About Stipple

We start with tenders across Australia and New Zealand: finding the work that fits, reading the pack, listing every requirement with the page it came from, and matching each one to the documents a supplier already has.

The reading, matching and drafting is done by AI agents working from the supplier's own documents. A person reviews every line before it goes out.

These are the checks that sit under a bid and are available on their own through our tools, REST API and MCP server.

Three things, all reading documents and text you did not write: AI-text detection that reports a likelihood with the specific tells behind it; citation fact-checking that follows every reference to see whether it exists and supports the claim attached to it; and document forensics that looks for tampering, checks a document’s internal arithmetic, and reads its structure and provenance.

Around those sit the practical tools that use them: structured field extraction, Australian 100-point identity checks, document-pack completeness, PII redaction, and adverse-media screening. Everything on the website is also available over a REST API and an MCP server, because an agent mid-task needs the same checks a person does.

Every tool here reports what was checked and what was found — not a truth verdict. A citation that resolves and supports its claim is a fact we can establish; whether the underlying claim is true is not. A document whose figures reconcile has passed a real test; that does not make it genuine. An AI-likelihood score is an estimate from writing style, and style is not authorship.

This is why results come with their reasoning attached. A score you cannot interrogate is a score you cannot defend to the person it is about — and these tools get pointed at real people’s work, applications and submissions.

The same principle governs what the tools refuse to do. They abstain on input they should not judge, they say when a check could not run rather than reporting a pass, and they publish their measured error rates in both directions — see the research, where the benchmarks include the false-positive numbers most tools leave out.

It depends on the tool, and the difference matters. The analysis tools — fact-checking and AI-text detection — do not store what you submit; only the result persists, keyed by a content hash. Documents uploaded for forensic verification are different: those files and their detailed results are kept for 120 days and then deleted automatically. If you sign in or create an API key we hold your email and a history of your checks — the verdicts, never the text or files behind them. The privacy policy and the security page set out the specifics, including what is logged and for how long.

Questions, corrections, or something we have got wrong: get in touch. If you are an agent developer, agents.md is written for you.