Perplexity has an open-source alternative now, and I'm hooked

It pains me massively to say (well, type) this, but the era of entering a query into Google, hitting Enter, and manually opening different websites to piece together an answer feels like it's slowly coming to an end. The way the average person looks for information online has changed considerably over the past few years, and I know you and I are both guilty of turning to AI-powered tools instead of traditional search engines every now and then. While I initially struggled to see the appeal of using chatbots like ChatGPT and Claude, AI-powered search engines are a bit of a different story.

Take Perplexity, for example. If an AI chatbot and a traditional search engine had a baby, Perplexity is pretty much what you'd get. Rather than just presenting you with a list of websites, it searches the web, pulls information from different sources, and uses AI to piece everything together into a comprehensive answer, complete with citations. Essentially, it does much of the digging you would otherwise have to do manually. Though I've now come to appreciate how convenient Perplexity's approach is, I've had my reservations about it. This is exactly why Vane, an open-source alternative to Perplexity, immediately caught my attention.

Vane is an open-source take on Perplexity

Perplexity's DIY cousin

Formerly known as Perplexica, Vane is a free, open-source, "privacy-focused" AI answer engine. Beyond being open-source, Vane is also self-hostable, allowing you to run your own instance on your computer rather than relying on a service hosted by another company. In fact, that's exactly what I did, setting it up on my MacBook Air using Docker.

Under the hood, Vane works on a similar principle to Perplexity. When you enter a question, it first analyzes your query to determine how to approach it. If the question requires information from the web, Vane uses SearXNG, an open-source metasearch engine, to retrieve relevant results, which an AI model then uses to generate an answer with citations.

For more complicated questions, Vane can take things a step further through its research capabilities, performing multiple searches and gathering information from different sources before presenting its findings. Perplexity follows a similar approach, though the key difference lies in how both tools retrieve information. While Perplexity has developed its own search infrastructure and massive web index, Vane relies on SearXNG to fetch results from existing search engines.

Like Perplexity, Vane also provides citations alongside its AI-generated answers, allowing you to trace information back to its original sources rather than simply taking everything the AI tells you at face value. Vane offers three different search modes depending on how much research you want it to perform: Speed, Balanced, and Quality.

Similar to Perplexity, you can also choose whether to search the web, discussions, or academic papers. The tool supports image and video searches, too, and you'll find a dedicated Discover section for browsing interesting articles and trending content.

You're not tied to a single AI model with Vane

Model hopping without the monthly commitment

Before I jump into the results I got while testing Vane, let me get the most important bit out of the way. If you've used Perplexity before, especially on its free plan, you might've noticed how little control you have over which AI model powers your searches. While Perplexity gives you access to different AI models, the ability to manually choose between them is locked behind its Pro subscription, which costs $20 per month. On the free plan, your queries are automatically routed to a model selected by Perplexity, meaning you don't get to decide which one handles your requests.

Vane, on the other hand, gives you considerably more flexibility. It supports models from major providers like OpenAI, Anthropic, Google, and Groq, and you can connect them using your own API keys. You can even run models locally through Ollama if you'd rather avoid relying on a cloud-based AI provider altogether.

Now, this doesn't necessarily mean you can use every AI model for free. If you're connecting to a paid API, you'll still have to cover the usage costs. The difference, though, is that you're only paying for what you actually use rather than committing to yet another monthly subscription.

So, if you're someone who only occasionally needs AI-powered search, this approach could potentially work out much cheaper. However, if this is a service you rely on heavily throughout the day, a fixed monthly subscription might be the more economical option, since API charges can add up quickly.

Vane gets the job done, for the most part

The occasional wrong turn

To see how well Vane actually worked, I started with a fairly straightforward question about MakeUseOf, asking it to explain what the publication is and find three recent articles. It managed to pull together a decent overview of the website, complete with citations linking to different sources. However, it struggled with the second part of my request. Rather than finding three recently published articles, it included a category page and a media coverage page among its recommendations, even acknowledging that it couldn't confidently verify the publication dates of three recent articles.

