Flowith

采用创新的二维画布交互模式，支持多线程创作和知识管理的 AI 生产力平台

writing tools

https://flowith.io/

Visit Website

In-depth Report

Flowith is an AI productivity platform developed by a Chinese team, dedicated to providing users with an AI work experience that integrates "thinking, creating, and executing." Different from traditional conversational AI tools, Flowith adopts an innovative two-dimensional canvas interaction model, breaking through the limitations of linear chat and allowing users to perform multi-threaded creation and knowledge management in a visual space. According to public information, Flowith was launched in early 2025. Within 2 days after its release, it gained more than 100,000 online users, and the number of Agent calls exceeded 500,000 times. As of March 2025, the platform has achieved annual recurring revenue (ARR) of US$1.3 million, with 200,000 registered users, and has achieved pure product-driven growth with zero investment. The platform is currently supported by a community of over 1 million users worldwide.
Flowith is an innovative AI collaboration platform created by a Chinese team and launched in 2024 by the team of serial entrepreneur Derek Nee. The philosophy of the founding team clearly stated that they "do not want to be a general-purpose Agent", but instead positioned it as "the ultimate creative workbench in the AI era", taking a specialized route rather than a general-purpose Agent. The team size is about 10 people, with offices in Shanghai and Silicon Valley. Future plans for the platform include the launch of an iOS client (supporting voice input), video workflow functions, and a new “command-type” Agent for long-term complex goals. At the technical level, Flowith adopts a pragmatic route: high-intelligence scenarios use third-party mature models (because the model iterates quickly and self-research is uneconomical), and low-intelligence and stable scenarios use self-developed models to control costs. The platform has certain advantages in cost optimization, and the Token consumption for the same task is lower than similar products.
Canvas-style interactive interface: The core innovation of Flowith lies in its unique "infinite canvas" interaction mode. Users can freely create, connect and organize nodes on the 2D canvas to visually manage the creative process. This interaction method breaks through the linear limitations of traditional AI dialogue tools, supports parallel and serial connection nodes, and realizes true multi-threaded creation. Users can interact with multiple AI models simultaneously within the same interface, significantly improving information processing efficiency. Users who have used the product said that the canvas function is "actually a little amazing". It can compare and interact with multiple conversation contents at the same time on one page, changing from traditional single-threaded creation to multi-threaded creation. Knowledge Garden: Knowledge Garden is the core knowledge management function of Flowith. Users can upload materials in various formats such as text, files, and web pages, and AI will automatically analyze the content and split it into "knowledge seeds" to reorganize it into a dynamic knowledge association network. This feature is built on RAG (Retrieval Augmented Generation) technology, which enables users to conduct non-linear conversational exploration around their personal knowledge base. Creators can also publish the knowledge garden to the platform community and set a price (0.5-2.99 US dollars) to charge, forming a "knowledge trading market" business model. Oracle mode: Oracle is an AI Agent mode launched by Flowith, which can independently perform task planning and disassembly. Unlike traditional AI tools, Oracle does not rely on users' pre-built knowledge gardens, but only relies on model capabilities and network knowledge to complete complex tasks. This mode supports executing tasks in series in steps, searching for information in parallel within steps, and generating previewable web reports or demos. According to the founder, Oracle uses a new algorithm architecture, including technologies such as context compression, overlay generation, and intelligent dynamic programming. Users generally believe that the autonomous task planning capabilities of Oracle mode are "very powerful" and can help users complete method-level thinking. Agent Neo: Agent Neo is a new generation of creative Agent launched by Flowith, focusing on the "uninterrupted millions of contexts" capability. This product supports continuous execution of tasks in the cloud, and tasks can continue to run after the user turns off the computer. Neo uses a self-developed "infinite memory" mechanism, which is different from the traditional RAG solution and can handle extremely long contextual content.Multi-model integration and collaboration functions: Flowith integrates more than 40 mainstream AI models such as GPT-4, GPT-3.5-Turbo, Claude-2-100k, Gemini, Kling, ByteDance series models, etc., and users can flexibly switch according to different scenarios. In addition, the platform also supports file analysis, OCR text recognition, multi-person real-time team collaboration and other functions. It has a built-in chat and comment system to facilitate team creation.
