Kagi May Drop AI Assistant From Its Pro Plan

What’s Happening

Kagi, the paid search engine that’s built a passionate niche following by promising ad-free, privacy-respecting results, is openly weighing whether to strip its AI assistant feature out of the Professional tier. The discussion, which has been percolating on Kagi’s community feedback forum, has drawn sharp reactions from subscribers — some who love the assistant, some who never use it, and many who just want their search to stay fast and affordable.

The move, if it happens, would redefine what you actually get when you pay $10/month for Kagi’s Pro plan. Right now, the Professional tier bundles unlimited searches with access to Kagi’s AI assistant (often compared to a ChatGPT-style conversational layer sitting on top of search). Removing the assistant would likely push it into a separate add-on or a higher-priced tier — a classic product unbundling play. CEO Vlad Prelovac has been transparent about the economics: inference costs for AI features are significant, and not every Pro subscriber actually uses the assistant. Why should light users subsidize heavy AI consumers?

This isn’t just a pricing tweak. It’s a signal. It tells us something real about the brutal economics of shipping AI features inside subscription products — and it should make every founder building an AI-augmented SaaS product sit up and pay attention.

Why Founders Should Care

If you’re building any product that layers AI on top of a core utility — search, analytics, CRM, you name it — Kagi’s dilemma is your dilemma. The fundamental tension is this: AI inference costs scale with usage, but flat-rate subscriptions don’t. When you bundle an expensive AI feature into a fixed monthly price, you’re making a bet that most users won’t hammer it. But power users always show up. And they will wreck your margins.

This is the pricing problem that Andreessen Horowitz has been warning about since 2023. Their research on AI inference costs showed that for many AI-native companies, the gross margins look more like a services business (30-50%) than a software business (70-80%). Kagi, a small independent company without the deep pockets of Google or Microsoft, is feeling that squeeze in real time. And their response — unbundling — is the most rational move available.

The market context matters too. The AI search market is projected to be worth over $30 billion by 2028, according to multiple analyst estimates. But that money isn’t going to companies that bleed cash on inference. It’s going to companies that figure out sustainable unit economics. For founders, the lesson is clear: how you package and price your AI features may matter more than the features themselves.

There’s also a timing signal here. We’re entering what you might call the “AI pricing correction” phase. The initial gold rush — where every SaaS product rushed to slap “AI-powered” on its homepage and bundle it for free — is giving way to harder questions. What’s the actual value of the AI feature? Who’s willing to pay for it separately? What’s the marginal cost of serving one more query? Kagi is being forced to answer these questions publicly, which makes it a useful case study for everyone else.

Who’s Already Moving

Kagi itself remains one of the most interesting indie search companies in the market. Founded by Vlad Prelovac, a serial entrepreneur based in Palo Alto, Kagi launched as a paid-only alternative to Google, betting that a segment of users would pay real money for search results free of ads and tracking. As of early 2026, the company has grown steadily — reportedly surpassing 50,000 paying subscribers — but it’s still tiny compared to the incumbents. The Kagi Blog has been refreshingly candid about the company’s challenges and strategic thinking, which is rare in tech.

Perplexity AI is the most direct comparison. Perplexity has raised over $250 million at a valuation exceeding $2.5 billion, and it bundles AI-powered conversational search into both free and paid tiers. But Perplexity faces the same unit economics question — it just has more runway to burn before it has to answer it. Perplexity Pro costs $20/month, and the company has experimented with usage caps on its more expensive model calls, which is its own form of soft unbundling.

You.com, led by former Salesforce chief scientist Richard Socher, has taken a different approach. It offers a multi-model AI assistant alongside search, with tiered pricing that explicitly separates basic search from AI-heavy features. You.com’s pricing model may end up looking prescient if the unbundling trend accelerates.

Brave Search is worth mentioning too. Brave has integrated AI features (Brave Leo) but kept them largely separate from its core search product, which remains free and ad-supported. Brave’s approach sidesteps the bundling problem entirely by treating AI as a distinct product line from the start — an architecture choice that gives it more pricing flexibility down the road.

