Horse Browser Review

[Dormouse]

What is Horse Browser?

"The internet browser that's designed for research. Organise notes and websites into projects, have all your work in one place, and get more done."
"Browse the internet with a simple sidebar and organise all your pages, tasks and projects into 'Trails®'. Trails® are nested groups of pages that capture the natural flow of your internet travels.
Yes, that means no more tabs, bookmarks or history.
Unfold, organise, name and add emojis to simple, meditative Trails® that fold away in a click."
"the browser built for professionals who demand focus and efficiency"

I use many browsers, and usually three or four of them are open. I have always found them pretty poor for focused research, so the premise made sense to me. I had already decided to set aside one browser specifically and it was almost easier to switch to something new. The key features required for research are maximising focus and keeping detailed records; most browsers enhance access to distractions and make record keeping effortful.

How does Horse work?

The major innovation in Horse is a vertical outline instead of tabs (horizontal or vertical). This is genuinely helpful. Outlines allow folding. The search process is easily visible through the trails and subtrails in a way that will already be familiar. The width of the left outline pane can be moved so that the details of each page can be read quickly, making it easy to leap back in even after being away for a few weeks. Pages can be renamed and moved around the outline.

Plaintext notes can be placed at any point in the outline to form a record of comments and conclusions.

And any point in the outline (with children) can be exported into a markdown outline (the notes are pure text anyway, but the web pages are exported in markdown format) so that there is a full record of the search process and conclusions.

Three extensions are incorporated - Ghostery, Dark Reader and Saddlepack (Horse's own password manager). Ghostery and Dark Reader work well and enhance focus. Saddlepack stores passwords but doesn't autofill; there's an intention to enable external password manager extensions, though I don't know if that would enable all Chromium extensions. Since most of the sites I need for research require accounts and logons, this is a major friction for me.

Notes Friction

There are many keyboard shortcuts for trails and web pages: you can expand, shrink, go up, down, left, right, in, out and do the hokey cokey (pokey in US). And admittedly the shortcuts can be complex 4 key combos; but you can edit the keys - though I've never persuaded the changes to save.

But there are no shortcuts for notes. None. At all. The move keys work (at least for move left and move right), but to create a new note you have to go to the tiny icon at the top of the left pane and click on it; this creates a new note at the bottom, which you can then move into position (with mouse being easiest). To directly create a new note as a subnote (the most frequent need), you need to click on the tiny hamburger on the current tab and then select New Note from the menu.

These are not UI design choices by someone who uses notes extensively. Most people do their browsing using their mouse predominantly; virtually all users write notes using their keyboard.

Pricing

The price is widely advertised as $60 annual subscription. In the past it has had lifetime subscriptions, and higher and lower prices (usually lower); changes appear unusually often.

When I looked at my own receipt, I noted that I had been charged £60 (as compared to an expected £45 to reflect the current exchange rate). Looking through their FAQs I noted a reference to a price of €60, so I suspect that's what many users in Europe will be charged.

Customer Support

There's the Horse Discord and emails to Support.

When I enquired about support for keyboard shortcuts for notes, I was directed to the list of keyboard shortcuts. When I pointed out that there was nothing for notes, my access to the Discord was removed. I had wondered why it was such an empty discord compared to many. That wasn't a great response, but oh well; Lattics has a long abandoned Discord and there's still active support through the app.

I emailed support about my overcharging. Firstly requesting a refund of the amount overcharged and then another suggesting a four month licence extension as an alternative (I thought that might be simpler for them). I received two email replies from their AI support bot. As is the way with AI, it invented it's own question and then answered that. It confirmed that the price was $60; it understood my concern about the pricing difference and pointed towards the full refund in the first thirty day option; "Unfortunately, our sources don't specifically address extending subscription end dates as an alternative to refunds"; it asked whether that helped or whether I wanted to talk to a person. I clicked the talk to a person button and emailed. There has been no response.

My personal experience has been of customer evasion rather than customer support.

So, where next?

