Chiron applies the encoded insights of your partners who have underwritten similar deals, identifies what doesn't hold together, and comes up with the questions to ask. No documents are ever stored.
An 80% gross margin in health tech SaaS usually has implementation cost buried in COGS. Ask where onboarding labor sits before you believe the margin bridge.
D. Keller, 22 platform dealsverbatim excerpt, interview 14, 31:06
The lens is accumulated judgment in structured form: how a good investor reads an 80% gross margin in health tech SaaS, what the third add-on tells you about a roll-up, what a reclassification looks like. It is built from interviews about deals people actually evaluated, including the ones that went wrong, and from your firm's own past decisions.
Not an anonymous assertion. Every insight carries the name and credential of the person who said it.
Each insight links to the exact words it came from, with an interview reference and timestamp, so you can check it.
Health tech is not multifamily. Insights fire only on deals in their category, and your lens is personalized to your firm.
Upload the CIM, the model, and the supporting documents. Chiron extracts terms, capital structure, projections, and operating metrics into one comparable record, each item traceable to the document it came from. Then it evaluates that record: structural inconsistencies the documents themselves reveal, plus flags triggered by your lens.
One structured, comparable record of the deal, with a source on every extracted item.
What does not hold together, and where a lens insight fired, who says so and why.
Every flag becomes a written question. Your team marks each resolved, open, or dropped, and records what came back. That record sharpens the lens on the next deal.
Every finding carries a weight, a source, and the name of whoever taught you to look for it.
A general model is very good at reading one document once. Everything below is what breaks when an associate does that forty times a year, across a team, against real data rooms, and has to defend the answer in an investment committee.
| Pasting documents into a chatbot | Chiron | |
|---|---|---|
| How much it can read | A handful of files before the upload limit, the context window or the request timeout stops you. Big underwriting models are the first thing to break. | Files are parsed in your browser, so size stops mattering. The run streams for as long as it needs and reports what it is doing, file by file and page by page. |
| Whose judgment | The model's. It has read everything ever written about private equity and underwritten none of it. | Your partners'. The lens is built from interviews with people who have actually done these deals, and each insight carries their name, credential and the verbatim excerpt it came from. |
| The same answer twice | Ask again tomorrow and the emphasis moves. Two associates asking the same question get two answers, and neither can say why. | The score is computed by code over the stored record. Same record, same number, every time, for everyone on the team. |
| Comparing deal 40 to deal 1 | Forty chat threads. Nothing lines up, because every conversation went somewhere different. | Every deal produces the same record against the same checklist, so a category becomes comparable once you have run a few of them. |
| What it did not see | It answers confidently from whatever it was handed. It does not tell you that the working capital build was missing. | Coverage is scored and shown. The confidence band around the score widens when the documents are thin, and the missing items are listed by name. |
| Defending the answer | A paragraph of prose. Ask where a number came from and you get a plausible-sounding reconstruction. | Every extracted item carries its source and a written note on what it means. Every score opens into a ledger whose lines add up to the number, each attributed. |
| Whether it improves | It does not. The fortieth conversation starts exactly where the first one did. | Every question you close records whether the answer mattered. Insights that keep mattering carry more weight; ones that never do stop firing. |
| Where the documents end up | Uploaded to a third party, often retained, sometimes reviewed, and out of your control the moment you hit send. | Parsed on your machine, processed in memory, never stored. Nothing is pooled across firms and nothing trains a model. |
| Who owns the method | The provider. It changes under you when the model changes. | You do. Every weight and assumption behind the score is published and editable, and changing one re-scores every deal you have. |
Chiron runs on the same frontier models you already have access to. The difference is what surrounds them: your firm's accumulated judgment, a fixed record shape, and arithmetic you control.
A chat thread is thrown away. A record, a question list and the answers that came back are assets. Two years in, the lens is the thing a competitor cannot copy.
For reading one document once, a chatbot is faster and free. Chiron earns its place when the same team evaluates deal after deal and needs the reads to be comparable, defensible and consistent between people.
This is architecture, not a policy page. Your counsel is welcome to verify it.
Files are processed in memory. Chiron extracts the structured data and discards the file, keeping only a hash so the same document is recognized across your team. The tradeoff, stated openly: there is no in-app source viewing, because there is no stored source.
No cross-firm aggregation, and no opt-in to one either. Your lens, your record, and your questions live in your workspace and nowhere else.
Access-controlled processing with no human review and no training on your data. Ephemeral, zero-retention processing at the model provider.
The full architecture is on the security page.
Chiron was the centaur who taught Achilles, Asclepius, and Jason. He was not famous as a warrior or a healer. He was the one whose judgment the great ones carried into every decision they made afterward, long after they left him. The Greeks set him apart from every other centaur by exactly that: the rest were impulsive, and Chiron was the one who knew things and passed them on. He taught different students different things, the way a lens is personalized by firm and by deal type. And his knowledge came from long practice, not a divine grant, which is the difference between accumulated judgment and a general-purpose model. That is the line this product stays on: judgment that compounds, attached to the person who earned it.
Powered by Argus Technologies Corp.
The fastest way to judge the engine is to run something where you already know the answer.
Sign in with your email