A radar that watches, an engine that acts and a co-thinker that makes sense of it all

The radar used to go through the watch list in order, with the same weight for every account.

Now each company gets a score and the radar looks first at the ones worth the most. No AI picks the order. It comes from a calculation anyone can review, using the company's size, its industry, its press presence and its history in the radar.

When a company adds or drops a tool, the change gets fixed points by category, capped at 100.

A math rule calculates it, not an AI. The same change always gives the same score, and that score picks which kind of message fits.

The points for each category are changed on a screen in the panel, and every change keeps its reason and its history.

The day's queue is ordered by that score, and recent empty searches subtract from it.

The radar works from highest to lowest and stops at the first company that leaves a publishable signal, or when it reaches the daily cap.

If a company left a valid signal, it leaves the queue for 30 days. That way the companies that are always in the news don't drown out the rest.

If the scan found nothing, the company comes back to the queue the next day, a little lower each time it comes back empty.

When the day's round finds nothing, the radar widens the search without relaxing the verbatim quote requirement.

First it looks at the last 3 days outside the watch list. If that isn't enough, it rewrites the search and extends it to 7 days. It stops at the first step that finds something.

A useful message needs three things at once: the right account, the person who decides on that topic and a reason that explains why now.

The reason comes from a signal with its source, not from an assumption about the company. Each contact gets a score from 0 to 100 for their role, their account, the company data, their email, the signal and any prior relationship, and the campaign is built by score band.

Getting contact data costs money. So the engine first decides whether the person is worth it and only then pays for their data.

The decision looks at what the person can actually get done in the company, not at the title on their profile.

When a model compares a text with a fact, it can only give one of three answers: it matches, it doesn't match or it's insufficient.

If the text is ambiguous or the model fails, the answer is insufficient and the system doesn't fill the gap. On the public radar, when the readings disagree, a person breaks the tie.

Before a campaign is loaded, its summary shows the checks that must be at zero: accounts with no angle, contacts with no opening reason, no counterpart by role or not yet reviewed, and excluded accounts. Paused contacts are shown and get nothing.

When sending, each block stops only the contact or account affected. A contact who already got a message is paused for 90 days.

When someone replies, their sequence stops on every channel, and if they unsubscribe they never get anything again.

Everything sent on behalf of a client, whether an email, a message or an automatic alert, starts as a draft.

Every email can be seen exactly as it will go out. The opening and the bridge are edited per contact, and the rest in each step's template. It stays held until a person approves it. Without that approval, the engine doesn't send it.

Per-item sign-off is mandatory while the engine is in training. When measured confidence is high, a flow can move to audited silence: it goes out without per-item sign-off and a person audits it every week, blind.

Autonomy is lost on its own. If the audit is overdue, a serious error shows up or a test goes undetected, the flow goes back to per-item sign-off without asking. Today every flow goes out with per-item sign-off.

Once a signal is signed off, it can trigger several actions at once.

The email goes out with a record of when it arrived, and the team gets an alert in the channel it uses. If the company adopted a competitor, it can leave your paid audiences: that removal is built and switches on once you connect each platform.

On professional networks the system never sends on its own: it prepares the task and a person does it.

Every message goes out tagged by angle, structure and email type. Variant b differs from a in just one thing, so you know which change moved the result.

In each cycle, once it is switched on for the client, the system looks at what happened after sending and improves the next version. What was learned stays in the Unified Signal Layer, not in someone's memory.

Each client has its own library of sources: texts, pages, videos and its own documents. The Co-Thinker talks a problem through with those sources in view and remembers what was discussed from one session to the next.

It doesn't live on a separate screen. From a signal, a news item, an account, a campaign, a message or a contact, it opens alongside, with that as context and without leaving what you were looking at.

The model that reads can't browse, search or research. It only reads the text it receives.

It doesn't give opinions either: it doesn't judge tone, morals or business value. It answers a single question, whether the source text supports the fact exactly as written.

Every public radar signal is always shown the same way: the company, the fact, where it came from, the verbatim quote with its source, three readings and a footprint that lets you check nothing changed.

The three readings are independent: one from an open-weights model, one from a commercial model and one from a person. Each says whether the quote supports the fact, and the person decides where the other two disagree.

Three kinds of footprint arrive by different paths: what is said about the market, what a company does and what changes in people.

They all converge in one place before reaching your CRM. That way there are no isolated copies that contradict each other.

Every paid query is logged with the module and function that requested it. That way you know what each part of the system costs.

There are three brakes: a daily cap, an overall ceiling that cuts off on its own and monthly Co-Thinker credits for each guest. A warning arrives before the search limit.