My little trip with Horse highlighted the value of a full record of my research journeys. It's helpful to know what I'm looking for and what strategies I have pursued. Of particular value is being able to revisit all the dead-ends and frustrated searches because they are the ones I am likely to want to return to when I have fresh ideas about what else I can do. Horse is actually not very good for this - it's lists of places visited rather than anything else.

I thought of Edge: I recalled mentions of it being good for vertical tabs and having notes; but I wasn't enthused. It's a long time since I used Opera. But I knew that Notes were quite a big feature in Vivaldi.

Horse has an article of the best browsers for research. Critiques Vivaldi - "Known as the most customisable browser out there, Vivaldi can be fine-tuned to suit you. That said, customisation takes time and can also end up in a distracted user experience"; "Free… But, again, there’s no free lunch here… While Vivaldi doesn’t track its users or sell data to the same extent as some other browsers, it does earn revenue through search engine and bookmark partnerships". My own experience of Vivaldi is that I pay no attention to the bazillion features that don't interest me; it's not entirely positive, as I've had issues with some webpages, but it floats in and out of being my most used browser.

So I determined to set it up specifically for research - I will suffer no loss of focus because I'm not using it that much on my desktop right now.

Vivaldi

Vivaldi has Workspaces; kin to Horse's Areas and folders. It has two level tab stacks. Stuff can be moved around. Usable. Adequate. But not as flexible or as visually descriptive as Horse's outline.

But the notes are infinitely better. When you have the notes panel open, you can write you note while still on a web page; you can cut and paste (or select and grab) between the two. A keyboard shortcut opens  notes in the main pane where a note can be written or edited in markdown rich text.
At this point, there are options. You can move notes into the tab stack; or you can copy the web pages into a note. For me, it's smoothest to write the note as I go along and copy pages into them; my heavyweight work is in my PKM notes app so the note was always my ultimate destination. And when I finish, I copy the history in too, so that I have a blow by blow record of my journey.

Vivaldi also has multiple navigation options.

And extensions. So I can use my password manager to login to all the sites I use for research. And upload articles to Readwise and my citation managers. Etc. Friction is minimised.

I wouldn't claim that Vivaldi is the best solution, even for my own work; idk. But I can definitively say that it is a far better solution for me than Horse, and Horse is unlikely to ever catch up.

Conclusions
Horse has a very neat idea but its potential is unrealised if only from the point of view of my own needs. Fundamentally, it's good for those who only want a curated and organised search history and record, with occasional incorporated notes; especially when they prefer to keep it all stored long-term in the browser.

It feels like a personal hobby project where substantial effort has been expended on monetising and marketing. The most recent incarnation of the website contains multiple messages that come over as empty AI garbage.

I came across  (this thread on Reddit).
"there's some dude and a gang of bots astroturfing this sub about "horse browser" which looks just like some typical chromium garbage. They're definitely astroturfing, and doing a really spray and pray approach, because they commented on a [4 month old post of mine](https://www.reddit.c...queeze_all_features/) on this sub with barely any comments, and magically 20 minutes later, that comment has 10 upvotes, and would you look at that, it's the top comment when you click on the post. I think there's some chatgpt shit going on because just look at the messages:"
"I'm the sole person behind Horse Browser, I want to clarify that this spam was not my doing but the result of an overly enthusiastic affiliate. I've just addressed this issue directly with them"
'Thanks. Removed some of the comments. If the account keeps it up, it will be banned'

Horse marketing


Unfortunately it doesn't feel as if the same effort has been expended on meeting - or even understanding - the needs of its customers. The investment of a heavy research user into a process and workflow is substantial, certainly greater than the price of an app, and there's a need for longevity and professionalism in the tools used. I have no confidence of either with Horse.

Final Conclusion

Obviously I requested a refund.
This message came up on the Stripe refund page "we would appreciate it if you consider canceling without requesting a refund; that way we can keep working on improving Horse Browser for when you come back later." I ignored it.

[Shades]

Perhaps you should take a look at Strawberry browser. Is on invite-basis only at the moment, not free either, because of AI, but the user interface appears to be well suited for (automating) research in combination with AI. Screenshots of the UI you can find here, as well as a complete description, a FAQ, some example animated Gif's, etc.

