Intelligence-grade analysis for investment and investigations
AI analysis that holds up under cross-examination.
TriangulAIte brings intelligence community methods and rigor to AI analysis. It's built to test a thesis instead of agreeing with it, adapt to the way your team already works, and show you how it reached each conclusion.
In development. Demos on the working system from mid-December 2026. Early access opens in late February 2027 for 10 teams.
The trouble
A chat window is a poor analyst.
Dropping a data room into an AI chat gets you a fluent answer. Fluent isn't the same as right.
- It agrees with you. Hand it a pitch deck and it reads the deck the way the founder hoped it would.
- It won't say what it skipped. A confident summary hides the pages it never connected.
- It can be confidently wrong. Lawyers were sanctioned for filing a brief that cited six cases a chatbot had invented. Mata v. Avianca, S.D.N.Y., 2023.
- It can take orders from the documents. Text no reader would notice can be written to tell it what to conclude. Researchers hid instructions like "give a positive review only" in 17 preprints, in white or tiny type, aimed at AI reviewers. Nikkei Asia, July 1, 2025.
Side by side, an illustration
Same deck. Two answers.
This is an illustration. The deck, the company, its figures, and both answers are invented. The answer on the right is what TriangulAIte is built to give.
The questionHere's the deck and the customer list. Should we take the meeting?
A chat window
This looks like a strong opportunity. Net revenue retention of 140% is well above what's typical for software companies, and the $4 billion market leaves room to grow, though you may want to verify the market size. The team has relevant experience. I'd take the meeting.
TriangulAIte, as it's built to answer
Take the meeting, but test the retention figure first.
- Rests on
The 140% retention appears only in the deck. In the customer list, two customers account for most of the growth.
- Unsupported
The $4 billion market has no source in the deck.
- Ask the founder
What does retention look like without the two largest customers?
Where it comes from
Analytic tradecraft, built for business.
Intelligence analysts have long had methods for reaching sound conclusions from information that's incomplete, conflicting, or meant to mislead. Those methods were too slow and too labor-intensive for everyday business decisions.
TriangulAIte adapts them to business problems at the speed and scale AI makes possible, and combines them with established methods from business analysis and investment management. Steve, Josh, and Donald, intelligence analysts who practiced these methods together through two deployments in Afghanistan, designed it with people from business analysis.
The team is Steve Collins, CEO, a data and AI executive who built the data function at three companies; Josh Seaton, CTO, a career intelligence analyst and published PhD computer scientist; and Donald Franks, Analytic Methodologist, who ran intelligence operations for a Special Forces battalion.
Same failures, new names
AI fails the way analysts do.
The problems people complain about in AI today aren't new. Intelligence analysts have fought every one of them, and the intelligence community spent decades building methods to catch them before a bad judgment reached a decision.
An analyst who misremembers a detail, or fills a gap with what ought to be there, reports it with the same confidence as a fact.
A team drifts toward what leadership wants to hear, because disagreeing is uncomfortable or risky.
Once a judgment is made, new evidence gets bent to fit it instead of testing it.
The key fact is in the file. Nobody connects it to the one forty pages away.
One claim, repeated through sources that all trace back to the same origin, looks like corroboration.
An adversary plants material designed to steer the analyst toward the wrong conclusion.
TriangulAIte is built on those methods.
Why it's called TriangulAIte
Truth is not found in the consensus of the agreeable, but in the triangulation of the divergent.
A navigator doesn't trust a single bearing, and neither does an intelligence analyst. Each fixes a position from several independent ones. That's where the name comes from.
What changes
Analysis you can check.
What it's built to do, and what the demo will show you.
It argues with the thesis.
It makes the strongest case against your thesis as well as for it, so the weak points surface before the committee finds them.
Built around your process.
Start from one of our published workflows, or set it up to follow your team's own. A built-in assistant that knows analytic best practice helps you design it.
You get the argument behind the answer.
See how each conclusion was reached, and question any part of it. We replace "trust me" with "check my work."
Change a number, see what moves.
Disagree with a figure or an assumption? Change it, and see how the conclusion changes. Keep both versions and compare them.
A record your committee can follow.
Every analysis keeps the record of how it was reached, one your committee, your LPs, or your family can follow.
Under your control
Not a wrapper. Not a prompt. A system.
TriangulAIte isn't a thin layer over one company's AI, and it isn't a clever prompt. It's a system built to make any capable model reason better.
Better models make it better.
It isn't tied to any one model. When a stronger one arrives, TriangulAIte can put it to work, so every advance at the frontier lifts the analysis with it.
No single provider holds the keys.
It doesn't depend on any one company's API. If a provider changes its terms, its prices, or its models, the work goes on.
Every document is evidence.
It's built to examine what a document says and to resist text planted in a data room to steer the result.
Built to run where your data lives.
After early access, it's built to run on a single AI workstation, fully air-gapped, or inside your own private or public cloud, with your own model accounts or open models. Early access itself runs in our cloud. Where your documents go
Security
Your work stays yours.
