Whalesync has always been about connecting your tools together. Back in 2021, we had an idea for a true "data hub" for your business. We started by building a data syncing tool, but the ambition was always bigger than that.
We were too early. The market wasn't pulling us toward a data hub, so we kept building sync features.
AI changed that overnight. Every company is trying to use AI for as much as possible, and AI needs context. The race is on to figure out how companies can best build their "AI brain": the unified body of company knowledge that AI agents draw from to do useful work.
We started building Scratch because we saw the need for a data hub to power that AI brain. We're not the only ones who noticed. Glean is building an AI knowledge hub. Notion's recent update repositions it as an AI-powered content hub. Dropbox Dash promises universal AI search across your tools. The category is clearly emerging, but every one of these is a cloud-side hub. Your data goes up, the AI runs there, and the answers come back. That doesn't benefit from the incredible power of local tools like Claude Code.
What Scratch does
Scratch is the missing piece for building your team's AI brain. It lets you:
- Download all of your content to a folder on your computer so it's dead simple to work with using your favorite AI agent.
- Edit with the AI agent of your choice: Claude Code, Cursor, whatever you like.
- Review changes using our content viewer, accepting or rejecting individual edits.
- Publish the approved changes back to your team's apps.
Even just the first piece, getting your content local, makes Scratch better than anything else we've seen and dramatically accelerates building your AI brain. But what makes it 10x better is the safe editing loop: download, edit, review, publish.
Why the loop matters
We've found that you usually need to work with your AI agent multiple times to get a set of changes exactly right before you publish. That iteration needs a safe place to happen.
This parallels the software development process that engineers have taken for granted for years: edit locally, review the diff, ship to production. It was never really needed for content because content was edited by humans in small batches, with publishing happening directly in the source application. AI agents have changed all of that. Suddenly content workflows look a lot more like code workflows, and they need the same safeguards.
Why we built it this way
A few observations shaped what we built:
Local is where all the power is. Hard drive space is cheap. Claude Code can ls, wc, and grep a huge codebase like nobody's business. We've experimented with searching data over MCP versus having everything local, and local almost always wins. The people at the bleeding edge already know this. Garry Tan, who runs Y Combinator, built gbrain (his personal AI brain) as a pile of markdown files in git, synced to a local database, queried by Claude Code over MCP. That's roughly the same shape as Scratch and they pair well together.
AI speaks raw JSON. For your AI brain to work, AI needs access to your data spread across your SaaS apps. Models have been trained on the API responses of these apps. Meeting AI where it lives means handing it the raw structured data.
Read-only is just the beginning. Powerful workflows almost always need editing. A knowledge hub that only lets AI read is leaving most of the value on the table.
Editing needs review. Unless you just want to push out AI slop, you need humans in the loop. The review step isn't a nice-to-have; it's the thing that makes the whole loop trustworthy.
That's Scratch. The bridge between your SaaS data and your local AI tools, with a safe editing loop baked in.