Sealion Projects

Tint every sentence by where it came from

A sample memo where each sentence is tinted by origin, opens its trace chain, and can be filtered to show what survives.

tool ยท desktop-firstPublished 13 Oct 2026ai-documentsprovenancereviewtrust

Interactive prototype

Product pitch

The problem

When a team drafts a memo, design brief or research summary with a model, the finished text hides its own history. A reviewer asks where a number came from and nobody remembers whether it came from the export, the prompt, the model's guess or a late human edit. The cost shows up later, in a review or an audit, when someone has to defend a claim they cannot trace. Existing track-changes tools show who typed what, not which file a figure came from or which sentences have no source at all.

Why now

Drafting with a model is now routine in ordinary teams, and the output reads equally confident whether it is grounded or invented. Reviewers have started asking for the receipts, but the only tool they have is rereading the chat log. Regulated teams and anyone handing documents to a client are the first to feel it, because they need to say what was machine-written and what was checked.

What it is

A document view that tints every sentence by origin: source file, prompt, model output or human edit. Each sentence opens a trace chain from the original file through each model pass to the final wording. A filter removes one origin to show what survives, and sentences with no source in their chain carry a plain label. It does not score quality or detect AI text. It records and displays history; it passes no verdict.

Who it's for

Product leads, analysts and compliance reviewers who sign off on documents someone else drafted with a model. It is not for teachers checking whether a student used AI, and it is deliberately not a detector. That use is adversarial, error-prone and needs a different product. It is also not for solo writers who do not need to defend their drafts to anyone.

How it makes money

Nothing yet, and that is correct at this stage. The value only exists if the history is captured while the work happens, which means integrations with the editors and chat tools people already use. Until a team is willing to route drafting through one of them, pricing is guesswork. If the capture works, per-seat pricing for reviewers is the likely route, since they are the ones who feel the pain.

Go to market

Start with compliance and risk reviewers in fintech and health startups, who already have to evidence how a document was produced. A reviewer who sees a tinted memo with a single untraceable sentence will forward it to the author that day. Offer a free reviewer view for documents exported from one popular editor, so adoption starts with the person asking the questions, not the person drafting. This demo does not capture history; it shows only what the view would make legible.

A focused prototype from Sealion Projects. Anything it saves stays in this browser, on this device.