Meet Kamban
Hi, I’m Kamban, founder of Elephas. Before this, I built FlatGA, a tool that pulled website analytics into a single view. That project taught me something I’ve carried ever since: professionals don’t want more information.
They want to stop losing time to information they already have. Elephas is now used by over 10,000 knowledge workers across research, law, healthcare, and consulting. What keeps me going is that the privacy problem we set out to solve keeps getting bigger, not smaller.
What inspired you to create Elephas?
GPT-3 dropped and I was up late experimenting with the API. The capability was obvious. The workflow was a mess. You had to stop what you were doing, go to a browser, paste text in, copy the response back. No memory. No context. Completely disconnected from your actual work.
So I built a simple Mac wrapper that let you use GPT-3 from anywhere on your computer. Posted it on HackerNews without much expectation. It hit the homepage and blew up overnight. That was the signal.
But then something shifted. People started telling me what they were pasting into these tools. Client contracts. Patient records. Unpublished research. And nobody knew where any of it was going.
That’s when the real product clicked. The question stopped being “how do we put AI everywhere on your Mac” and became “how do we do that without your sensitive data ever being exposed.” That’s what Elephas is now.
How does Elephas help Mac users get more done?
The honest answer is that the productivity gains come from the privacy architecture. Most knowledge workers today are doing one of two things: they paste sensitive documents into ChatGPT and hope for the best, or they keep their most important work completely out of AI because they don’t trust it. Both options cost them hours.
Elephas gives them a third option. Before anything reaches a cloud AI, your Mac strips out personally identifiable information. Names, emails, phone numbers, identifiers. Stripped on-device before anything goes anywhere.
And we have a zero retention policy across the board: your content never trains anyone’s AI model, never sits on a vendor’s server, and never passes through a third-party reviewer’s screen.
When you can trust the tool with your real work, the productivity follows. Users save 5 to 10 hours a week just from that shift.
What sets Elephas apart from other productivity tools on macOS?
Three things.
First, the privacy model is structural. Before anything goes to a cloud AI, Elephas runs automatic PII redaction on your Mac. Sensitive data is stripped locally before it goes anywhere. We back that up with a zero retention policy: your content never trains AI models, never sits on vendor servers, never passes through a third-party reviewer’s screen. That’s three specific guarantees, not one vague promise.
Second, no hallucinations. Elephas only answers from what you’ve uploaded. It doesn’t pull from the internet. It doesn’t invent citations. If the answer isn’t in your documents, it says so.
Third, it works with the AI you already use. Most people are already on ChatGPT 5.5, Claude Opus 4.7, Perplexity, or Grok. Elephas doesn’t replace those. It wraps them with a privacy layer. You keep your model. You just stop sending raw sensitive data straight to it.
And for users who don’t want to use a cloud model at all, Elephas provides built-in local LLM models. Everything stays on your device. No external connection. Total privacy, no compromises.
Are there plans to expand Elephas to other platforms besides macOS and iOS?
The Mac focus is deliberate. The privacy guarantees we make, local processing, offline mode, PII redaction on-device, those require being close to the operating system. A web app can’t do what we do.
iOS has gotten more capable. Users can now query their work from iPhone and iPad with full sync. But the Mac is where serious work happens and that’s where we’re putting most of our effort.
What role does AI play in the functionality of Elephas?
Elephas isn’t a standalone AI. That framing matters.
You bring your AI model. Claude Opus 4.7, ChatGPT 5.5, Grok, whatever you already trust. We put the privacy layer between your sensitive data and that model. The data gets redacted on your Mac before it goes anywhere. The AI gets enough context to give you a useful answer.
You get that answer without your documents ever being exposed, and our zero retention policy means nothing from that session is stored, trained on, or reviewed by anyone.
For users who want zero cloud involvement, we provide built-in local LLM models. Everything runs on-device. No external connection required. The goal was AI that your most sensitive work could actually go through.
What are the biggest challenges for a SaaS product in the current market?
Trust. That’s the real one.
The AI market is full of tools making the same claims. “We take privacy seriously.” “Your data is safe.” Users have heard those lines enough times that they don’t believe them anymore. They’re right not to. Most of those claims don’t describe any specific architecture. They’re just reassurances.
Our challenge is showing that the privacy model is structural. It can’t be changed in a terms-of-service update. The architecture itself prevents it: local first, redacted on your machine, zero retention, then to the AI. That story takes more than a tagline to tell.
The other hard part is positioning a product that doesn’t fit one category neatly. Call it an AI writing tool and you undersell it. Call it a second brain and it sounds like a lifestyle product. The people who get it arrive right after they’ve pasted something sensitive into ChatGPT and realized they have no idea where it went.
What advice would you give to someone starting a SaaS product today?
Build for one specific person with one specific painful problem. Not a market segment. Build the thing that fixes those people’s painful moments first.
On privacy: treat it as an architecture decision from day one. Products that earn real trust over the next few years will be the ones where trust is structural, not a promise. A zero retention policy means nothing if it’s just text in a terms document. Build it into how data actually flows through your system.
Talk to your users constantly. Not to confirm your roadmap. To find out what they’re actually doing with your product. Some of our best features came from watching users work around problems we didn’t even know we had.
What do you think the future of productivity tools looks like?
The knowledge base becomes active. Right now most people’s second brain is really just a well-organized pile. Everything filed. Almost nothing used when it actually matters.
The tools that win will work on your actual documents, run at least partly on your device, and get more useful the more of your real work you put into them. Local LLM models are going to become a serious part of that picture, not a niche feature.
The bottleneck today isn’t capability. AI models can do remarkable things. The bottleneck is that users won’t feed their most sensitive work into tools they don’t trust. The future is private-by-default AI with real zero retention guarantees, not just promises.
Did you enjoy our interview? Do you have anything to say to our community?
Thanks for having me. These were good questions.
To the community: the most useful thing you can send us isn’t “I love the app.” It’s “I tried to do this and got stuck here.” Those conversations have shaped Elephas more than anything else. Tell us what’s broken. What’s missing. What you expected and didn’t get. That feedback is what makes the product actually work for the next person who finds it.
And if you haven’t tried it yet, if you’ve been keeping your real work out of AI tools because you don’t trust them, that’s exactly who we built this for. Give it a week.