I then decided to make things a little more personal and asked Vane to tell me about myself. To my surprise, it did a pretty good job. It found my MakeUseOf author page, referenced my work at XDA Developers, and correctly identified several other publications I've written for, including SlashGear, Laptop Mag, and Android Police. Perplexity did a similar job, too!

But I wanted to see how Vane would handle something considerably more complicated than a basic web search. So, I switched to its more advanced research capabilities and asked it to compare the privacy implications of using Perplexity, Vane with an OpenAI cloud model, and Vane with a locally hosted Ollama model.

This time, Vane went through 11 research steps and gathered 45 sources before generating a detailed answer. There were a few inaccuracies, but the response was fairly impressive as a whole.

Vane's research capabilities are surprisingly impressive

65 sources later...

I also put Vane to the test with a more niche technical question: how to connect an original Xbox 360 Kinect to an Apple Silicon MacBook Air. It correctly identified the need for a separate power and USB adapter, recommended the appropriate open-source driver, libfreenect, and highlighted potential USB compatibility issues. It even gathered 65 sources for the question, which was impressive in itself.

What I particularly liked was being able to follow Vane's research process as it happened. Rather than simply performing one search and generating an answer, it broke the question down into smaller parts, searched for relevant information, and used its initial findings to determine what it needed to investigate next. For instance, it started by looking into the differences between Kinect models and then moved on to researching Apple Silicon compatibility, power requirements, and reports of successful setups.

Vane's research process is quite similar to what I've noticed when using Perplexity. Both tools can break complicated questions into smaller research tasks, perform multiple searches, and build on their findings before generating a final answer. The fact that Vane manages to recreate so much of this experience using open-source components (and while running on my own hardware) is particularly impressive.

However, when I asked Perplexity the exact same question, its answer was noticeably more useful. For instance, while Vane suggested compiling libfreenect from source, Perplexity pointed out that I could simply install it through Homebrew. It also discovered FreenectTD, a project specifically designed to help use Kinect cameras with TouchDesigner on Apple Silicon Macs, which Vane had missed despite gathering considerably more sources.

The setup process was a pain

The DIY experience comes with DIY problems

One extremely underrated benefit you get with tools like Perplexity that most people overlook is just how frictionless using them can be. Beyond creating an account (and getting a subscription if you need to), there's practically nothing else you need to do. You simply open the website, type in your question, and get an answer. There's no software to install, no complicated configuration to figure out, and certainly no need to worry about what's happening behind the scenes.

Unfortunately, we can't say the same about self-hosted alternatives, including Vane. Especially for someone like me who isn't very experienced with using Docker or managing self-hosted applications, getting everything up and running can be quite intimidating. While Vane's setup process seemed fairly straightforward on paper, my experience was anything but. All I was supposed to do was install Docker, run a command, and access Vane through my browser. However, what I expected to be a relatively quick setup turned into hours of troubleshooting.

For starters, while I managed to get Vane running initially, its bundled search engine wasn't returning any useful results. Whether I asked it about MakeUseOf or practically anything else, it would consistently return zero search results. After some digging, I discovered that several of the search providers used by its underlying search engine, SearXNG, were blocking requests or returning errors. I eventually installed a newer version of SearXNG separately, which finally got things working.

I also ran into networking issues that prevented me from accessing Vane through my browser, and my attempts to troubleshoot everything ended up consuming so much storage that my MacBook Air was left with barely over 100 MB of free space!

Of course, some of these problems were specific to my setup, and your experience might be considerably smoother. Still, it's worth keeping in mind that the flexibility of self-hosting comes with the responsibility of maintaining everything yourself.

As impressive as Vane is, Perplexity certainly has the upper hand when it comes to convenience. However, for everything else, I do think Vane is a solid alternative worth considering, especially if you want more control over the AI models you use without committing to another monthly subscription.