Work with agents inside the same folders and app that owns design, content, and assets. The agent reads your real project, proposes edits, and waits for your review.
What you get
We have built extensive tooling that enables agents to efficiently create, edit and review your site.
Users will visually see what's changed, and decide to keep it or not
We don't want to be a middleman, bring your own key for Anthropic, OpenAI and OpenRouter.
Chat history stored with the site, building on the context of the AI.
Why this outcome matters
Prompting a coding tool can ship a first page. Ongoing marketing work needs review, constraints, and a place where changes become publishable history.
Guessing from a pasted snippet creates drift. Grounded edits on the live project keep the site coherent.
Splitleaf does not insert itself as a middleman tax on every token. You bring the key. You pay the provider. The agent stays a feature, not a new subscription layer.
How it shows up in the product
The agent works on the currently open site. File access stays sandboxed to that site root.
Edits are staged with baselines. Writes use precise replacements. Nothing is committed to Git by the agent. You review before publish.
Connect your model API key in site settings. Keys are stored for that site's agent connection. Splitleaf is not the LLM bill.
Sessions and staging live under .opt-one/agent/ in the site. Usage can follow the site across devices. The conversation stays near the work.
In practice
Install Splitleaf on your machine and sign up free, no credit card, ready in minutes.
Every feature is unlocked while you try it out, the designer, the CMS, the AI agent.
Click and connect with a few clicks to store the project in GitHub, and publish it on your own domain with Vercel.
Keep what's right, undo what isn't, and publish through the normal review flow whenever you're ready.
FAQ
Keep AI speed with a human still on the merge button.