Your AI workforce

Eleven AI agents.
One team that never sleeps.

You never assign work to one of them. You ask a question and the right specialists pick it up - sharing the same claims, the same evidence and the same standard of proof, so what one of them learns, all of them know.

10 of 11 working on your market todayForecaster is still being built - named below, with why
Collecting sources

Scout

Finds what is on the record

Thirty credible sources, read for every company you watch. The list of sources, and their licences, is on the Trust page.

Pricer

Watches pricing and packaging pages

Dated captures of a rival’s pricing page, compared over time. Every change opens the archived page it was read from.

Recruiter

Reads hiring as intent

Job postings, from the boards companies post them on. A requisition is a company stating in public what it will pay for.

Regulator

Tracks filings and rules

SEC filings and US Federal Register documents, each one opening the filing itself.

Reading and connecting

Sifter

Decides what matters

Every signal carries the score that let it through, and everything held back stays in a log you can read.

Historian

Reads the archive backwards

Ninety days read backwards, with near-duplicate coverage of one story merged into one story.

Mapper

Keeps the decision-makers current

People named only where a document names them, and the document is always linked. Nothing is bought, and nothing is guessed about anybody.

Verifier

Counts how many independent sources actually say it

Every signal carries a verdict: how many independent sources have it, which ones, and how many copies of the same text were discounted. Confirmed, on the record, one source reprinted, or single source - and the sentence explaining which.

what it does not doIt counts agreement, not truth. Two independent sources can both be wrong, and the verdict says "reported independently", never "verified as fact". It also cannot see a shared origin it has no evidence of: two outlets briefed privately by the same person, who then write their own sentences, read as two voices here.

Writing the intelligence

Analyst

Writes the brief

The brief, written from the shortlisted signals, with every line tracing to a dated source.

Delivering it

Router

Delivers it by email, and as a file you can use

The brief arrives by email on your schedule, and everything on every screen exports to Excel, CSV, PDF or slides.

what it does not doEmail and file export are the whole of it. There is no connection to Slack, Teams, a CRM or a webhook, so if the plan is for a brief to land in a channel or a record to appear in your CRM, that is not what this does. The upside is that no token of yours sits here.

Why there are no accuracy scores here

Because nothing has measured them yet. A percentage next to an agent’s name is only worth reading if something recorded what it predicted and what then happened, and we have not been running long enough to have that ledger. When it exists, the numbers will appear here and you will be able to see the calls they were computed from - including the bad ones. A figure we could type in today would tell you nothing except that we were willing to type it in.

Delivered by the team, not by the platform

The one that is a service, not a subscription

No model here predicts anything, so this one is done by people: scoped to the decision you have to make, quoted for that scope, with the assumptions written down.

Forecaster

paid service

Calls what happens next

Forecast is a specialized paid service rather than part of a plan: the scope, the price and the turnaround are agreed for what you actually need. Speak to the team and it is quoted against your subject, your competitors and your timeframe.

Nothing in this codebase predicts anything, and that is deliberate: a prediction is only worth reading next to a record of how often past predictions were right, and no such ledger exists. So the forecast is done by people, priced per requirement, and the assumptions come written down with it.

Speak to the team

How the team works

Shared claims, shared evidence, one standard of proof

A team is only worth having if the parts agree on what they are looking at. These do, because they read the same index rather than each fetching their own.

They read one index

What Scout collects, Sifter scores and Analyst cites. Nothing is fetched twice and nothing is remembered in two places, so two agents cannot quietly disagree about what a document said.

Only a subset runs

Asking for one lens dispatches the agents that lens needs and no others - 7 to 9 of them, not all eleven. Talent needs no Pricer; Market needs no Mapper. It is faster, it costs less, and every agent in the run has a reason to be there.

Every claim keeps its document

A signal that reaches you carries the dated page it came from, and the ones that did not clear the bar stay in a readable log. You can disagree with the reading and check the fact underneath it.

Every source class, and what we will never do →

Put them on your market

The 10 that are working go on your companies from tonight, on every plan.