Flowith offers four subscription tiers, with specific pricing as follows: Starter is free (300 Credits one-time), Pro is $19.90 monthly or $214.92 annually (22,000 Credits), Ultimate is $49.90 monthly or $508.98 annually (55,000+ Credits), and Infinite is $499.90 monthly or annually $4,799.04 (550,000+ Credits). All payment plans support monthly or annual payment, and annual payment can enjoy a 10%-20% discount. Pro plans and above can obtain commercial licenses, allowing users to use AI-generated content for commercial purposes. Target users include content creators, knowledge workers, enterprise teams, and individual users who have needs for AI-assisted creation.
Positive reviews: Based on media reports and community feedback, users have given Flowith high marks for its canvas-based interactive experience. Users generally believe that this interaction method is "very novel", breaking through the single-thread limitation of traditional AI dialogue, and allowing thinking to no longer be limited to a question-and-answer model. The Knowledge Garden function has also received positive feedback. Users believe that its full link coverage is complete, forming a good closed loop from the knowledge production end to the consumption end. The platform currently has 13 plug-ins covering scenarios such as search, image generation, and academic research. Suggestions for improvement: Users have also pointed out some deficiencies in Flowith. First of all, the Chinese translation coverage is not complete enough. After switching to Chinese mode, some pop-up windows still display English, which affects the user experience. Secondly, the filtering function of Knowledge Garden needs to be strengthened. When there are many categories, it is difficult to quickly find the target content. Some users also mentioned that some interaction details need to be optimized, for example, the action bar displayed by hover will not automatically disappear after leaving. Some in-depth users pointed out that the branch node logic of the product is not yet fully interoperable, and the flexibility in modifying generated nodes and workflow steps needs to be improved. In addition, the regular basic functions on the canvas (such as node editing, movement, etc.) are still missing, and ordinary users cannot yet fully rely on this platform for creation.
Several Chinese technology media have conducted in-depth reviews on Flowith. The article "Everyone is a Product Manager" pointed out that the core value of Flowith lies in its "canvas-style" interactive innovation, which can break the linear narrative and form a relatively unique business model with the subscription-based community and AI-assisted understanding capabilities. But it also pointed out that the product has not yet formed a complete ecological closed loop, and ordinary users may jump to external platforms. The author of "Minority" believes that Flowith is "more interesting than buttons" and the overall product experience is "dazzling and overwhelming", but there are still interactive details that need to be polished. From the perspective of market positioning, Flowith has made it clear that it “does not want to be a universal Agent”, but focuses on becoming “the ultimate creative workbench in the AI era.” This differentiated positioning is in sharp contrast to generic Agent products such as Manus and Lovart.
Flowith is suitable for users who need to create multi-threaded content, need knowledge management and accumulation, and are willing to try new AI tools. For users who only need simple questions and answers, traditional conversational AI tools may be more concise. Alternatives include Manus (universal Agent), Coze (visual workflow), Dify (self-deployment), etc. If users aren't interested in canvas-based interactions, it may take some getting used to.
As an innovative AI productivity tool, Flowith takes canvas-style interaction and knowledge garden as its core functions, taking a specialized route that is different from general-purpose Agents. The product has obvious features in interactive innovation and knowledge management, and the pricing strategy covers a diverse user group ranging from free to high-end. Users' overall evaluation is positive, believing that it brings a new experience that is different from traditional AI tools. They also point out that translation, filtering functions, interaction details, etc. need to be optimized. With its differentiated product positioning and pragmatic technical route, Flowith has shown certain development potential in the AI creation tool track.

