Apptisan #39 - Talking with FlowDown’s Creator: Beyond Laggy AI Assistants, How This App is Incubating a Truly Smooth Personal Agent
Indie developer Lakr created FlowDown after two years of frustration with laggy AI chat apps. This personal agent leverages iOS capabilities to deliver web search, tool calling, and system...
Name: FlowDown
Developer / Team: Lakr
Platforms: iOS, iPadOS, macOS
Read this newsletter issue in Chinese (中文) .
Please describe your product.
At its core, FlowDown is an AI chat application.
To give you an example: you can drop an image into FlowDown and tell it, “Help me search the web for this.” FlowDown will then recognize the image, find relevant information online, process the webpages, filter out extraneous content, and present you with the specific information you were looking for.
Contextualizing it within the current tech landscape, you could say FlowDown is an incubating personal Agent. It’s being developed to leverage the native capabilities of the iOS system to perform tasks on your behalf.
For FlowDown, what does “personal Agent” specifically mean? And how do you envision FlowDown evolving into such a product?
For FlowDown, “Personal Agent” is currently a rather generalized concept, which makes it difficult to define in precise detail right now. If I had to state one concrete thing, drawing from the basic definition, an Agent is something that does things for you.
FlowDown is still very much in its incubation phase. I have to admit my own limitations – I don’t yet have a perfectly clear vision of the definitive path forward. It’s entirely possible FlowDown might not evolve into the idealistic version of a personal Agent often discussed.
For now, our team is focused on strengthening the foundational features, fixing bugs, and polishing the user interface. Once we have a clearer understanding of the direction, perhaps it won’t be too late to share more then.
On a personal note: I hope FlowDown becomes an Agent that isn’t overly intrusive. When I need it, I want to be the one to initiate the interaction, have it manually gather the necessary information to solve my problem, and then deliver a convenient result. I don’t want it anticipating my needs or constantly monitoring my activities.
Was there a pivotal moment that inspired your product’s creation?
The desire to build something like FlowDown started about two years before we actually began development. Back then, I saw it as a monumental challenge as we might say – and honestly hoped someone else would tackle it. I waited for two, maybe three years, but the similar apps that emerged were, frankly, one worse than the next. They suffered from several major pain points:
First and foremost, lag. Basic performance is crucial. When I use a mobile client, I expect it to run smoothly. During the development of FlowDown’s community edition (our tech validation version), we achieved a level of smoothness in less than a week that virtually no competitor could match. That proved to us that the flashing screens and choppy scrolling weren’t technical impossibilities; it was just that no one was taking performance seriously.
Second, unbelievably, not a single product offered integrated web searching within the local client. It seems straightforward – help the user open a browser, search for something, summarize, and organize the results – yet no one was doing it. Interestingly, shortly after I completed a technical research project for this feature, ChatWise implemented it. That was the first time I saw a third-party client with this capability.
Then there’s tool calling. The current state of Siri is just bafflingly bad. It’s so difficult to use. After trying numerous clients, I still couldn’t find a reliable product that could simply help me add a calendar event, check my location, or open a link. I later realized this relates to the concept being heavily hyped this year (2025): the “Agent.” What I found most absurd was that these are relatively simple functions, yet they weren’t being implemented properly.
Finally, apps capable of running offline models universally lacked a good user experience. They typically only supported model downloading and the most basic chat interactions. This was unacceptable.
Those were the main functional pain points. Additionally, I noticed that when I opened competing apps, one had Telemetry, another had Firebase. Several network requests fired off the moment the app launched. In my view, this is unacceptable for software handling potentially sensitive personal information.
I simply couldn’t stand it anymore. Coincidentally, a few friends offered some funding, mentioning they needed technology in this area. I thought, perhaps it was time. Very quickly, we put together a rough prototype. Compared to existing apps, it felt incredibly smooth and comfortable. The moment that prototype came together solidified my decision to commit to building FlowDown properly and making it truly usable.
The name “FlowDown” (浮望) is quite different from typical tech product names. Could you share the story behind this name and how it connects to your product vision?
That’s an interesting story.
The English name “FlowDown” was a bit of a spur-of-the-moment idea. “Flow” represents data streaming past your eyes, and ”Down” signifies it landing, becoming accessible. That became the project’s codename. Later, we looked for a Chinese name with a similar sound, initially settling on “浮荡” (Fúdàng - ”float/drift”). However, some users criticized the character “荡” (dàng), pointing out its association with the word “淫荡” (yíndàng - lewd), which wasn”t ideal. So, I changed it to “浮望” (Fúwàng), replacing “荡” with “望” (wàng - ”gaze/hope/expect”).
