On October 21, Personal AI company Today AI announced it has raised more than $200 million to date. Investors include IDG Capital, Tencent, 5Y Capital, Ant Group, Sequoia China, Transsion Holdings, Tongge Ventures and Shanghai Jiao Tong University's Hanyuan Asset.

Founder Qi Junyuan also founded the collaboration tool Teambition, which Alibaba acquired for $100 million in 2019. He went on to work at Alibaba and then ByteDance. In late 2025, Qi struck out on his own again, betting on Personal AI.

Today is also among the earliest teams in China to explore Personal AI. It opened an overseas public beta in September and launched the domestic version on September 24, only recently disclosing its funding. Over the same period, Silicon Valley saw a wave of buzzy Personal AI products including Muse, Cue, Dots and Instinct.

Yet the Today team struck a restrained tone as it announced the round. Personal AI, they say, is still at a very early stage, and some of the current excitement is emotion-driven.

The Use Cases Taking Off in China Aren't the Same as Overseas

A month after the domestic launch, some rather different use cases have emerged.

One user preparing for China's CET-6 English exam got a last-minute notice about volunteer training. The session was the next morning, clashing with her scheduled vocabulary study. Comparing her earlier study plan with the newly added event, Today proactively suggested moving vocabulary review to the evening.

Another user, a blogger and a mother, uses Today to organize both work and her child's schedule. One time, a newly added client meeting collided with her kid's baseball class. Before she noticed, Today had already flagged it a day in advance and drafted a message to send to the father to sort out logistics. Once she confirmed, the calendar updated with a single tap.

During the National Day holiday, a user living in Tibet was mulling over where to go the next day. Before he even asked, Today's evening digest recommended a temple frequented by local Tibetans, along with the best windows for setting out and taking photos. The little-known temple matched his interests exactly, even though he had never mentioned the trip to Today. It turned out the suggestion came from an earlier interaction: he had once asked Today to organize photos and travel materials, and those experiences and preferences had been remembered and put to use on this trip.

The three scenarios look different, but they solve the same class of problem: many things don't wrap up in a single conversation, and life keeps throwing changes at plans already made. Nearly a month after the domestic launch, the clearest value Today has seen lies precisely in these tasks that stretch across weeks or months and keep absorbing new information along the way.

That differs from the use cases Personal AI products found first overseas.

For products like Instinct and Muse, many of the magic moments come from execution: booking tickets, handling expense reports, canceling long-unused SaaS subscriptions. Instinct's founder once revealed that half of the platform's GMV comes from travel. A large share of overseas services still runs on the web, so a browser can cover plenty of long-tail needs, and an agent with the right permissions can act on the user's behalf.

China is another story. The country moved into the mobile internet era early, and booking tickets, ordering food delivery and hailing rides are already extremely convenient. Many services are also wrapped inside separate apps and relatively closed ecosystems. Having an agent open a browser to book a ticket delivers far less of an experience gain than it does overseas, and it is harder to pull off.

What the Today team has observed is that Chinese users show demand fastest in scheduling, to-do lists, knowledge management, and some CRM-like scenarios.

Someone who never uses a calendar can simply tell Today they might have something on next week. The system turns that unstructured sentence into a calendar entry, then keeps adjusting it as things change. A teacher can use it to track the progress of different students. Long-running study plans, exercise and family affairs all suit a system that remembers how things got here.

Among user feedback, one person writing a doctoral dissertation has been using the product since the March closed beta. He uses Today to log his daily state, organize his thesis and references, manage to-dos, and track his workouts and his cat's health. Another runner connected his COROS watch, and Today worked out a training plan with him, adjusted it around the weather, and then proactively recommended his next race three weeks after he finished one.

Clearly, as long as something requires sustained follow-up and keeps getting reshaped by new information, Personal AI can help. It isn't only founders and investors with packed calendars who need it.

As for whether these use cases amount to finding PMF (product-market fit), the Today team answers carefully, preferring to call this stage Product-Value Fit. Some users genuinely need it and the product is genuinely useful, but whether that value translates into durable retention, extends to a broader population, and eventually becomes revenue still needs proving.

Personal AI today has shown it can be useful. Becoming something people open every day is still some distance off.

