Brello vs Perplexity: two ways to answer from the web

At a glance

Brello 1.0 and Perplexity both answer from the web with numbered citations. They differ in where the searching, reading and writing happen, and in who receives your question.

How each finds and reads sources

Both search the web and answer with numbered citations. Brello 1.0 does the searching, reading and writing on your phone; Perplexity receives your question and does that work on its side, drawing on its own search index.

Brello 1.0 first decides whether a question needs the web at all. It skips small talk, arithmetic and writing tasks, then rewrites the question into a search text: “What’s new in AI this week?” becomes “latest AI news October 2026”. The phone tries DuckDuckGo and then Bing in an invisible browser, and falls back to direct requests to DuckDuckGo, Google News (for time-sensitive questions only), Brave Search, Bing and Wikipedia until one returns relevant results. The top six results become sources [1] to [6].

It fetches up to four of those pages in parallel, each with a 7-second timeout and a 1.5 MB cap, strips navigation, ads, cookie banners and comments, and keeps up to about 24,000 characters of the main text per page. It splits that text into passages of about 420 to 700 characters, ranks them with BM25 (k1 = 1.2, b = 0.75) plus small bonuses for covering more of the question’s terms and for lead paragraphs, and keeps at most two passages per source, within 3,400 characters for the Gemma 4 models or 2,600 for Qwen3. The model on the phone reads that numbered block, labelled “Web results” with the retrieval date, and writes an answer whose inline citations, such as [1], open the source. Answering from the open web, without a server describes each step, and BM25, explained covers the ranking.

Perplexity describes four steps: understanding your question; searching the internet for information from sources such as articles, websites and journals; compiling the most relevant information into an answer; and citing sources, so that “each answer includes numbered citations linking to the original sources”. It uses language models “like GPT-5 and Claude 4.6 Sonnet”, and Pro Search lets you pick the model, with “Best” as the default.2 Its crawler, PerplexityBot, “indexes pages similarly to other search engines” and respects robots.txt, and Perplexity “partners with third-party crawlers to help build our search index”.3

The two designs make different trade-offs. Brello reads at most four pages per question and writes with a model small enough to run on a phone, so each answer rests on a narrow slice of the web. Perplexity’s Research mode, by its own description, “performs dozens of searches automatically, reads hundreds of sources” and delivers a report “in just 2-4 minutes”.2 Retrieval-augmented generation, explained covers the general technique of answering from retrieved text.

Brello can also answer without the web. After the one-time model download it works offline, web search stays off unless you turn it on or approve it, and if a search fails or comes back empty, Brello answers from its own knowledge. None of the Perplexity pages we checked describes an offline mode.

Who sees your query

With Brello, the search engine and the websites receive requests straight from your phone, and no Brello server sees anything. With Perplexity, your question goes to Perplexity, which works with model providers under agreements it describes.

When Brello searches, the search engine that answers receives the search text, sometimes with up to 10 words of the previous question added on a follow-up, and up to four websites receive ordinary page requests. These requests go directly from the phone, carry Do Not Track and Global Privacy Control headers, and show the phone’s IP address, as any web request does. Source cards use letter marks drawn on the phone, so no request goes out for site icons. With web search off, a question that looks time-sensitive brings up a card titled “Search the web for this?” with two buttons, “Search the web” and “Answer offline”; Asking before going online explains that choice.

Perplexity’s privacy notice says it processes “prompts, queries, uploads, and any other data you choose to provide”, and that its services “may collect your general location from the IP address from which you connect” for location-specific queries. It also records device data, including your IP address, and shares data with service providers that help run its services. Submitting data means agreeing to its processing outside your country, “including, but not limited to, the United States”. The current notice clarifies that Perplexity does not “send your queries, prompts, or conversation content to advertisers”.1 For the language models, Perplexity says its “agreements with third-party model providers like OpenAI and Anthropic prohibit using Perplexity data for training their models”, and that data processed through Perplexity “is not retained or used to further train external AI models”.4

Brello puts no company between you and the search engine, but the engine and the websites see your phone’s IP address. With Perplexity, one company receives your whole question and the details its privacy notice lists, and its pages describe the limits it sets on model providers. How AI assistants handle your data explains these patterns in general terms.

What is kept, and for how long

Brello 1.0 keeps chats only on the phone, and search results only in memory, for 15 minutes. Perplexity keeps personal information while your account is active and removes it within 30 days of account deletion, with legal exceptions.

Brello writes chats, including any “Thought process” text, to one file, conversations.json, in its private app storage, excluded from Android cloud backups and device-to-device transfer. Search results are cached in memory for 15 minutes to avoid repeat lookups and are gone when the app closes, and the invisible search browser is torn down after 3 idle minutes, with its cookies, cache and local storage wiped. You delete a chat by swiping left on it, or every chat with “Delete all chats” in Settings. Nothing is synced, so if the phone is lost or reset, the chats are gone.

