Kombai - The first AI agent built for real-world frontend tasks
The first domain-specific AI for frontend tasks beyond prototypes. Build entire app UIs, refactor code, or create components from Figma, text, image or code. Understands your codebase. Works in your IDE. Vastly outperforms generic agents + frontier models.
Replies
Kombai
Hello Product Hunt!
The Kombai team is back. We’re thrilled to launch our new domain-specific agent for frontend tasks and can't thank you enough for checking us out! 🙏
WHAT WE ARE LAUNCHING
Today, we’re introducing Kombai’s new avatar, the first domain-specific AI agent for Frontend development. It's a major evolution from our popular Figma-to-code tool, PH’s most upvoted devtool in 2023.
THE MOTIVATION
Today’s models, and by extension general-purpose coding agents, are great at tasks represented in benchmarks, e.g. SWE bench and Terminal Bench, as well as competitive programming platforms, e.g. Leetcode and Atcoder.
However even the frontier models struggle with real-world Frontend tasks, e.g. building a large component or a page using a given set of components and libraries; while adhering to best practices. For example, in our evaluation of large Frontend tasks, SOTA models were unable to produce compilable code 30-40% time, even when coupled with best-in-class MCPs.
On the other hand, AI prototyping tools are great for quickly producing complex-looking V0 (no pun intended) for a page or app. But their outputs can rarely, if at all, be used for long-term production codebases.
KEY CAPABILITIES & FEATURES
Kombai is laser-focused on the frontend domain and has highly specialized optimizations that make AI work for Frontend codebases. In real-world frontend benchmarks, Kombai consistently beats general-purpose agents, even when coupled with frontier models and popular MCPs.
Some of the interesting features:
Human-like Codebase Understanding: Learns your repo like a human developer to get parametrized, domain-specific context, going beyond simple embedding search.
Best-in-class Figma Interpretation: Uses an advanced engine to interpret complex, imperfect Figma designs that humans create.
Understands best practices of 30+ frontend libraries: Uses built-in, human-tested context engine.
Plans before it codes, preview before you save: Get editable plans before coding, see visual previews in a sandbox, and iterate safely with a no-write-by-default policy.
Smart Model Routing: We continuously benchmark frontier models, so our agent automatically picks the best one for your task, saving you time and cost.
NEXT STEPS
We are committed to improving with your help. Please give the agent a try and share your feedback. Our team will be live all day to answer your questions. We can't wait to build the future of frontend development with you!
-----
Dipanjan
Co-founder & CEO, Kombai
Sprinto
@dipanjan_dey very cool! Congrats on the launch.
@dipanjan_dey Congratulations on the launch! Love the domain-specific approach - general-purpose agents often miss the nuances of real-world frontend development that you're tackling.
@dipanjan_dey It's not clear from your description if this works with Android/iOS UI or web only?
Kombai
Folks, we have a generous free tier for this launch. Very excited to get all your feedback. 🎉🎊
Kombai
Kombai turns UI into production-grade code, and we’ve spent months fine-tuning it with real-world workflows in mind. We’re launching today and would love your feedback—what you love, what could be better, anything!
Huge shoutout to the rest of the team for making this happen. I’m here all day if you have questions, want a demo, or just want to chat code.
Thanks for checking us out! 💙
Kombai
looks like a serious step! does it adapt to project specific conventions?
Kombai
@mohsinproduct Yes, it does! It can understand the tech stack, best practices, and component library from your repo. It also has access to search tools - so it can find relevant code and config files needed for a specified coding task
Kombai
@mohsinproduct Great question! and yes! Kombai scans your project and identified your tech stack and makes sure your tech stack is honoured when it generates code. You can furthermore add custom instructions as well!
@avirup_mukherjee @dipanjan_dey thx, does it also flag inconsistencies or suggest improvements if existing code deviates from best practices or the defined conventions in the project?
Tidyread
Congratulations on the launch of Kombai! 🎉 The smart model routing feature is intriguing. Can you share more about how Kombai determines the best model for a specific task? Is there a way for users to override the automatic selection if needed?
Kombai
@jaredl great question!
LLM models from major providers evolve rapidly, and their benchmarks often don't represent real-world frontend tasks. Discovering the best model for a given job is often time and cost-intensive.
We continuously benchmark the latest frontier models to understand their capabilities and limitations on various frontend tasks. Based on our regularly updated benchmarks, Kombai's agent has built-in opinions on the best model for different task categories. Once it receives a task, it first quickly evaluates its category and then uses the most suitable model for that category.
Currently, users are not able to override this. A part of this reason is cost; some of the Frontier models we use for specific tasks can be hugely token-expensive for some other tasks - e.g. indexing a large codebase with Claude's latest models will be prohibitive. There are also some challenges on the interoperabilities of some models that we chain together for output and cost efficiencies.
However, we are actively collecting users' feedback on this; and are open to putting model-control in our roadmap, if stars align :-)
This looks seriously powerful - amazing work by the team!
One question though how deep does kombai’s codebase understanding go? Can it reason about folder structures, custom hooks, or internal design systems?
Kombai
@lak7 Thank you so much! Kombai is built to truly understand complex codebases. It can reason about folder structures, custom hooks, and even your internal design systems. We wanted Kombai to fit right into real-world workflows. If you have a specific setup in mind, please feel free to reach out!
Nanonets
New frontend framework comes every other day, specialised frontend AI agents just make sense, otherwise it's a lot of manual context feeding and hand holding.
Kombai
@aman_rusia that was the idea! To have an AI agent thats fine-tuned for frontend tasks!
Jupitrr AI
Kombai
@lakshya_singh thanks!
Extrovert
This is dope! giving it a try for sure
Kombai
@oleg_sobolev thanks a lot!
I really appreciate this philosophy because it solves the hidden "cleanup cost" of most AI dev tools—where you end up spending more time refactoring messy, non-semantic code than you saved. Prioritizing augmentation over automation, and positioning Kombai as a "senior colleague" that produces genuinely maintainable code, is the perfect answer. This is how you build the trust needed for a true human-AI partnership.
Kombai
Thanks a lot @alexis_luo , really appreciate it!
Triforce Todos
Kombai makes it so easy to turn Figma designs into clean frontend code. It looks very useful for saving time on projects.
Kombai
@abod_rehman Really appreciate it! It sure would cut down on dev time while delivering clean code. Please feel free to reach out for any questions or feedback.
Haimeta
Whoa, the fact that Kombai actually *understands* your codebase and works right in your IDE is wild—no more copy-paste chaos, tbh. This is next-level smart!
Can it copy figma designs? Or just for react code?
AppStruct
Impressive work team Kombai!
How does the agent maintain design consistency when generating new UI components across large projects?
Huge congratulations on the lunch, Good luck!!
Looks really good! Will give it a try
Eternal AI
Congrats on the launch! So exciting to see a domain-specific AI for FE devs. Do you have plans to add features for automatically updating code based on new FE frameworks or libraries as Kombai grows?
Kombai
@sig_eternal Thank you for the encouragement. We will keep on supporting the latest versions of popular libraries and frameworks. For example, you can already use Kombai for MUI v7 and chakra v3, libraries that came out after the cut-off dates of many latest SOTA models.
Growth Hacking Resources
Kombai
@thehirenthakkar thank you so much!
Ugh, I was just stressing a few days ago about getting a pixel-perfect front-end for my designs. Tried Cursor and Lovable but was kinda let down. Off to try this new thing now, let's see!