Comet browser from Perplexity is a shame

[flackz]

I recently installed CometAPI browser from Perplexity to test its capabilities. To be honest, I didn't expect such a disappointment. The built-in “AI assistant” can't do anything on its own and constantly gets stuck when automatically filling out even simple forms. It tries to insert made-up data into them, ignoring instructions. It constantly asks for confirmation and then displays a message stating that the free usage limit has been reached and offers to purchase a Pro subscription. Subscription to what? To a tool that does nothing but cause harm? Thanks, but no.

 

[IM Dude]

It's a scam. It's perplexity free version with it's own browser while it says its perplexity pro when you sign up. When you first install and run it shamelessly ask you permission for spaying you like most of your device activity. Unbelievable people are promoting this scam.

 

[eagleeye111]

You need to have Pro subscription for full Comet capabilities, free tier is not good enough.

You can find 1 year pro for a few bucks.

 

[buzzbandit]

Having tried it for a week or so now I do concur. For the sheer amount of potential additional security risks it is perplexingly unusable as a product. If I have to babysit a toddler failing continuously at simple tasks I'd rather do them myself

 

[Spicy Chromosomes]

Next time don't create any expectations about those AI browsers, just another slop to attract people using AI hype

 

["Surendra Parmar "]

I had the same experience. It looked promising, but the actual assistant behavior felt super unreliable.

 

[flackz]

The problem is that these companies are trying to attach random text generators to various tools, such as browsers etc. Some even attempt to control their computers or servers with a tool designed to generate plausible text. The primary purpose of language models is auto-blogging, not making decisions for humans. What is now commonly called AI is not actually AI. It is not intelligence in the conventional sense. These models cannot think, much less make decisions. They are designed that way.

Language models are great for autoblogging and programming. They can generate high-quality text in any language, both human and machine. That's all they're designed for. Any other applications are doomed to fail. Therefore, this whole hype bubble will soon burst.

 

[ZenWolf]

Also i have heard rumors of these browsers getting hacked easily. These browsers are at early stages, I don't think anyone can take chrome at this point.

 

[Starblazer]

Unbelievable people are promoting this scam.

They were getting like $20 per each install from USA. It's easy money if you are an influencer with considerable following.

 

[learnandearnwithseo]

I recently installed CometAPI browser from Perplexity to test its capabilities. To be honest, I didn't expect such a disappointment. The built-in “AI assistant” can't do anything on its own and constantly gets stuck when automatically filling out even simple forms. It tries to insert made-up data into them, ignoring instructions. It constantly asks for confirmation and then displays a message stating that the free usage limit has been reached and offers to purchase a Pro subscription. Subscription to what? To a tool that does nothing but cause harm? Thanks, but no.

It is a new tool, will improve with time. If they allow , submitting feedback, you can submit these issues to them.

 

[flackz]

It is a new tool, will improve with time. If they allow , submitting feedback, you can submit these issues to them.

The concept of the browser itself prevents its improvement. All of these language models from OpenAI, Anthropic, xAI, and so on, are merely statistical, random text generators. They can't perform actions or make decisions. They just continue writing a given text endlessly. That's all they're designed to do.

 

[appnetmedia]

That guy raised million dollar investment and gives too much hype on facebook ads for it as "google competitor" but it is nowhere.

 

[VAT]

The problem is that these companies are trying to attach random text generators to various tools, such as browsers etc. Some even attempt to control their computers or servers with a tool designed to generate plausible text. The primary purpose of language models is auto-blogging, not making decisions for humans. What is now commonly called AI is not actually AI. It is not intelligence in the conventional sense. These models cannot think, much less make decisions. They are designed that way.

Language models are great for autoblogging and programming. They can generate high-quality text in any language, both human and machine. That's all they're designed for. Any other applications are doomed to fail. Therefore, this whole hype bubble will soon burst.

It does not work like that, it isnt some text generator hooked to a browser. The underlying is AI capable of reasoning and math, the core logic is not rule based.

 

[conradellis]

I have the Pro version, but I didn't like it at all. Poor in my opinion.

 

[flackz]

It does not work like that, it isnt some text generator hooked to a browser. The underlying is AI capable of reasoning and math, the core logic is not rule based.

No, none of the current AI models are capable of real reasoning and decision-making. They are built on the transformer architecture invented in 2017, which is designed primarily for sophisticated statistical pattern matching in text, not thinking or logic at all. These models process input as sequences of probabilities to generate human-like text. However, they do not actually "understand" or "reason" in the common human sense. They only produce responses based on patterns found in massive datasets they were trained on. Think of them as an extremely advanced version of T9 autocomplete rather than true intelligence. Any apparent capability in math or logic is always an imitation, not a genuine mental process.

Today's "AI" is merely a sophisticated random text generator, not a thinking entity. Any claims of genuine reasoning or decision-making ability are mistaken. These models are great for auto-blogging and tasks like that, but not for making decisions.

 

[CCIS]

Next time don't create any expectations about those AI browsers, just another slop to attract people using AI hype

There's gotta be a way to program an ai to be much more capable and takes the tasks of all the low level tasks. also im thinking since agentic browsing is new with subsequent updates agentic browser will be come much more effective and understands how to browser better, After all all ai needs is just more data.

 

[Saadii]

I recently installed CometAPI browser from Perplexity to test its capabilities. To be honest, I didn't expect such a disappointment. The built-in “AI assistant” can't do anything on its own and constantly gets stuck when automatically filling out even simple forms. It tries to insert made-up data into them, ignoring instructions. It constantly asks for confirmation and then displays a message stating that the free usage limit has been reached and offers to purchase a Pro subscription. Subscription to what? To a tool that does nothing but cause harm? Thanks, but no.

CometAPI browser’s AI assistant is disappointing—it struggles with forms, makes up data, repeatedly asks for confirmation, and blocks free usage, pushing a Pro subscription for little value.

 

[donelyardi]

I recently installed CometAPI browser from Perplexity to test its capabilities. To be honest, I didn't expect such a disappointment. The built-in “AI assistant” can't do anything on its own and constantly gets stuck when automatically filling out even simple forms. It tries to insert made-up data into them, ignoring instructions. It constantly asks for confirmation and then displays a message stating that the free usage limit has been reached and offers to purchase a Pro subscription. Subscription to what? To a tool that does nothing but cause harm? Thanks, but no.

Your experience matches what many people report - unreliable agent behavior + constant upsell = bad combo.
AI automation should simplify workflows, not break them.

If you’re still open to testing other approaches, I can recommend one tool that focuses on privacy + stable agent behavior. It’s called Sigma Browser, and the main thing people like about it is the complete anonymity of prompts and much smoother task handling.

 

[saphiron]

I recently installed CometAPI browser from Perplexity to test its capabilities. To be honest, I didn't expect such a disappointment. The built-in “AI assistant” can't do anything on its own and constantly gets stuck when automatically filling out even simple forms. It tries to insert made-up data into them, ignoring instructions. It constantly asks for confirmation and then displays a message stating that the free usage limit has been reached and offers to purchase a Pro subscription. Subscription to what? To a tool that does nothing but cause harm? Thanks, but no.

That’s the main issue with using LLMs for browser automation right now. They hallucinate data and struggle with specific DOM elements. For reliable form filling, hard-coded scripts are still miles ahead of these 'AI assistants'.

 

[cg019]

TBH Perplexity itself is pretty bad (even the good models, they just don't feel as good as the APIs or the official websites)