Proactivity Is the Aha Moment Right Now

Traditional chatbots and agents generally wait for a task and then deliver a result. Today's Personal AI wants to go a step further: continuously understand a person's goals and state, spot what's worth handling, speak up at the right moment, and when needed, see the task through.

That is what gradually separates Personal AI from chatbots in product design.

The Today team believes the standout feature in the current version, and the one most likely to produce an aha moment, is proactivity. But being proactive isn't about pushing out more notifications; it has to rest on an understanding of the person.

"The experience gain from proactivity comes largely from personalization. And personalization comes largely from how memory is designed, so these capabilities are coupled together," the team told Founder Park.

Ask an AI to research a product and general-purpose tools like Doubao or Manus can do it, but it feels like completing a relatively objective, standalone task. A Personal AI that already knows who you are, what you do, and why you're researching will carry that context into deciding what to focus on and surface the part you actually care about.

In the user feedback Today sees, what comes up again and again isn't speed or feature count. It's "it remembers" and "it gets me." One user fed three years of notes into Today and found it began citing ideas he had written before, unprompted, in conversation. Another mentioned just once that he dislikes morning meetings, and three weeks later Today automatically steered his schedule away from mornings.

Looking at the industry as a whole, the Today team sees good conversational experience and long-horizon task capability as settled consensus for Personal Agent products. How humans and agents should actually interact, though, is still an open question every company is probing differently.

Today's main entry point right now is a persistent primary conversation. But the chat box may not be Personal AI's final form. Natural language is the best way for people to interact with each other, yet when the counterpart is an agent skilled at writing code and expressing things graphically, cramming every interaction into a chat box isn't necessarily the best fit.

Internally the team is also discussing formats like dynamic UI and generative UI. In their view, as GenAI products mature, each person may end up seeing a different product. Some need schedules and tasks, some care about health trends, others want work information front and center. The system could dynamically generate a dashboard tailored to each individual.

Before the interface can truly vary by person, though, Today has to prove something more basic: that what it remembers is accurate, that it speaks up at the right moment, and that this experience holds up consistently across different users.

Vertical Is the Starting Point, General-Purpose Is the Endgame

Understanding a person requires knowing enough about them. For the Today team, that is the crux of competition among Personal Agents: context. The race comes down to two things: who can obtain as much of a user's context as possible, and, given the same context, who can deliver the better experience.

That is why Today chose to build a general-purpose Personal AI from day one. The team argues that a product that can only access information from one narrow domain will struggle to understand a person's work, health, family and offline life all at once.

That information isn't a set of unrelated folders; the pieces shape each other daily. Health data alone won't reveal how months of overtime are wearing someone down, and work data alone won't explain why one last-minute meeting upends the school pickup.

"There's a chemical reaction between different dimensions of context. In a sense, that reaction is one of the product's moats," the Today team said.

The general-purpose route is heavy, of course, and not every team can start there. For startups with limited resources, beginning with a vertical use case is more practical: the scenario is easy to explain, the target users are clear, customer acquisition and computing power costs are easier to control, and willingness to pay is easier to test. Town is a case in point, entering through the concrete need of email before gradually expanding to founders, parents and other groups.

The Today team respects that approach but is clear-eyed about its limits: a vertical use case works as a product's starting point, but rarely as its destination. Models themselves generalize strongly, and a Personal AI product that stays confined to one narrow scenario may eventually be absorbed by general-purpose products.

Going general means facing messier, long-tail demand. Even with domestic and global giants piling in, general-purpose Personal AI still has plenty of unresolved questions: operating costs remain high, delivery capability isn't strong enough, and whether it ultimately becomes an app, a phone, or something else entirely is still undecided.

On Personal AI itself, Today is firmly convinced. The team believes agents will embed themselves ever more deeply in daily life, and that Personal AI will become a sufficiently important product category.

What Separates Products Is Evaluation, Not Memory

Aiming to "understand users better" brings concrete engineering problems: which information is worth remembering long-term, and which needs to be judged stale? When should the system speak up, and how should it pick up a task after an interruption?

Many people expect Personal AI companies to produce proprietary memory technology. The Today team is blunt: Muse, Today and Instinct are solving similar needs, and the optimal engineering architecture is fairly consistent. At least for now, no one holds a generational lead in memory.

