Updated on
Andi: reviews and analysis
A free AI search engine with no ads and no account, attributing and linking the source of every answer.
Affiliate link · no extra cost to you
Our verdict
Andi does what this segment’s question asks and does it for free: it answers by citing and linking the source, carries no advertising, requires no account and does not log searches. There is real work underneath in intent classification and extraction, so it is not a text box over somebody else’s model. What it does not sustain is the promise on its home page: the default answer is an extract from a single page rather than a synthesis of several, the selection misfires fairly often, and the accuracy figure it displays comes from a 2024 measurement with the runner-up left out of the chart. Two people run it, the browser extension it promotes has gone untouched since 2024, and the team’s recent work is in neighbouring products.
Best for: The one-off query you would rather not attach to a profile, with the source in view and nothing to pay.
What the internet says
First sweep. Andi is the small project of this segment and it genuinely answers the segment's question: a free search engine with no advertising, no account and no search logging, with the source named and linked in every answer. Verification was run against its live API rather than only its marketing site, returning results dated in 2026 with attribution on all of them. There is real work underneath, with intent classification by query type, recency scoring and its own page extractor, so it is not a wrapper over somebody else's model. The problems are three and none of them is about price. The first is that the default answer is not what the home page promises: instead of a multi-source synthesis by language models, what arrives is an extract from a single page, and in three of nine test queries the chosen extract did not answer the question. The second is size and focus, because the declared team went from ten people in 2023 to two, the browser extension promoted on its home page has not been updated since April 2024, and most of its 2026 posts concern neighbouring products. The third is coherence: its policy states it never uses third-party analytics while the app loads a product analytics platform with a persistent per-visitor profile, served from its own domain, although it trims the query text before sending and the whole thing can be switched off.
What the web repeats in favour
- It is genuinely free with no hidden trade: no card, no account, no ads and no search logging, which nobody else in this segment offers
- Attribution is systematic rather than decorative: the interface prefixes an «according to» with the site name and its link, so checking an answer takes one click
- There is a real product underneath, with intent classification by query type, recency scoring and its own page extraction, plus a search API the company sells separately
What the web repeats against
- The default answer is a single-page extract rather than the multi-source synthesis the home page advertises, and selection misfires often: in three of nine test queries the extract did not answer the question
- The accuracy figure on its home page comes from an external measurement run in August 2024 against the models of the time, is published in the present tense with no date, and the chart leaves out the runner-up, which was the closest rival
- Weak continuity signals: a two-person team against the ten declared in 2023, a browser extension untouched since April 2024 with a thousand users, a paid plan announced in September 2025 that still has no price, and recent work poured into neighbouring products
- Its published policy states it never uses third-party analytics while the app initialises a product analytics platform with a persistent per-visitor profile. It is proxied through its own domain, the query is trimmed and it can be disabled, but the policy and the code do not say the same thing
Sweep sources: Official pricing · Docs · Review sites · Press · Press · App stores · GitHub
Pros / Cons
Pros
- Free, with no account and no advertising of any kind
- Every answer arrives attributed and linked, so it is one click to check
- It does not log searches or use cookies, and trims the query before it reaches its analytics
Cons
- The default answer is an extract from a single page, not the multi-source synthesis its home page promises
- Extract selection misfires often enough to force more checking than you would like
- Two people run it, the extension has not been updated since 2024, and the announced paid plan has had no price since September 2025
TLDR: Andi is a free AI search engine that answers by naming and linking the source behind each reply, shows no advertising, asks for no account and does not log searches. It is the small project of this segment, run today by two people. It works, and for a one-off query you would rather not attach to a profile it is a genuine option. As a main tool it falls short: the answer is usually an extract from a single page, and maintenance is patchy.
What Andi is and how it works
Andi answers questions instead of returning ten blue links, and every answer arrives with the name of the site it came from and a link to check it. It came out of Y Combinator’s winter 2022 batch and had generative answers running months before ChatGPT existed, which in this segment is a piece of history rather than an advantage today.
Two things set it apart. The first is that there is no advertising of any kind. The second is that no account is required, and that one is genuinely unusual: almost every AI search engine asks you to identify yourself by the third query, because writing answers costs money and somebody has to be billed for it. Here you arrive and you ask.
Now for what it actually returns, because the site promises more than it delivers by default. The home page talks about answers synthesised from the best sources by an ensemble of language models. What comes back when you ask is something more modest: an extract from the page it treats as primary, attributed with an “according to” and its link. The AI synthesis does exist, but as a separate action you request with a button to summarise or explain. The distinction matters, because a well-attributed extract is useful and honest, and it is not the same thing as an answer written from several sources.
