Blink Agent Builder - You can now vibe code agentic AI apps
by•
Blink is the first vibe coding platform that builds AI agents. Describe what you want — Blink creates an agent that thinks, uses tools, and completes tasks end-to-end. Built-in web search, code execution, vector database, sandbox, and 180+ AI models. We used it to recreate Cursor, Perplexity, and Shortcut in minutes. Now it's your turn.
Replies
Blink.new
Hey everyone! 👋
Kai here, founder of Blink.
When we launched Blink, something interesting happened. People weren't just building landing pages and dashboards — they kept trying to build AI agents. Research assistants. Coding copilots. Support bots that actually do things.
But building agents meant stitching together APIs, managing context windows, handling tool calls, deploying infrastructure. Real engineering work.
So we asked ourselves: what if you could just describe an agent and have it work?
That's what we're launching today.
Blink Agent Builder lets you create agentic AI apps from a prompt. We tested it by rebuilding Cursor, Perplexity, and Shortcut — fully working, no code.
Everything's included: web search, code execution, vector database, sandbox, image and video generation, 180+ models, webhook tools for any API. Plus human-in-the-loop, multi-agent, and context engineering so your agent stays sharp.
Would love to hear what you think — and what agents you'd want to build. I'll be here all day.
—Kai
PS Grab our PH launch discount 🎁
Kalendar.work
@kf_builds All the best with your launch Kai!
Blink.new
@masoodtalha7 Thank you so much!
@kf_builds This is great Kai, congrats on the launch. How do you handle midsize corporate use cases that are niche?
Blink.new
@zolani_matebese Great question Zolani! There are a lot of midsized corporations using our service to create bespoke software tools - a lot of them are created to replace their internal CRMs and ERPs (like Salesforce etc.) People are often amazed at how cost-efficient and easy it is to personalize software for their own needs. The traditional enterprise SaaS model is starting to crack. For premium customers we have very good technical support to help them with any issues they can't solve on their own, so customers have been very happy.
@kf_builds hey, this is Saul; I’m a graphic designer, and I used Blink for the first time this week. I built a nutrition app for cyclists called BackPocket.
I was honestly impressed by the speed of execution compared to other similar platforms. One thing worth noting is the limited number of credits during the trial period, although I also understand the high costs involved in LLMs wrapping and building agents.
I also think there’s room for stronger visual curation, as many templates tend to look quite similar. This could potentially be handled or curated by real art directors—something hybrid between AI and human touch (maybe something to consider in the future).
I’d love to get in touch and talk. Congrats on the product.
Blink.new
@saul_suaza Saul, thank you for trying it out and for the thoughtful feedback. Indeed, visual design is very important to a lot of users and that's why we invested heavily into building the best-in-class AI designer agent right within Blink. We are curating lots of user-generated examples today, take a look at https://blink.new/explore
And of course we would love your help to make it better. Also hoping to launch more templates for users soon.
Regarding free credits - they reset daily! So you can always come back tomorrow to try again if you want; upgrade to a premium plan to build without limits whenever you are ready.
@kf_builds congrats on the launch! Is it possible to access the code as well?
Blink.new
@austin_heaton Thank you Austin! Do you mean the source code of agentic apps made by Blink? Yes of course. You can download the code from the project's code tab and also able to achieve two-way sync with your Github account.
@kf_builds Congrats on the launch and running lots of upvotes!
As someone running a remote services business, we see a lot of excitement around agents but the real challenge is reliability once they’re embedded into day-to-day workflows.
How does Blink handle guardrails, human-in-the-loop, and error recovery when agents are used for repeatable operational work not just one-off tasks?
Blink.new
@aileen_gallinero I totally resonate with your comment, as our company is also fully remote! Having reliable agents is critical as your operations rely on it (ideally). We've built and deployed lots of agentic apps to help us with marketing and operations tasks and it's done wonders.
Sharing some example use cases:
generating marketing copies - before our agents went online, team members and interns would all run with their own process, some are just using chatgpt with generic prompts. we standardized everything into verified and tested agents for copywriting for different use cases and the output and effectiveness of marketing content has been through the roof.
customer support answers: building knowledge-base vector database and using it to answer customer queries is something that blink agent builder does very well out of the gate. using it we are able to bring our support up a notch and helping customers resolve issues significantly faster.
Guardrails, HITL, etc. - these can be achieved easily with additional fast and lightweight agent / ai generateObject calls that serve as verifiers. Ask Blink agent to help you build them!
Congrats on the launch — this looks like a dream tool for solo builders and small teams who want to go from idea to production-ready agentic apps in minutes, not weeks.
@zeiki_yu I was thinking about the same thing, speed is great, but removing the hidden complexity feels like the real unlock.