There is a limited free tier, and 2 priced tiers. Added myself to its waitlist for the free tier a week or so ago. No clue how long that will take, Did something similar for the Manus AI, and that took 2 months or so. Lets hope that Strawberry doesn't disappoint in the same way.

[Dormouse]

Perhaps you should take a look at Strawberry browser. Is on invite-basis only at the moment, not free either, because of AI, but the user interface appears to be well suited for (automating) research in combination with AI. Screenshots of the UI you can find here, as well as a complete description, a FAQ, some example animated Gif's, etc.

-Shades (May 30, 2025, 09:46 PM)

Thanks. I'll take a look; I'm certainly going to be paying more attention to browsers generally.

doesn't disappoint in the same way.

-Shades (May 30, 2025, 09:46 PM)

It may not seem like it, but I'm still very pleased about Horse. I'll admit that I now question whether the wonderful workflow I envisaged was somehow my own invention, but at the very least it was inspired by Horse. It should have been obvious - it was obvious! - but it had never occurred to me before.

Every now and then an interest in family history and genealogy takes hold of me. Because of the big gaps, I cannot remember the precise details of the work I have done or not done - especially failed searches. I know I should make detailed records, but I only ever do it incompletely, sometimes, and I'm unlikely to do better because it is a significant amount of drudgery for the remote prospect of a possible payoff in the distant future. This mostly automates the whole thing.

Off the top of my head, I recognise three types of search/research that seem to crop up in all domains.

• One is planned and systematic. Defined data gathering, statistical analysis, maths. What most people think of as typical science, though science uses all these types.
• The second is like a wood sculptor beachcombing for driftwood with potential. Probably one of the most common web activities. Despite being sometimes vilified as mindless collection, the items are often chosen as potential triggers for the imagination.
• And the third is like the hunt for Bigfoot; you know it's there, you're sure it's there, it's just very, very hard to find and most trails lead nowhere. And you just need a bigger collider.

And it's only the third type that I imagined Horse helping with. Not that it matters. Now that I have the idea, the Vivaldi method will work perfectly well enough.

Family history research is a funny beast. Most of the data you work with is in archives, which constantly acquire new data. So search results may change over time. And sometimes the site will only cough up the data if you search x way rather than y way because search algorithms can be glitchy. The data is often mistranscribed and the original respondents were often misheard, or had limited knowledge, or downright lied. Does this record belong to that person?; does that person, or that record, truly belong in your family tree? Nothing is certain. Everything you 'learn' should have a probability estimate, but all you can do is write it down and keep some type of probability in your mind. You increase probabilities by triangulation. Preferably supported by copies of actual written records and not just transcriptions or someone else's assumptions. You constantly look for stuff to extend, but more often you look for supporting evidence, and even more often disconfirming evidence. A disappointing but useful outcome.

So covering old ground is the norm. And remembering the sites or the precise search terms is hard, but writing down every slight change in location or date range is tedious beyond belief. It had never occurred to me to store that information in a browser. Partly because the searching can seem haphazard - I tire of not finding Gobble Grimstone in Derby, exhaust my supply of variants, Gimstone, Gimston, Gimson, Jemson; and switch to Abel Turkey (probably Tukey oc) in Nottingham; and round and round in circles. Even worse I do find them, but with records that show they're not one of my ancestors and I have to start again. Partly because tabs always seemed haphazard themselves and using multiple browsers didn't help because I never saw them as central to my workflows. Seeing tabs as an outline was a real eye opener; I can shuffle and rename, even use emojis. Reorganise in the browser. And the power is amplified by moving the searches into a PKM notes app.

I appreciate the lure of AI in all of this; genealogy sites are using it directly in a limited way. But I'm also very wary. The data is already probabilistic. AI's internal logic must function around probabilities. I fear a tendency to be overly certain or even a willingness to invent a record that it has decided must exist. Even inventing an image of a Dog Latin birth register from an Irish parish wouldn't be beyond it. And I would have no way of weighing the AI's probabilities.

PS That was interesting. I looked at the Strawberry link. I'd never quite realised that Lattics would just open URLs in one of its own windows.