Perplexity says: “We retain your personal information as long as your account is active. If you delete your account, your personal information will be removed from our servers within 30 days.”5 Deleting the account from settings disables it while all account and personal data is “permanently deleted within 30 days”, and signing back in during that time cancels the deletion. You can also delete a single thread or all threads, and uploaded “files are kept for 30 days by default”.6 The privacy notice sets the general rule: Perplexity keeps personal data “for only as long as necessary to fulfil the purposes in this Notice unless a longer retention period is required or permitted by law”. On deletion requests, it says “in many cases we must keep your personal data to comply with legal obligations, resolve disputes, enforce agreements, or for other business purposes”.1

Model training settings

Brello 1.0 can’t train on your chats, because they never leave the phone. Perplexity may use data for AI training through its “AI data retention” setting, which is on by default for Free, Pro and Max users and can be turned off.

Perplexity says: “For Free, Perplexity Pro and Perplexity Max users, AI Data Retention is enabled by default.” The toggle is under Account settings › Preferences, and “if the toggle is enabled, your data is being collected for AI training”. Opting out has limits: “Opt-outs only apply to data collected after the opt-out date”, and “Previously collected training data cannot be deleted or removed”. With the setting off, data “may still be processed for service operation, legal compliance, and product improvement”. Perplexity Enterprise data, it says, “is never used for AI training purposes”.7 The privacy notice lists among its uses to “improve or create services and products, including our AI models”.1

Two further statements set the scope. Perplexity’s agreements stop outside model providers such as OpenAI and Anthropic from training on Perplexity data.4 For the websites it indexes, Perplexity says it “does not build foundation models”, so their content is not used for “AI model pre-training”.3

Brello 1.0 has no account system, no analytics and no crash reporting, and no Brello server receives your questions, so there is nothing for us to train on.

Accounts, history and Incognito

Perplexity can be used without an account, and offers an Incognito mode that doesn’t save search activity across sessions. Brello 1.0 has no accounts and no incognito mode; its chat history stays on the phone.

Without signing in, you can use “some but not all features” of Perplexity; you need an account for Pro Search, to “upload up to three files per day” and to save your sessions. You create one with an Apple or Google account, or with an email address.9 Accounts can “save query history”. Incognito, available in the app and in parts of the website, “increases your privacy by not saving your search activity across use sessions”,1 and “all incognito sessions expire within 24 hours and are not recoverable”.8

Brello 1.0 saves every chat automatically on the phone and lists it in a history drawer titled “Chats”, grouped as Today, Yesterday, Previous 7 days, Previous 30 days and Older. There is no sign-up, and no chat is kept anywhere else.

What Perplexity offers that Brello 1.0 doesn’t

Perplexity offers features that Brello 1.0 lacks. These are the differences its own pages describe.

• Deeper research. The Research feature “performs dozens of searches automatically, reads hundreds of sources” and delivers a report in minutes.2 Brello 1.0 reads at most four pages per question.
• A choice of models. Pro Search lets you choose the model, including OpenAI’s GPT-5 and Anthropic’s Claude 4.6 Sonnet.2 Brello 1.0 offers three models small enough to run on a phone.
• Its own search index. PerplexityBot and third-party crawlers build Perplexity’s index.3 Brello 1.0 depends on public search engines answering requests from a phone.
• File uploads. A free account includes three file uploads.9 Brello 1.0 takes one photo per message and no files.
• Saved sessions. Signed-in users can save sessions,9 and accounts can save query history.1 Brello 1.0 has no sync, backup or export.
• More places to use it. Perplexity runs on its websites, in mobile apps and in its Comet browser.1 Brello 1.0 runs on Android phones and iPhone.

Where Brello is weaker

Brello 1.0 is a prototype with small on-device models, and its web answers rest on fewer sources.

• Fewer sources per answer. At most four pages are read, and at most two passages from each reach the model, within 3,400 characters for the Gemma 4 models and 2,600 for Qwen3.
• Smaller models. On-device models are far smaller than frontier cloud models: they can be wrong, have a knowledge cutoff and are weaker at long or complex reasoning. Brello’s system prompt asks the model to admit uncertainty, which is no guarantee.
• Web search depends on others. Third-party engines can take up to about 14 seconds per attempt, block the request or return nothing, and Brello then answers from its own knowledge.
• Your IP address is visible. Search engines and websites see your phone’s IP address. Brello doesn’t hide it, and adds no identifier of its own.
• Short memory. A 4,096-token context window; each answer sees only the last six messages, clipped.
• Availability and hardware. On Google Play and the App Store. On Android it needs Android 8.0 or newer on a 64-bit ARM phone and a one-time model download of 977 MB to 3.66 GB.
• No sync, backup or export. If you lose or reset the phone, the chats are gone.

Other comparisons

We answer the same questions for four other assistants, each from its provider’s own pages.

• Brello and ChatGPT: training, Temporary Chat and who sees your searches
• Brello and the Gemini app: activity storage, human review and Gemini Nano
• Brello and Claude: model training and how long Anthropic keeps chats
• Brello and Microsoft Copilot: where your questions go

Sources, with the dates we checked them

We opened each page ourselves on 5 October 2026. Where a page shows its own date, we give it too.

Facts about Brello describe Brello 1.0, version 1.0.0 (build 1), as of 5 October 2026. Product names are trademarks of their owners; Brello isn’t affiliated with them.