Blink.new
@zeiki_yu Thanks! That's literally it - we got tired of AI tools that generate code that doesn't actually work at scale. Wanted something that solo founders could actually use to ship real products.
Interesting direction. The “describe an agent and it just works” idea sounds powerful, but also hard to get right. Curious how much control users have once things get more complex.
Blink.new
@shreya_chaurasia19 Great question, honestly, this was one of our biggest design challenges.
The "describe and it works" part is the entry point, but you're right that real-world use cases get messy fast.
We've built in layers of control for when you need them. You can configure which tools your agent has access to, manage context windows, set up human-in-the-loop approvals for critical actions, and even orchestrate multi-agent workflows where different agents handle different parts of a task.
Blink.new
@shreya_chaurasia19 Great question Shreya! Users have the complete control over their agentic ai app behavior after initial build. This is because Blink is an ai app developer that can always follow user requests to make changes and iterate on the app. Think of it as your own expert dev team.
Elser AI
The claim about recreating Cursor and Perplexity in minutes is bold. Does it actually generate production ready code, or does it need significant cleanup and debugging afterward?
Congrats on the launch!
Blink.new
@sarahjiang Yes I couldn't believe it at first either but really, we did it.
Check out what Blink made:
Excel agent: https://shortcut.sites.blink.new/
Powerpoint agent: https://gamma.sites.blink.new/
Research agent: https://perplexity.sites.blink.new/
Coding agent: https://cursor.sites.blink.new/
Thanks so much for your support!
Blink.new
@sarahjiang Oh and yes it generates production ready code, which is a unique selling point compared to others like Lovable, Replit and Base44. The code generated is well decomposed, significantly more scalable and enables complex use cases and full-on commercial software. It's a new era.
Bundling web search + code execution + vector DB + sandbox into one “agent app” builder is the right abstraction. The scale pain is runaway tool calls + nondeterministic runs as agents get more capable; best practice is a run ledger (all tool I/O + model/version pinning), budgeted tool policies (timeouts/quotas), and eval gates before deploy 🔥 Are agent workflows represented as an explicit graph/state machine under the hood, and can users export/replay a run (for CI + regression) across different models?
Blink.new
@ryan_thill Yes you are right! Internally we already have very robust eval practices and with our users this is on our roadmap to help them achieve really great agentic building best practices as well. Thank you for the thoughtful input, well said!
Blink.new
@ryan_thill Now regarding your question, agents are defined as the Agent class and you can create multiple agents in your codebase and call them whenever you want. This object oriented approach achieves greatest flexibility and ease of management.
The best part is even a non-technical user can easily read it and edit the prompt, tools array, etc. if needed.
Blink.new
Hey Product Hunt!
I'll be honest – I didn't fully get the agent hype until a few months ago.
Everyone was talking about AI agents like they were the future, but when I actually tried to build one for our marketing workflows, I gave up after two days. Too many moving parts, too much technical overhead. I just wanted something that could research competitors and pull data from the different channels and data sources we have, not a PhD in API management.
When Kai told me we were building the Agent Builder, it immediately resonated – because I'd lived this exact problem as a user.
Now we have agents running all over our stack. One pulls engagement data and writes analysis. Another monitors our community channels and flags important conversations. Another helps draft and iterate on copy. They're not perfect, but they're useful.
The bigger realization is that whether you're rebuilding something like Perplexity or Cursor, or just automating a specific workflow, the barrier has always been the same – the technical complexity of making agents actually work. That complexity shouldn't be what stops good ideas from existing.
I'm really curious what you'd build if the technical overhead wasn't in the way. What's the agent you've been wanting to create but haven't had the time or resources to actually build?
Around all day for questions or just to chat 🚀
Agno
Congrats on the launch. Its slick and fast.
Great experience
Blink.new
@monali_dambre1 Glad you liked it!
Blink.new
@monali_dambre1 Thank you Monali! It's great to hear you're liking it. Let us know how we can make it better for you :)
Blink.new
@t0ny_ns That's a great idea! With Blink you can do that but for Chrome extensions likely it will involve a bit more hacking and testing to access the user browser and take actions such as scrolling and clicking. We'll likely build this next as a challenge!
Congrats on the launch just curious how do you handle long running agents memory across sessions ?
Blink.new
@farhan_nazir55 Great question, Farhan. Agents created with Blink run on our managed backend service and can handle long-running tasks since it's a node based service. Memory across sessions is achieved easily through Blink SQL database which every Blink project gets out of the gate.
Congrats. In the screenshots what prompt did you give to create that excel clone?
Blink.new
@ahmedrana Hey, you should be able to get a prototype with "Build me an Excel Copilot like Shortcut". It might require a few iterations to perfect. Excited for you to try it out!
Blink.new
@ahmedrana Check out this page for the prompt and build process!
https://blink.new/p/shortcut