There is real work underneath, not a text box over somebody else’s model: it classifies the intent of each query by type, scores the freshness of results and pulls page content with its own reader. The company is also building a web search API for developers, with published prices and waitlist access. That API changes nothing for someone who only wants to search, but it does explain where the money is meant to come from, which is exactly the question to ask about a free product.
What using it is like day to day
There is no learning curve and nothing to configure: search box, question in plain language, short answer with its source. You can ask it to summarise or explain a page, and it works across several languages.
Quality is uneven, and that is better shown with examples than adjectives. On specific informational queries it returns what you asked for, with the source in view. On others the choice of extract misfires: ask why the sky is blue and you may get a sentence about what colour the sky was 2.5 billion years ago, and ask who runs a country and you may get a stray line from an encyclopedia’s footnotes. These are selection failures rather than fabrications, and you spot them immediately because the source is right there. They still mean more checking than you would like.
Maintenance is patchy, and this weighs most on the final judgement. The engine responds quickly and its status page reports normal service, but the browser extension it promotes on its own home page has not been updated since 2024, there is no native mobile app, and one of its published policies is still dated late 2024. Looking at what the team shipped in 2026, most of the movement is in neighbouring products rather than the search engine. And the team, which reported ten people in 2023, is now listed at two.
Two more warnings about what the marketing leaves out. First: the home page displays an accuracy percentage it says beats the big players, but the figure comes from an external measurement run in 2024, against the models of that time, and the chart leaves out the runner-up, which was the closest rival. Presented in the present tense with no date, that number misleads. Second: some results carry affiliate links, which its own assistant admits when asked and which appears nowhere in its marketing pages.
Pricing and plans
The search engine is free, with no card and no account, and its own site says so in those words. There is no paid tier unlocking features, no announced query limit, and no ads in the list or in the answer.
The part to watch is the part that does not exist yet. A premium plan for heavy users has been announced with no price and no date, and it has been in that state since September 2025. Its terms of service, updated in 2026, already cover charges, auto-renewal and a rates page that is not published today. The current deal is good and it is not guaranteed.
On privacy the picture needs sharpening, because it is the selling point and it does not hold up whole. The verified part is good: it does not store searches, it uses no cookies, there are no ad networks in the app, and the query text is trimmed before it reaches its analytics. What does not add up is the absolute claim. Its policy states that it never uses third-party analytics, while the app loads a well-known product analytics platform, served from its own domain and keeping a persistent profile per visitor. It can be switched off, it collects far less than any large search engine, and the policy and the code still say different things.
Who it is for (and who it is not for)
It suits the one-off query you would rather not attach to a profile, and anyone who wants an answer with its source in front of them without creating an account or seeing an ad. As a second tool, the one you open for a specific question, it does the job and costs nothing.
It is not for people who search all day for work. There the uneven answers and the lack of depth cost real time, and both Brave Search at $3 a month and Kagi at a higher price go considerably further. It is not for anyone who needs continuity guarantees, integrations or support that answers, because a two-person team neither offers nor promises those. And if what you want is to question your own documents, this is not the segment.
Alternatives to Andi
Inside the segment the comparison comes down to what each one charges and what each one controls. Brave Search has a verified index of its own and sourced answers in its free tier, with ads removable for very little money. Kagi goes to the far end of control and paid privacy, with no free layer at all. Andi is the option where you neither pay nor register, in exchange for a smaller product with fewer guarantees and a less elaborate answer than it promises. The full picture is in the best AI search engines, part of AI assistants.
Frequently Asked Questions
Is Andi really free?
Today it is, with nothing hidden: no card, no account, no ads. The caveat is about the future, because a premium plan announced in September 2025 still has no price and no date, and its terms already provide for charges.
How does it make money then?
Not from advertising. The declared plan is charging for its web search API for developers, which has published prices and runs on a waitlist. On top of that, some results carry affiliate links, which its own assistant acknowledges but which is not disclosed on its site.
Is the answer a synthesis of several sources?
By default no, and this is the gap between what it promises and what it does. What you normally get is an extract from a single page, attributed and linked. AI synthesis is available, but as a separate action you request with a button.
Does it really not track me?
Almost. It does not log searches, uses no cookies, and trims the query text before sending anything to its analytics. It does load a product analytics platform with a persistent per-visitor profile, served from its own domain, while its published policy states it never uses third-party analytics. It can be turned off, and the contradiction is worth knowing about.
Can I rely on it as my main search engine?
With serious reservations. It is a two-person project whose recent work has moved to neighbouring products, the browser extension has gone untouched for two years, and extract quality is uneven. As a free secondary option it holds up. As your only search tool, it does not.