Ycode is a visual development platform that enables creators to build custom web projects without having to hire developers. Ycode's mission is to make visual development as powerful as coding by hand—just faster and more accessible.
This is the 5th launch from Ycode. View more
Ycode AI Agents
Launched this week
Connect Claude, OpenAI, Gemini or Grok to Ycode and use your preferred AI directly inside the website builder. Describe what you want to change, and the agent can update designs, manage CMS content, and build or improve components for you
Free
Launch Team / Built With
Ycode
Hey Product Hunt! 👋
We built Ycode to give designers and developers more control over how they create websites, and this launch takes that idea further.
You can now connect Claude, OpenAI, Gemini or Grok and use your preferred AI directly inside Ycode. The agent can help you refine designs, manage CMS content, and create or update components without leaving the editor.
Instead of generating a website once and starting over, the AI works with your existing project and helps you keep improving it.
We’d love to hear what you think, how you would use it, and what you would like us to build next. Thanks for checking out Ycode! 🙌
the bring-your-own-model part is the interesting risk here. we run a browser-driving agent and swapped the loop model across four candidates on one fixed task, and the failure that cost us most was not a model that errored. it was one that reported every step as successful while committing nothing, so the run read clean and the output was empty.
if your agents write to the CMS, that shape is worth guarding against, because a tool result saying it updated something is the model's own claim rather than the CMS's. do you read the record back after a write, or take the tool return at its word?
@whateverneveranywhere We guard against exactly that. Tools execute server-side against the project DB, so a tool return is the database's report, not the model's claim — errors propagate into the loop instead of being narrated away. Then every turn ends with an authoritative snapshot read back from the DB (state + server-compiled CSS), and that's what the canvas renders — so a silently failed write would be visibly missing, not invisibly "successful". Final backstop: everything is draft-first with a per-turn Changes card built from actual writes, and only a human can hit Publish.
Huge congrats🙌 on the launch @Ycode team.. giving creators choice over their AI model directly inside a visual builder is a game-changer qq Is there a version history rollback for AI edits, so if a prompt yields a messy layout we can instantly restore the canvas to the previous state?
Ycode
@vikramp7470 Yes! When the AI makes a change, it creates a checkpoint of your canvas first. So if a prompt gives you a layout you do not like, you can hit Undo right in the AI chat and instantly snap back to how things were before that prompt, no manual cleanup needed. 🙌
@tristan_mouchet1 Perfect, love that safety net, Huge win for productivity
Giving users the freedom to work with Claude, OpenAI, Gemini, or Grok directly inside the builder is a strong level of flexibility. Can users easily switch models within the same project while keeping the previous context?
Ycode
@nico_mandera Yes, absolutely! You can switch models anytime, right from the chat (even mid-conversation). Pick from Claude (Opus 5, Fable 5, Sonnet 5), GPT (5.5, 5 Mini), Gemini (3.1 Pro, 3.5 Flash), or Grok (4.5, 4.3), and swap between them whenever you like. The full conversation context carries over, so you can start on one model and switch to another without losing your thread or repeating yourself, and it's all working against your current project state either way.
@nico_mandera Yes, you can switch models right in the chat composer, mid-conversation. The session context carries over, so the new model picks up where the last one left off.
Since you support Claude, OpenAI, Gemini and Grok, are users bringing their own keys per provider, or would you consider a single unified endpoint to abstract the model routing and billing? Curious how you're handling failover when one provider's API is degraded.
@tian_yi1 Bring-your-own-key per provider. Each project can hold a key for Claude, OpenAI, Gemini, and Grok side by side — keys can be shared with the whole project or scoped "only me" so contributors bill their own accounts. We deliberately didn't put a unified endpoint in the middle: Ycode is open source and we don't want to be a billing middleman or add markup — you pay your provider directly and see exactly what each session cost in the usage badge.
No silent cross-provider failover (we'd rather not swap models behind your back mid-build). Instead, provider errors surface as readable messages in the chat, and if a provider is degraded you can switch models from the picker mid-conversation — the session context carries over, so the new model picks up where the previous one left off.
Love that I can use my own Claude/OpenAI key instead of buying AI credits with a markup. You pay your provider directly and Ycode charges nothing for AI usage.
Most AI websites tools optimize for convenience @lunenas . This seems to balance convenience with flexibility