What actually determines whether a product delivers, and what separates experiences, is whether a team can build a solid evaluation system and accurately attribute problems when the experience breaks down.

A coding agent works against a relatively closed repository, where whether the program runs and whether tests pass are objective standards. Personal AI's tasks are often open-ended and subjective.

Take a user saying, "I want to buy some clothes soon, help me figure out what suits me." The answer depends on the person's job, taste, budget, life circumstances and the occasions they dress for. The same recommendation can feel spot-on to one person and useless to another.

Personal AI has to recall the right memories, judge which of that information is relevant to the question at hand, and then give advice that fits the person's situation. Models don't yet generalize well enough to read all the subtext on their own, which makes the product manager's role more important, not less.

Today has poured a large share of its effort into evaluation. Product managers have to understand different users and personas, and encode human prior knowledge into sets of rubrics, in effect acting as judges on behalf of different users. Subjective impressions like "does it get me" and "was that proactive nudge appropriate" get broken down into dimensions an agent can understand, observe and quantify.

Evaluation is only the first step; attribution is harder. When an experience degrades, is the model at fault, or memory recall, the harness, or some product policy? Without pinpointing the specific link in the chain, a team can't know what to optimize next.

To test how the system actually delivers over long horizons, Today built a set of simulated users with continuous life histories, comparing three states: no history, a static personal profile only, and 30 days of accumulated interaction history. The team wants to see whether, as context accumulates, the agent's help really gets closer to that specific person and actually gets the task done.

The other part of its technical investment goes into models: Today has started training its own.

Mainstream large models today mostly focus their iteration on coding and complex reasoning, and don't necessarily prioritize conversational quality. In practice, models still ramble, follow instructions inconsistently, and struggle to adapt to an individual's way of expressing themselves. For a Personal AI that interacts with people every day, these small interaction flaws directly affect whether users stick around.

Cost is another practical factor. Today's harness calls on multiple models at once, and different tasks have different requirements for intelligence, speed and price. Even if a top-tier model that can do almost anything exists in the future, using it for every single task would be impossible if the cost is too high.

So Today isn't training models to chase frontier models on raw intelligence. It's looking for a better balance within the impossible triangle of intelligence, performance and cost, using its own scenarios and data to jointly optimize model and harness.

Prove It's Useful First, Then Talk Monetization

The team has thought a lot about monetization, but says Personal AI makes it hard to settle on a clear plan at this stage. Today's overseas version already runs on a subscription; the domestic version is free for now, with pricing to be adjusted "dynamically" based on the competitive landscape.

Advertising, transaction commissions and subscriptions are all possible paths to revenue, and each would reshape the relationship between product and user. The Today team's firmest principle is that Personal AI has to be on the user's side. The biggest difference from platform products lies in that word "personal": no monetization experiment can come at the cost of user experience and trust.

Follow that principle and the tradeoffs across business models become clear.

Advertising easily erodes trust, as the product may start favoring whoever bids highest. Commissions on fulfillment stand a better chance, provided the agent genuinely helps users get the right service at a lower price or higher quality, and takes a cut of that.

Subscriptions may not be the final answer, but in a product's early days they may do the least damage to the experience: when the money comes straight from users, the product has more reason to create value for them rather than for platforms, merchants or advertisers.

It's early to be debating any of this, though. Personal AI's costs are still high and the value it creates isn't big enough yet. The Today team's read is that the most important thing for the industry in the near term is still creating value: finding more real needs and serving more users, without rushing to pin down ARPU (average revenue per user). Real competition may only begin once model and inference costs keep falling and products can scale more cheaply.

"A lot of users question whether this kind of product really delivers that much value. Answering that question first matters more."

Asked where this round of funding will go first, the Today team's answer is simple: "The product is the leading digit for Today, the '1.' Everything else is a '0' behind it. Only with that '1' in front do the zeros mean anything."

Ample funding lets a company stay at the table longer. Whether it ultimately stands depends on answering the most fundamental question: why do users need a product like this at all.

Today has found some answers. Some people need it to remember long-term goals. Some need it to untangle work and family. Others simply want a system that notices, when life shifts, the things they don't have time to notice themselves.

Whether that value can hold on to enough people, and whether it can support a company building general-purpose Personal AI, will take time to prove.