[Shades]

You can run a (smaller) local AI/LLM easily enough. Depending on your GPU hardware. Or lack thereof. From this and other posts, I gathered that you stored all of your previous research already. That data could be put into a vector database (RAG) and this vector database can then be coupled to this locally running AI/LLM. Once that is done, you will see that even these smaller local LLM's are pretty good for helping you out finding what you need, collect this data and "feed" that into an external genealogy database.

You could even find out which research paths were a dead end, or maybe less of a dead end than envisioned, with a few simple prompts. Or tell the AI/LLM that those paths were already marked as a dead end, so not to be investigated (in an much more automated) way.

Smaller models do tend to hallucinate more than online ones, but if the data in your RAG solution is solid, you'll find there will be no to hardly any hallucination. The "garbage in, garbage out"-concept is very much a thing with AI/LLM. The very large online versions are usually filled with better/more coherent data, making those look good in comparison with smaller models.

But you will be very pleasantly surprised how well those small models perform, when you can let it go loose with your own proper data. And those will not rob you blind with subscription fees, token consumption limitations and possible overcharge fees.

Just get a free tool like 'LM Studio" (GUI tool for Windows, Linux and Mac) and/or "Msty" (GUI tool for Windows, Linux and Mac) or even "Ollama" (PowerShell/terminal-based text tool for Windows, Linux and Mac). All of these also have a server-like function. Meaning you can connect LLM web-interfaces (such as 'Open-WebUI') to these tools. Then you can use your local AI/LLM with any device in your LAN (computers, laptops, tablets, phones, even a smart TV if it has a decent enough browser).

Personally, I went the "LM Studio"-way, because it also has an excellent LLM model search function build-in. Where I discovered model 'ui-tars-1.5-7b', which is surprisingly sound of logic (without giving it a system prompt to tweak it) given it's size. And even manages to output between 4 to 5 tokens per second on a desktop with a 10th generation Intel i3 CPU (5 years old by now), no GPU of any kind, a small and simple 2,5" SATA SSD drive and 16 GByte of 3200 MHz RAM.

Fit such an old PC with a GPU that contains 6 GByte of VRAM and this model can be loaded into VRAM instead. The 4 to 5 tokens/sec output is too slow for person who reads. When the same model is loaded in VRAM, the output goes to around 12 to 15 tokens/sec. And that is fast enough for adept readers. Maybe not for speed readers, but given the state of today, there aren't that many persons anymore that have and/or use that skill.

Sorry for ranting on and on about this. Thought I mention all of the above, because you already did a lot of legwork and have the data.  And in this case, I expect (local)AI/LLM to be a big boon for you. Just need to figure out the RAG solution for your collected data. Tools like 'Rlama' and 'LlamaIndex' are likely to be a great help in finding the right solution and/or help you build your RAG solution, as both can deal with PDFs, images, images in PDF, word and excel documents, text and MarkDown, etc.

[Dormouse]

Sorry for ranting on and on about this.

-Shades (May 31, 2025, 07:22 PM)

Please, please continue.

I am aware that I need to move in that direction with this. Not that I will move quickly.

When I was a student, I saw that the best understanding came from knowing all the details from the ground up, and that thinking consequently became superfast. I always worked to understand my data and processed through the underlying equations and maths before using formulae. I always distrusted blackbox packages. Still amazes me that Stats courses at uni concentrate on formulae, interpretations and packages with minimal effort given to teaching data types or shapes; afaics the most common failures irl come from a failure to understand the basic data. I suppose now they're moving(ed?) to AI.

Which is fair enough. Understanding and appreciating the basics isn't the same thing as wanting to do them manually every single time.

you already did a lot of legwork and have the data.

-Shades (May 31, 2025, 07:22 PM)

Legwork 'yes'; have the data 'no'; at least not in family history. But I will be recreating it; fairly straightforward because I do have all the finish points and remember the routes and already knew I needed to do that.

You can run a (smaller) local AI/LLM easily enough.
...
you will be very pleasantly surprised how well those small models perform, when you can let it go loose with your own proper data. And those will not rob you blind with subscription fees, token consumption limitations and possible overcharge fees.

