Dishaya / Why
Why We Are Building Dishaya
A short letter on work you must stand behind, and what it takes for AI research to be worth sending.
There is a difference between work you produce and work you must stand behind. A note to yourself can be roughly right. A report with your name on it, sent to a client, a board, a professor, or a regulator, cannot. The moment someone else will act on what you wrote, every sentence becomes a small promise, and the cost of a broken one lands on you, not on the tool that wrote it.
Current AI is remarkable at producing and unequipped for standing behind. It writes with the same confidence whether it is right or wrong, and it does not tell you which paragraphs are which. So the honest workflow becomes: generate in seconds, then recheck by hand for an hour. Anyone who has traced a plausible statistic back through search results at midnight knows how this ends. The rechecking erases the time the fluency saved, and the people who skip the rechecking are quietly spending their own credibility to save it.
We kept returning to one question: what would have to be true for AI research to be sendable? Not impressive. Not fluent. Sendable, as in you attach it, press send, and sleep.
Three things, we think. The receipts have to be attached: every claim carries its source, and you can tap through to the exact passage it rests on. The failures have to be disclosed: when a claim could not be supported, the document says so plainly instead of smoothing it over. And the cost has to be honest: you should know what a run will cost before you start it and what it did cost when it finished.
So we built Dishaya that way. Every claim is checked against its exact source passage before it is written. Each one carries a label: verified, partial, unverified, or contradicted. A source ledger lists them all, claim by claim, and a Confidence Ledger closes every report with how the evidence held up. What fails the check is disclosed, never hidden, because a tool that only ever says "verified" is a rubber stamp, not a check.
We also publish what the labels do not promise. Verification is a rigor tool, not an oracle. A cited source can itself be wrong, and a label tells you how well a claim is supported by the source it cites, not whether the universe agrees with it. We keep those limits in a public methodology, because trust built on an overclaim is just a slower way to lose it.
A word on the company, because the company is part of the product. Dishaya is small, independent, and early, and we say so plainly. We keep it cheap to run so that it answers to the people who use it rather than to a burn clock, and we would rather grow slowly than grow in ways that cost trust: no dark patterns, no invented traction, and no training on your content, ever. Tools in this category tend to die by being acquired, pivoted, or quietly degraded. We are trying to build the boring alternative: a product that works, priced honestly, still here and still trustworthy in its tenth year. Most decisions here get made by asking what the tenth year needs, not what this quarter wants.
None of this is finished. The labels will get sharper, work packages will cover more kinds of work, and some of what we believe today will turn out to be wrong. That is where you come in. Try it: one question in, and a report, a deck, and a source ledger out, with every claim carrying its verdict, all of it waiting in your Library. Vote on the public board, which is a real roadmap that moves with real votes. And when we get something wrong, tell us. A company that asks you to check its work has no business being precious about its own.
That is why we are building Dishaya. Work you must stand behind deserves a tool that can stand behind itself.
The Founder, Dishaya