“The biggest mistake SaaS founders make with AI features is treating inference costs like they’re fixed. They’re not. They scale with your most engaged users, and those users will always use more than you model for.” — A common refrain among AI infrastructure investors in 2025-2026.

The Startup Opportunity

Kagi’s unbundling moment opens up several concrete opportunities for founders who are paying attention:

• AI usage metering and billing infrastructure. If every SaaS company with AI features is going to need usage-based pricing, someone needs to build the tooling. Companies like Orb and Metronome are tackling this for general SaaS billing, but there’s a specific gap for AI-aware metering — tracking tokens, model calls, inference time, and tying it back to customer value. This is picks-and-shovels territory.
• Modular AI feature marketplaces. Imagine a world where a search engine, CRM, or project management tool lets you plug in the AI capabilities you actually want — and pay for them separately. The opportunity is in building the connective layer that lets SaaS products offer AI features à la carte without rebuilding their entire billing and integration stack.
• Niche, vertically-focused AI search. If horizontal AI search is a margin-crushing business, vertical AI search might not be. A search assistant purpose-built for legal research, medical literature, patent analysis, or academic papers can charge premium prices because the value per query is dramatically higher. The cost of inference stays roughly the same, but the willingness to pay goes through the roof.
• Lightweight AI alternatives. Not every user needs GPT-4-class responses. There’s an opportunity in building AI assistants powered by smaller, cheaper models that deliver 80% of the value at 20% of the cost. Kagi’s problem is partly that it’s serving expensive model outputs to users who might be perfectly happy with faster, cheaper answers. A startup that nails the “good enough AI” tier could undercut everyone.

The meta-opportunity here is in AI product architecture — helping companies figure out which AI features to bundle, which to charge for separately, and how to build the technical and billing infrastructure to support that. It’s not glamorous, but it’s where the money is.

Watch Out For

Let’s be honest about what could go wrong — both for Kagi and for founders chasing the unbundling trend.

Churn risk is real. Kagi’s subscribers chose the Professional plan with the understanding that it included the AI assistant. Removing it — or charging extra — could feel like a bait-and-switch, even if the economics justify it. User trust is Kagi’s most valuable asset, and this move could erode it. Community reactions on the Kagi feedback forum already show a split: some users are understanding, others feel like they’re losing something they paid for.

Google and Microsoft aren’t standing still. As The Verge noted in its early coverage of Kagi, the fundamental challenge for any paid search engine is competing against a product that billions of people use for free. Google’s AI Overviews and Microsoft’s Copilot in Bing are bundled at no additional cost to the user, subsidized by ad revenue and enterprise licensing. Indie search companies don’t have that luxury. Every pricing change is existential in a way it never is for Google.

The “AI tax” fatigue is growing. Consumers and businesses alike are starting to push back against being charged extra for AI features that feel like they should be table stakes. If the industry moves toward aggressive unbundling, there’s a risk of user revolt — people simply opting out of AI features entirely rather than paying piecemeal. Founders building AI add-ons need to be crystal clear about the value proposition, or they’ll face brutal conversion rates.

Inference costs are falling, but not fast enough. Yes, model inference is getting cheaper. Custom silicon, quantization, and more efficient architectures are driving costs down. But they’re not dropping fast enough to make the flat-rate bundling model sustainable for small companies right now. Founders betting on “costs will drop and solve our margin problem” are playing a dangerous game. You need a pricing model that works today, not one that depends on next year’s hardware improvements.

The Bottom Line

Kagi’s potential unbundling of its AI assistant isn’t just a niche pricing debate among search nerds. It’s an early, visible example of the reckoning that’s coming for every software company that rushed to bundle AI into existing subscription plans without doing the math. The economics of AI inference don’t lie: if you’re offering unlimited access to expensive model calls at a flat rate, you will eventually face this exact choice — charge more, cap usage, or unbundle.

If you’re a founder, the actionable takeaway is this: design your AI features as separable modules from day one. Build your billing infrastructure to support usage-based pricing. And don’t assume that bundling everything together is a competitive advantage — in 2026, it’s increasingly a liability. The companies that win won’t be the ones with the most AI features. They’ll be the ones with the most sustainable AI economics.