Regrettably, in terms of pure product functionality, there isn’t a grand vision directly tied to the name. Apart from coding and occasionally asking questions, I don’t heavily involve AI in my personal life. Beyond the app’s features, however, I do hope FlowDown can play a part in the process of helping my users reclaim ownership of their digital lives.
What makes your product unique compared to others in the market?
The most immediate difference people feel is FlowDown’s smoothness. Before FlowDown existed, every configurable chat app we researched and used felt sluggish. Just scrolling might involve waiting several seconds. In our eyes, this is unacceptable.
Then, we noticed that existing third-party apps generally don’t support system interface integration. FlowDown invested significant effort into implementing tool calling. This capability, which arguably should be standard, was something FlowDown prioritized for its launch. If you ask FlowDown, “Where am I?”, it can actually query your location and tell you.
Finally, we believe FlowDown has numerous subtle refinements and small improvements. They might not be individually worth highlighting, and I might not even be able to list them all, but once you’ve used FlowDown, going back to other apps feels like a real downgrade.
Extra Point 1: Our software is designed with privacy at its core. We aim to be the cleanest third-party chat app available. We collect absolutely no user data and actively teach our users how to verify these claims. We hope to set a positive example for the industry.
Extra Point 2: We recently held a week-long “Open Source Week” (ending March 10th). We hope this encourages the emergence of more high-quality software and helps developers avoid reinventing the wheel. This is something other paid software providers are generally unwilling to do. (Note: The open-source week has concluded; all released code can be found on the official website.)
You’ve emphasized FlowDown’s smoothness as a key differentiator. What specific technical challenges did you face in achieving this performance, and what solutions did you implement that other developers might learn from?
Honestly, there’s no groundbreaking technology to boast about. We simply did the fundamental work correctly that others perhaps should have done but didn’t. We handled user interactions diligently, without cutting corners. That’s basically it.
If I must talk specifics, it boils down to three main principles:
Don’t perform heavy or blocking tasks on the main thread.
Cache everything that can be cached.
Implement custom handling for dynamic height changes and scrolling behavior.
As a side note: three years ago, for a project called Kimis, I had unixzii (who’s also on the FlowDown team) write a text rendering framework called Litext. It wasn’t deployed back then because it didn’t support direct text selection. We repurposed it for FlowDown, fixed a few memory leaks, and found it performs exceptionally well. It’s truly a blessing to have reusable code available.
Regarding “benefiting other developers,” I’m not sure if you’ve seen our official website, but we launched an “Open Source” page where we’ve released some of our frameworks. This includes Litext and the first-generation Markdown rendering engine used in the FlowDown community edition. This is the most direct way we can benefit developers and potentially the entire industry.
I believe open source promotes technological advancement and helps me better fulfill my mission: to help everyone reclaim ownership of their digital lives.
How have you marketed your product, and what key lessons have you learned?
Actually, I’m not particularly fond of widespread promotion.
The main reason is that navigating the registration and compliance requirements for chat applications in mainland China can be quite troublesome. We prefer to attract high-quality users – people who are responsible for their actions – and focus on serving them well.
In fact, on the day FlowDown launched and the day after, we did absolutely no promotion. Users found it organically, with some even posting in group chats saying, “Hey, it’s released, I’m going to purchase it.” :P
Looking ahead, we plan to grow things slowly. We believe a good product will naturally promote its own value. Of course, this might also be because I don’t have the bandwidth to juggle multiple roles; I have more important things to focus on.
What has been some memorable feedback since your product’s release?
Honestly, no, not really! No particularly memorable feedback; it has all been within expectations – requests for supporting specific models, reports of issues needing fixes.
Are there any products out there that you feel deserve more recognition?
I don’t know. I genuinely don’t know. I tend to be quite adaptable and accepting, and currently, there isn’t any software that makes me feel compelled to recommend it widely.
I do, however, think the friends who built this software with me deserve to be known by more people.
Could you introduce the friends who built FlowDown with you and describe your collaboration style?
How we met feels like destiny. Regarding the software itself, two core developers were involved this time. My meeting with the other developer was purely serendipitous, not something you could replicate. It’s a small circle, we happened to know each other, that’s all.