-Shades (May 31, 2025, 07:22 PM)

Is local best for this? I know it's very personal data, but it's not private - it's virtually all derived from public databases after all.
Is the downside of the online models mostly to do with cost? In which case, I'd need to compare it with the cost of local. Without being slowed.

And that is fast enough for adept readers. Maybe not for speed readers, but given the state of today, there aren't that many persons anymore that have and/or use that skill.

-Shades (May 31, 2025, 07:22 PM)

I suspect I'm fast; idk about speed reader. Usually read 10+ books a week, frequently +++, as well as everything else.

Depending on your GPU hardware. Or lack thereof.

-Shades (May 31, 2025, 07:22 PM)

Lack entirely; not being a game player. And even shifted to basic level PCs in recent years because I don't (didn't) need power for anything. I really like my little Geecom mini.

you will see that even these smaller local LLM's are pretty good for helping you out finding what you need, collect this data and "feed" that into an external genealogy database.

You could even find out which research paths were a dead end, or maybe less of a dead end than envisioned, with a few simple prompts. Or tell the AI/LLM that those paths were already marked as a dead end, so not to be investigated (in an much more automated) way.

-Shades (May 31, 2025, 07:22 PM)

And that would definitely be good.

token consumption limitations and possible overcharge fees.
...
'LM Studio" (GUI tool for Windows, Linux and Mac) and/or "Msty" (GUI tool for Windows, Linux and Mac) or even "Ollama" (PowerShell/terminal-based text tool for Windows, Linux and Mac). ... LLM web-interfaces (such as 'Open-WebUI') to these tools.
...
LLM model search function build-in. Where I discovered model 'ui-tars-1.5-7b', which is surprisingly sound of logic (without giving it a system prompt to tweak it) given it's size. ... 4 to 5 tokens per second on a desktop with a 10th generation Intel i3 CPU (5 years old by now), no GPU of any kind, a small and simple 2,5" SATA SSD drive and 16 GByte of 3200 MHz RAM.
...
Just need to figure out the RAG solution for your collected data. Tools like 'Rlama' and 'LlamaIndex'

-Shades (May 31, 2025, 07:22 PM)

There's a whole new language for me to learn, as well as the content, as well as how to apply it.

--

My current plan is:

• Gather the data. Organising and reorganising the searches in the browser.
• Saving each organised sequence (aka "trail") into markdown - both Lattics and Tangent/Obsidian.
• Using something else to help me visualise it and help me go forward. Might be some sort of data oriented PKM app, or AI or something else.

I appreciate that an AI oriented browser might be able to do the lot. But I'm extra hesitant to trust the whole process to a newly developed program with unknown (possible) pitfalls.

[Dormouse]

It confirmed that the price was $60 ... I clicked the talk to a person button and emailed. There has been no response.

-Dormouse (May 30, 2025, 09:29 AM)

I finally received a response by email:
"I have fixed the bug on the homepage, it should now correctly display either $80 or £60, not “$60”. "
The € price has been upped to €70.

Presumably this will help address this issue mentioned on their marketing page:
"Conversion Rate
The percentage of trial users who pay up. We're running at a ridiculous 25% versus the 10% industry standard—meaning we're still undercharging even after all our price experiments."

idk if they're going to let all the sites quoting $60 know

[Dormouse]

tbf the trails idea is good, and clinking a link as a subpage is fast and effective. Useful for some types of research.

In rests between thinking about, and setting up, my Vivaldi workflows - and reading about AI - I've look at various tree style browser solutions, and they appear to lack clarity of purpose and effective default behaviours. They don't seem adequate alternatives.

Having notes in the tree is also a very good idea - and not one I've come across  elsewhere. But the default behaviours are poor.

Which means that it's a system that might work well for those who want to keep all their info in the browser, but not for those that want to work with the notes.
And more a one-trick pony than a packhorse. And none of the expected comforts of a saddle horse either.

It's already apparent that for me Vivaldi is far faster and has fewer frictions for doing this type of research. This is partly because of the excellent inbuilt note options, and partly because of the availability of all my usual extensions. And a password manager that works. And it's not costing me an annual $60 $80.