User Reviews

PeterHoward_X

—

用了一段时间 Neo，做研究报告确实省事。以前自己到处搜资料再整合，现在给它一个目标就能跑完整个流程。但积分烧得是真快，深度研究一次下来几百积分就没了。

Kenneth.BaileyIII

—

画布的概念是真的好，但实际执行起来经常有各种小 bug。任务跑到一半卡住不走了，刷新又丢进度，来来回回搞了几次就不太想用了。

LEwal

—

从去年用到现在的老用户了，说句公道话——这团队确实在认真迭代，Canvas Cowork 出来后可以直接接 Claude Code 在画布上干活了，这个方向我觉得是对的。

Judy_Campbell16801

—

用了大半年，从最初的画布模式到 Neo 再到最近出的 Canvas Cowork 和 Matrix，团队迭代速度确实可以。画布这种交互形式对做复杂研究的人来说非常友好，分支对比、多模型并行、知识花园长期沉淀，这些能力别家确实没有。但槽点也不少：Web 端太卡了，节点一多就各种卡顿崩溃；积分体系让人没安全感，简单跑一次报告就几百积分没了，Pro 档每月也就两万积分，放开用几天就见底。简单查个东西我还是会切回 ChatGPT，Flowith 只适合做深度研究。

CoreyRobinson

—

别的不说，界面是真的好看。但好看不能当饭吃啊，Web 端卡得要命，节点一多就卡成 PPT，4060+i7 都带不动。

POwar2024

—

试了下 Matrix 的 0 人 OPC 概念，想法很超前。让它自动跑一个 YouTube 频道还真的跑起来了，虽然中间有些环节还要微调，但方向没问题。

DEwat

—

生图用 nanobanana pro 还行，但等图等到崩溃，经常一张图等十几分钟告诉我失败了，浪费感情。黑五本来想入会员的，想想算了。

ChristineSimmonsK

—

40多个模型随便切确实方便，不用在各家之间来回切换。但说实话大部分人根本用不到这么多模型，感觉军备竞赛大于实用。

d73oert

—

我最常用的场景是让 Neo 做竞品分析报告，然后把报告里的点拆开继续深挖。这个多层级探索的体验是别的 AI 工具给不了的。

EmmaWood_525

—

观猹了一圈，现在所有 agent 产品的核心技术都是上下文工程和交互设计。Flowith 的 context playground 是我用过最有创意的，Canvas 模式把上下文的玩法展现得淋漓尽致。

BSmith520

—

无限画布方式挺好用的，一次多图生成可以横向对比效果，但经常生图失败，一两次就算了，一直这样就是严重卡效率。

Lawrence834

—

找官方客服反馈问题，Discord 上回了一两次就再也不理人了。积分扣了事没办好，体验太差。

yellowrabbit243

—

试过几次 Neo，用它做了一篇行业研究报告的初稿。我在画布上扔了几篇参考文章、一个大概方向，它自动规划步骤、搜索信息、整理框架、生成报告，整个流程大几十分钟没断过。质量嘛，七十分吧，宏观结构和关键数据点都抓对了，但深度和专业性还是需要人补。算是很好的起点，省了我至少三四个小时的信息搜集时间。不过这次跑下来烧了快八百积分，按 Pro 档算差不多小两美元——不算贵，但心里没底。

WThompson_2021

—

上传文档让 Neo 处理，明明传上去了它说没看到文件。重新传了一遍终于开始干活了，等了一小时出的方案还行，结果一看执行记录——「由于用户未能提供原始文档，我将直接开始执行……」，血压当场飙升。积分没少扣，事情没办好，无语。

Scott_Garcia_66

—

作为创意工作者，画布多线程的模式非常对味。一个项目有不同的产出需求，画布上铺开来做很清晰。不过最近 AI 工具竞争太激烈了，我在 Flowith 上的使用时间已经减少了。

Johnny_Gonzales_7

—

先说结论：Flowith 是我今年用过最惊艳也最让人烦躁的 AI 工具。惊艳的是画布交互和 Agent Neo 的自主执行能力，第一次看到它自己规划步骤、拆解任务、一个一个节点跑出来的时候，确实觉得这才是 AI 该有的样子。烦躁的是稳定性和积分问题，任务跑到一半卡住是常事，生图失败率不低，积分消耗又快，免费 300 积分开个对比模式就没了。团队审美在线、迭代快是好事，但能不能先把基本体验做稳了再加法？

M_onique288

—

这种类似 ComfyUI 的工作流模式上手需要不短的时间适应，自由度很高但用起来麻烦了不少。容易爆内存、积分消耗高，上下文也容易丢失。模型调用失败率还挺高的。

q3xgd

—

Flowith 适合做复杂项目，研究、内容创作、项目规划都挺好。但学习成本确实高，习惯对话式 AI 的人看到画布结构第一反应是懵的。

SaturnSwap657_lab

—

画布式交互不是噱头，做完一个复杂项目之后就会发现回不去线性聊天了。但移动端体验差，手机上看画布基本不可用。

DianeCollins_X

—

多模型在一个面板里并排对比效果省心。以前在 ChatGPT 和 Claude 之间来回切，现在一个画布里搞定。

Diana_Carter168950

—

免费版 300 积分真的不够用，开个对比模式跑两个模型就快见底了。想认真评估产品还得充钱。

Olivia.Simmons_77

—

Agent Neo 的思考过程可以实时看到，能从中学到很多领域内的思考方法。既是效率工具也是学习工具，好评。

BrendaPhillips_66682

—

联网搜索效果不太行，信息源是硬伤。对于一个主打无限输出的产品来说，搜索质量直接影响最终产出的质量。

HenryHill777

—

用对比模式让 GPT-5 和 Claude 同时回答同一个问题，效果是好，但 GPU 占用拉满风扇起飞，回答速度也明显变慢。

moWAR

—

深度用了三个月，说个真相——真正能坚持用 Flowith 的人不多。大部分试了两天就回 ChatGPT 了。但对留下来的人来说，它是无可替代的。

FrankCollins

—

Flowith 的知识花园（Knowledge Garden）功能很强大，自动把上传的资料建关联。但处理速度太慢，超过 50 页的文档要等半天。检索效果也一般，有时候关联出来的东西跟主题完全不搭边。团队在画布和 Agent 上投入了很多，但知识库这块感觉还需要打磨，毕竟对深度用户来说，高质量的信息管理才是长期留存的理由。

天涯539

—

网上吹得天花乱坠，实际体验下来只能说想法很好完成度六成。每次新功能发布都让人期待，用起来总有各种小问题。画布模式下写长内容还是不顺手，不如 Notion 加 AI 来得直接。真正能坚持用它做复杂项目的人，我觉得是少数。剩下的大多数人试了试觉得「也就那样」就走了。

MsMustafaDağdaş_88

—

凌晨两点还在研究积分消耗。一个对话加上输出才一千多中文字就扣了四五十积分，用 Claude Sonnet 的档位。简单算了下，如果一天深度使用两小时，Pro 档两万积分大概够用一周半。这样的消耗速度让人不敢放开用，每次点生成之前都要想一下值不值。积分体系如果再不透明一点，我觉得很多中度用户会被劝退。

Katherine.KellyJr

—

Canvas Cowork 发布后第一时间试了，Claude Code 接进来在画布上写代码，体验还不错。就是跨工具协作偶尔会断链，比如生成图片的节点有时候会卡住不往下走。总的来说方向是对的，让人和 Agent 在同一个画布上协作这个理念比纯聊天框进了一步。团队在产品审美和迭代速度上确实强，希望稳定性跟上来。

SaraDavis168

—

Flowith Neo 出来后再次入坑，现在多了智能体模式，很多节点不用自己设计了。团队确实一直在进步，值得肯定。

Flowith

In-depth Report

User Reviews

Similar Products