Skip to content
MakerGain
← Back to project

After switching off Claude Opus: AI Flow Chat puts content systems on a whiteboard—and usage bills on the app owner

Edited by MakerGain

AI Flow Chat lets creators put videos, PDFs, websites and other references on a whiteboard, group them to define a style, and build a repeatable content system delivered through scheduled tasks or CMS embeds. Embedded-app usage is charged to the owner. After a VC-backed startup failed, founder Alexander Van Le turned to bootstrapping, disabled Claude Opus because it cost ten times as much, and replaced three usage tiers with a single credit system to control model-call costs.

Embed an AI app on your website, let visitors use it, and the bill lands in your account.

AI Flow Chat spelled out that rule in its v0.4 update. Founder Alexander L. explained that users could embed their finished AI apps in their own CMS using an iframe, then added that all embedded-app usage counts against the owner's account. Once the daily credits run out, the app is disabled until the next day. (AI Flow Chat Features Update v0.4[1])

That sounds more like a usage notice than a feature pitch. But it clarifies the product: AI Flow Chat sells a repeatable content system that can be embedded in a website. Once it is there, someone must keep paying for each run.

The website positions it as an “AI canvas” for creators and marketers, generating business content from existing viral videos, PDF playbooks and websites. (AI Flow Chat - Scale Your Content with AI[2]) Its “How it works” section describes the steps: put references—videos, PDFs, websites, TikToks, Reels and Shorts—on an AI whiteboard, then group resources to define a style and tone. Users can build repeatable AI systems without code, use community templates, or save their own prompts in a personal library for reuse. Notion integration brings documents, wikis, meeting notes and SOPs into the workflow. According to the website, it can connect multiple Notion workspaces, including personal, team and client workspaces. (Same source[2])

Image: official product website[2], obtained on 2026-11-18.

Grouping references to define style and tone is a documented use. The resulting system can be embedded in a CMS for visitors to use, as described in v0.4, or run on a timer using app schedules. (v0.4 update[1] · [2])

The cost logic sits inside that workflow. Model calls generate real bills, and v0.4 explicitly assigns embedded-app usage to the owner. Several of the founder's later decisions relate to that fact.

Alexander Van Le says he previously cofounded a VC-funded FinTech company that failed. He says the experience left him somewhat biased against the VC mindset; he now prefers using capital to scale rather than to survive. (Bootstrapping a $20k/mo AI portfolio after his VC-backed company failed[3]) The interview introduction says he had been an employee, subsequently went viral, quit, and switched to bootstrapping AI Flow Chat and Starpop. (Same source[3])

On the product side, that left a series of changes focused on usage costs.

In v0.7, he replaced the three usage tiers—Fast, Standard and Premium—with a single credit system. At the time, he listed 11,000 monthly credits for Basic and 33,000 for Pro, with GPT-4.1 costing 10 credits per call and GPT-4.1-nano 1 credit. In the same update, he acknowledged that he had disabled Claude Opus because it cost 10 times as much as other Premium models. The new credit system was intended to make AI Flow Chat more sustainable. (AI Flow Chat Features Update v0.7[4])

“More sustainable” is a mild phrase for a substantial decision: switch off the most expensive model and move from three tiers to more granular credits. Deductions could now correspond to particular models and individual calls, with sustainability as the product goal.

The current website has changed its pricing again. On its 2026-11-17 page, Basic was $29 a month, with the first 7 days at $0 for new customers, 8,000 monthly credits and 5 app schedules. Scale was $59 a month with 18,000 credits and 10 schedules. Pro was $119 a month with 40,000 credits and 15 schedules, plus removal of embedded-app branding. Extra credits have two prices: Basic and Scale pay $6 for 1,000; Pro pays half that rate, receiving 2,000 for $6. (AI Flow Chat - Scale Your Content with AI[2]) Basic and Pro's monthly credit counts have therefore changed since v0.7.

Another charging model offers a comparison. FLORA Starter costs $18 per seat per month, including text, image and video models, API and MCP access, and real-time collaboration. Its documentation says each seat includes an $18 monthly usage budget; Pro at $50 and Max at $200 each include a matching budget, drawn down at published model rates. (FLORA Pricing[5] · How Pricing Works[6]) Neither company can escape the real cost of model calls. Their meters differ: AI Flow Chat turns calls into credits deducted per use; FLORA puts a US-dollar budget into each seat's monthly allowance. Which fits better depends on deployment, call volume and budget arrangements. Price alone does not tell the whole story.

Now look back at growth. In v0.4, the founder reported 255 users, up from 58 at the previous update a month earlier. That update also offered the code FOUNDINGPROMPTENGINEER, reducing Basic to $10 a month with that price retained permanently. The offer expired on 2025-06-30, was limited to 10 redemptions, and had not sold out at the time. (AI Flow Chat Features Update v0.4[1]) Those are historical figures and should be read as such.

The interview introduction reports combined MRR of $20k a month for AI Flow Chat and Starpop. (Bootstrapping a $20k/mo AI portfolio after his VC-backed company failed[3]) The monthly recurring revenue figure belongs to the two products together. It is not profit, and it does not reveal AI Flow Chat's share. Public materials do not disclose its standalone MRR, gross margin, cost structure or refund rules. They therefore do not establish whether this business makes a profit.

Its commercial shape is roughly this: content creators place scattered references on a whiteboard, define a tone once, then repeatedly run the system through schedules and apps embedded on websites. The payer is whoever builds that system—perhaps the creator, perhaps someone producing content for a client inside the client's Notion workspace. The output can be embedded in a CMS for visitors or triggered on a schedule.

That arrangement makes the purchase clearer: a content system that runs on metered usage, rather than an unlimited chat box.

Continue reading

AI publishing
The first pound came before the website: Eclipse makes content approval a monthly system

Mark Ridgeon received his first pound before he had a website or clickable product. The problem he saw was not a lack of drafts: even after stitching several tools together, agencies could not defend their content claims to clients or boards. Eclipse therefore connects research, approval, publishing, and proof of visibility. It began with that payment; the website and clickable product followed, while a defensible, auditable approval process runs through research, publishing, and visibility evidence.

AI publishing
They decided to close a product that still had users, and build a marketing department for a company of one

Adam Nathan and Eddie Wu decided to close Almanac while it still had users and focus on Blaze. Built for teams of one, Blaze handles weekly cross-channel content creation, review, scheduling, publishing, and performance checks: users decide, Blaze executes. A hundred customer interviews helped them narrow the audience to small-scale founders, with brand feel as their differentiator. Public materials leave open how approval and automatic publishing interact.

AI publishing
A small task, two businesses: how PostNitro splits up carousels

PostNitro turns articles, topics, or X posts into carousels: it proposes an outline, then after confirmation writes copy, applies a brand kit, and exports or publishes to social platforms. Seerat left on May 31, 2025; Muneeb bought his shares and ran the business alone, with his wife handling marketing. The blog said v2 was expected in July, with a beta for a small group first. The editor is also embedded in other platforms; the API is available on paid plans and charges credits per slide.

AI productivity
Klipy, the sales AI that waits for your approval

Klipy takes over follow-up after sales meetings. It captures customer conversations from email, meetings, and instant messaging, drafts follow-ups and CRM updates, and prepares for the next meeting. Follow-up messages go out only after the salesperson approves them. Founder Jung Hong Kim is a Korean serial entrepreneur who took on debt when COVID sank his retail venture. He met his cofounder while consulting and launched Klipy.ai in three months. It went live in November 2024 as an AI CRM that logged communications automatically. In June 2025, it shifted toward carrying out next steps for sales teams and tied billing to specific actions users approve.