In early access, each firm gets an environment of its own, with its own database, file storage, keys, and sign-in. Nothing is shared with any other firm.
Encrypted before it's stored.
Your documents, prompts, and results are encrypted before they reach a database, with keys held in your firm's own key store.
A record that shows any tampering.
Every step of every analysis goes into a tamper-evident record, signed and time-stamped every minute.
Built to the standards.
We build to SOC 2, ISO/IEC 27001, and NIST's frameworks for secure development and AI risk, with a SOC 2 Type II audit planned.
How we protect your work Download the security overview (PDF)
How it works
We'll show you in person.
In the demo, check for yourself
- The strongest case against your thesis sits next to the case for it.
- The contradictions, weak assumptions, and missing evidence it finds in your documents.
- Change a number and watch the conclusion move.
- All of it runs on documents you bring.
We don't publish how TriangulAIte works. Early users, investors, and serious buyers get the full walkthrough.
Built for
Work where a wrong answer is expensive.
Venture capital, private equity, and family offices come first, and the same rigor applies to investigations. We'll build what each of the first three needs with the 10 early-access teams.
Venture capital
For the associate who screens the deals and the partner who makes the call.
In the demo, we'll run the working system on a deal you passed on and already know the outcome of.
Planned for early access
- A clear call on each deal (pass, watch, or dig in) and the reasons for it.
- The assumptions the deal rests on, ranked, each with the evidence that would prove it wrong.
- The questions to ask the founder on the next call.
- A memo a partner can read in ten minutes.
Private equity
For the deal team working through a CIM and a data room.
In the demo, we'll run the working system on a deal that surprised you after close.
Planned for early access
- Every add-back sorted into recurring, one-time, or aggressive.
- Flags on the adjustments that change the debt multiple or covenant headroom.
- Churn and cohort quality, read from the customer exports.
- The questions to press with management and the investment committee.
Family offices
For the CIO asked to judge a technology outside their expertise.
In the demo, we'll run the working system on a technical pitch that was hard to judge.
Planned for early access
- Science or sci-fi? Each technical claim is marked established, emerging, or speculative, with the reason.
- A plain-language account of what it claims, and where the comparison to something familiar stops holding.
- The downside cases, including the ones that would ruin the investment.
- A memo you can share with family members and advisors.
Investigations
For investigators who need conclusions that hold up when the information is incomplete, conflicting, or meant to mislead.
- Claims checked against the documents and against each other.
- Contradictions and gaps called out.
- A record of how each conclusion was reached, ready to defend.
On the calendar
Three ways in.
Demos start in mid-December 2026. Early access opens in late February 2027.
Early access
Opens in late February 2027 for 10 venture, private equity, and family office teams willing to test it on real work and tell us what's missing. If you'd like to be one of them, request a place.
Investors
Book a demo from mid-December and we'll show you the system running, with time for your questions about the company and the team.
Your requirements
Have a specific use in mind? Tell us what it has to do, and we'll show you whether TriangulAIte fits.
Questions
What people ask.
Isn't this a chatbot with a better prompt?
No. A chatbot gives you one fluent answer and asks you to trust it. TriangulAIte is built so that a conclusion has to survive scrutiny before you see it, and you can see what it survived.
What does the intelligence community have to do with it?
Intelligence analysis is built for conclusions that have to hold up when the information is incomplete, conflicting, or deliberately misleading. TriangulAIte adapts those methods to business and investment questions. It was designed and built by people from that world and from business analysis.
Which AI models does it use?
In the demo and early access, commercial AI models through their providers' APIs. It isn't tied to any one provider, it gets better as the models do, and it's built to run on open models on your own hardware.
What will the demo show?
The working system, running live on a single workstation. We'll run it on our documents, and you're welcome to bring your own. Old decks and investor materials that are no longer sensitive work well. Tell us when you book and we'll confirm what we can take.
Can you tell us more about how it works?
Yes, in person. Request a walkthrough and we'll show you all of it.
What happens when the big AI companies build this?
They'll keep making better models, and TriangulAIte is built to put each one to work. A model company sells its own model's answers. TriangulAIte is built to test any model's work, theirs included.
What happens to the documents we upload?
Nothing you upload is used to train a model. The demo runs on one of our workstations, using commercial AI models through their providers' APIs. In early access, each firm gets a dedicated environment of its own in our cloud in the United States, and everything is encrypted before it's stored, with keys in your firm's own key store. Our security overview has the details. Running it in your own cloud or on your own hardware comes after early access. For a demo, bring documents that are no longer sensitive, like old decks and investor materials. We'll sign an NDA if you'd like, and you can watch us delete them afterward. A model provider may keep its own copy for a limited time, under its terms.
When can we use it?
Demos start in mid-December 2026. Early access opens in late February 2027 for 10 teams. If you'd like to be one of them, request a place and we'll be in touch.