From day one, we discussed profit sharing, as I firmly believe the software needs to be financially sustainable to benefit more people in the long run. We then created an initial task list and rough plan. I started working on UI building blocks – the text attachment editor, popups, menus, etc. My partner identified performance bottlenecks in the first-generation rendering engine and took on the massive task of rewriting it entirely. Later, he delved into optimizing TableView (a crucial list view layout engine) and ScrollView (we rewrote its scrolling logic to completely eliminate jumping issues caused by content height updates).
Our division of labor evolved very naturally, driven by both technical needs and individual strengths. If I felt something wasn’t good enough, I’d build a better version. If I was proficient in a specific area, I’d tackle it. We didn’t rigidly plan everything; tasks got done when time permitted, and delays weren’t critical. Initially, I met up with him daily to code together, but later we only convened for key integration points.
Beyond the app itself, we have fantastic teammates handling website development and design. Our designer, who thought the initial pigeon logo wasn’t great, created the current one for us. Meeting them was also serendipitous, and task allocation happened organically – everyone just knew what needed to be done.
I want to add that every member of this team is overqualified, some significantly so. In software development, both my partner and I are capable of independently delivering complete, user-approved, high-quality products. On the web side, our teammates can each independently design, develop, and operate an entire website. Our designer is currently the lead designer at a well-known note-taking software company in Hangzhou. They are all incredibly talented; I might actually be the least technically skilled person on the team.
Tell us a bit about yourself and what you envision for the future.
Hi everyone, I’m Lakr. Our team has five people involved in the FlowDown project, though they prefer to stay low-key. We’ve worked on many projects together, and we’ve poured a lot of care into bringing you FlowDown this time. As for future plans... there are no plans. We prefer to work on projects spontaneously, without too many rigid structures or limitations.
We don’t feel there’s much more to introduce; we just hope everyone enjoys using the app!
What’s your perspective on the current trend of using AI for programming and rapid product launches? Has this influenced your development process for FlowDown?
At its core, AI currently operates on principles of statistics. It predicts the most probable next character, or one among several probable characters.
Unfortunately for us, humans are also products of statistics. Our neural networks are cells that react to external stimuli by adjusting activated neurons based on past memories.
When an AI’s context window becomes sufficiently large, its activated neural network sufficiently complex, and its reward mechanism sufficiently human-like, it will be able to function much like a person. Through vast memory and computation, by analyzing the roots of your joys and sorrows, it will inevitably be able to create products perfectly tailored to please you.
We can’t escape this. Even the singer Stefanie Sun has chosen to acknowledge and accept AI’s role.
Based on this perspective, I believe this trend is inevitable and unstoppable.
Before going further, I need to add my view on what makes a product successful. It must possess at least one of these qualities:
It helps others make money.
It helps others save money.
It fulfills others’ emotional needs (provides emotional value).
It fulfills others’ physiological needs (provides physiological value).
As AI continues to develop, it first helps people save money, then helps them make money. Around the same time, or perhaps slightly before or after, it starts fulfilling emotional needs. If it can help make money, fulfilling physiological needs becomes easier, which in turn facilitates emotional fulfillment. Therefore, based on AI’s current trajectory, it’s primarily in the stage of helping people save money and beginning to help them make some money.
Given this, understanding how to leverage AI tools to save money for oneself and make money for others is something everyone who doesn’t want to be left behind by the times must engage with and discuss.
In summary: I see AI programming as an undeniable trend that everyone needs to learn, master, and understand. There’s no need for excessive anxiety, because replacement, in some form, is inevitable. Finding your unique value and purpose is the only antidote to this anxiety.
Apologies if this topic got a bit philosophical; my thoughts on it aren’t perfectly organized yet.
Finally, did it affect me? Yes, but only positively.
For instance, some translation tasks became unnecessary – just hand them off to AI. Generating structured content became straightforward. Simple logical components could be entirely planned and designed by AI. This significantly shortened FlowDown’s projected 6-month development timeline to just one month, delivering returns that far exceeded the effort invested.
About Apptisan
Apptisan is a portmanteau of “application” and “artisan”, signifying “a weekly exploration into the world of apps and the passionate artisans who create them.” Each issue is a conversation with global creators, aiming to uncover and present intriguing products to a wider audience.
For those who prefer Chinese, you can subscribe via Quail. Creators interested in featuring their products are encouraged to submit them through our form, and we’ll be in touch promptly.
Supporter Status
If you enjoy Apptisan, consider showing your support in these ways: