrives with the User-Agent AISearchBot/1.0 (+https://mvp-search-k3gecv3cwa-uc.a.run.app/bot; managed site search) — protection rules match it by that substring. In Cloudflare: WAF → Custom rules → Skip on the condition "User-Agent contains AISearchBot", and turn off Bot Fight Mode / Browser Integrity Check for it. In other firewalls — add the bot to the whitelist / good bots by the same User-Agent. The bot's static IP is 35.223.115.47: you can add it to exceptions both by the User-Agent AISearchBot and by this IP.

Two plans: Start — $15/mo ($144/yr) and Pro — $39/mo ($374/yr), each with a fixed volume of searches included. Queries above the limit are $0.02 each. No hidden charges.

One visitor question that the service generated an answer for. Follow-up questions in the same conversation are separate queries. Service and blocked (spam/abuse) requests are not billed.

By default, queries above the limit are billed at $0.02. You can set a hard cap — then, once the budget is spent, the widget gently pauses until the next period or a top-up. There's denial-of-wallet protection: rate limits prevent malicious budget draining.

Subscription and top-ups go through an external payment provider. We don't store card details. The dashboard shows your balance, payment history, and double-entry accounting of costs.

Paid periods and already-consumed queries are non-refundable except where required by law. You can cancel the subscription at any time — it simply won't renew for the next period.

Prices are in US dollars. Payments are handled by Paddle as Merchant of Record: applicable taxes (VAT, etc.) are calculated and remitted by them based on your jurisdiction at checkout.

The queries included in your plan reset each period and do not roll over — the unused remainder expires. Need headroom beyond the plan — buy a block of +1000 queries for $20: it goes into your wallet and does not expire. The subscription allowance is spent first, then the wallet queries.

On Google Cloud infrastructure, in the selected region. Indexing, search, and answer generation are performed by Google's services (Vertex AI Search, Gemini).

No. We do not train models on your content or queries. Under its enterprise terms, Google does not use data submitted to Vertex AI and the Gemini API to train its models.

The architecture is built for it: the site owner is the controller of visitor data, we are the processor, and Google is a sub-processor under its Data Processing Addendum. A full Policy and DPA are being prepared for launch. Details are in the Privacy Policy.

The query text is sent to Google to generate an answer and used for quality metrics. It is not sold to third parties and not used to train models. Retention periods are in the Privacy Policy.

Queries are sanitized (prompt-injection protection), with rate limiting and strict security headers. Cheap checks run before any paid call, which also protects against denial-of-wallet.

Yes. Each project runs against its own index and key; answers are grounded only in that site's content.

Google Vertex AI Search (Discovery Engine) for search over the index, and Google Gemini for answer generation. We are a managed layer on top of that technology: widget, indexing, dashboard, billing.

We provide access to Google's technology and take on all the hard wrapping: crawling and keeping the index fresh, multilingual answers with citations, abuse protection, billing, and a dashboard. The "intelligence" is Google's; the value is a ready, configured, maintained solution.

Answers are generated by Google's models. We do not control or guarantee their specific content. We tune the pipeline (citations, an honest "no answer" when there is none), but AI output should be verified — see the Terms.

The service depends on Google Cloud availability. We are not responsible for outages or changes on Google's side, but we work to minimize their impact. Google's commitments apply directly under its terms.

The service runs on Google Cloud's managed infrastructure. We do not provide a separate availability guarantee (SLA) unless agreed in a separate contract; the service is provided "as is".

Roles are split the way GDPR expects. The site owner is the controller of their visitors' data and their content: they decide why it is processed. We are the processor, acting on their instructions. Google Cloud is a sub-processor (indexing and answer generation in Vertex AI and Gemini). For your own account data — email, payments — we are the controller. Separately: we also use query texts to a limited extent for our own purposes — quality metrics and abuse protection; in that narrow part we act as an independent controller. Details are in the Privacy Policy.

On Google Cloud infrastructure, in the chosen region (for EU clients — the EU by default, to minimize cross-border transfers). Search, indexing, and answer generation run on Google services (Vertex AI Search, Gemini) under their Cloud Data Processing Addendum. The service operator is established in Israel: EU-to-Israel transfers rely on the European Commission's adequacy decision, and transfers to Google's services (US) rely on Standard Contractual Clauses and/or the EU-US Data Privacy Framework. We do not store card details — payments are handled by Paddle as Merchant of Record.

The question text is sent to Google to generate an answer, and is used by us to a limited extent for quality metrics and abuse protection. It is not sold to third parties and not used to train models — neither by us nor by Google under its enterprise terms. Query text is kept for a limited period and then deleted or anonymized; the IP address itself is not stored — it is processed only transiently in memory for abuse protection. Specific retention periods are in the Privacy Policy. Tip: don't ask visitors to enter sensitive personal data into the search.

The architecture is built for these requirements: clear controller/processor/sub-processor roles, processing on instructions, no training on your data, limited retention periods. By launch we provide a Data Processing Agreement (DPA), accepted at onboarding, which also covers the CCPA "service provider" requirements. We cannot contractually override mandatory GDPR/CCPA obligations — liability is allocated by law. This is not a claim of "full compliance in all cases": final documents are being prepared with a lawyer.

It depends on whose data it is. For visitors' data, the request goes to the site owner as the controller — we, as the processor, assist them: we help locate and delete data, support log clean-up and project deletion in the dashboard. For your account data (email, payments), contact us directly via the contact form. We handle such requests within the timeframes set by applicable law.

Answers are generated by Google's AI based on your site's indexed content. They may be inaccurate, incomplete, or out of date, even when they look reliable, and are not professional advice — you should verify them before use. In the widget interface we indicate that the AI is answering. We tune the pipeline (source links, an honest refusal when there is no answer), but we do not control or guarantee the specific content the models output. You are responsible for your site's content that serves as the source. More in the Terms of Service.

The service is provided "as is", without warranties of accuracy, uninterrupted operation, or fitness for a particular purpose — as is standard for AI services. Liability is limited to the extent permitted by law.

In the Terms of Service and Privacy Policy. These are pre-launch drafts — final versions are published before launch.