jataíEnxame. Qualifies with evidence. Replies with the person’s own words.
Enxame (agents) is Jataí’s set of agents. They do three things: qualify each organization Pólen (discovery) finds, citing the evidence used; write each touch of the Rotas (sequences) for the specific person; and triage the reply — classifying it and proposing a response that only goes out within the operator’s approval limit. It isn’t an "AI SDR": nothing goes out without a trail.
Agents that show the why behind every decision: the evidence that qualified, the angle of the first contact, the reason for the reply. You approve according to your approval limit — and it goes up as your trust does.
An "AI SDR" decides for you. Enxame shows what it saw.
The market sells autonomous agents that "handle everything" — at US$ 3K a month or US$ 36K a year — and that touch your pipeline without saying why. When they get it wrong, the team stops trusting the CRM. That doesn’t come back.
Enxame works by one rule: every decision carries its evidence. Qualified? It shows the evidence for each signal, and a score without cited evidence is rejected. Wrote something? It shows the angle. Proposed a reply? It shows the reason. And no agent decides the next step of a Rota: that’s a deterministic rule, not a model.
Three agents, three jobs.
The one that qualifies.
It reads the Pólen record and the brand profile, and answers: in or out, and why. Output: the fit (0–100), the tier (A to D), the signals evaluated — each with the cited evidence and the source — and a justification. The angle — the specific reason the offer matters to this organization — is calculated once per organization, and stored.
Instituto Horizonte de Formação · fit 84 · tier A · IN • "new class opening in February" — notice, institutional site, 11/09 • "3 locations (Centro, Setor Sul, Jardim Novo)" — Locations page • "academic coordination: Helena Prado" — Team page Angle: new class in 5 months → teaching material decision by November
The one that writes.
It writes touch 1 and the following touches of the Rota for that person, in the brand’s tone, starting from the angle. Each mold has a word limit and a list of prohibitions: no "hope you’re doing well", no manufactured urgency, no number that isn’t in the context. And no agent text goes out until the facts it cites are checked against the context.
The one that triages.
When a reply comes in, it classifies it into one of nine classes: positive · information request · not now · not interested · unsubscribe · out of office · referral · auto-reply · hostile. It gives a temperature from 0 to 100 and states the reason. For the first two, it proposes the reply and the lead goes to the CRM. An unsubscribe suppresses the contact immediately, without going through a human. And when it isn’t sure, it escalates to a person.
The reply, with the reason next to it.
Helena Prado · Instituto Horizonte de Formação · today 09:12 "Hi, thanks for reaching out. I forwarded this to our academic committee, they meet on the 22nd. Can you send me pricing?" REPLY PROPOSED BY ENXAME Of course, Helena. I’ll send the price table with the three class tiers today — plus a one-page summary prepared for the committee on the 22nd. anchored in: "they meet on the 22nd" · class: information request · Entrega: lead to CRM
The reply is a proposal, never a send: approval comes from the operator’s approval limit.
You choose how much you trust. Per operator.
Trust has a name in Jataí: approval limit. It lives in the person’s link to the brand, not in their role. A new operator starts with a zero approval limit — everything they approve becomes a draft for the manager. As trust grows, the approval limit grows: how many sends per day go out without review, and up to what lead value. A new brand starts in shadow mode: Enxame writes, and nothing goes out.
And by default every email Enxame writes waits for human review: it lands in the Review queue, where subject and text can be edited before release. The brand can switch to review by exception — then only the message that cites a fact outside the context, the low-confidence one and a minimum daily sample go to the queue. See Deliverability.
Each brand, its own tone and its own knowledge base.
Each brand has its own tone and its own knowledge base, and one brand’s Enxame never reads another’s. A school writes like a school; an agency, like an agency. See Multi-brand.
Three things Enxame doesn’t do.
It doesn’t negotiate.
Asked for a proposal, a discount, a contract? It goes to your salesperson via Entrega (handoff). Enxame only answers what’s already in the brand’s material.
It doesn’t send what the approval limit doesn’t cover.
A reply is a proposal. The operator approves it, within the approval limit they have — and whatever exceeds it waits for the manager.
It doesn’t learn from your data for other brands.
What your brand teaches stays in your brand.
Which models, where they run.
Enxame uses language models from providers (list in Security and LGPD) to write and classify, and deterministic rules to decide stops and suppression. Each task has a timeout and a deterministic fallback path: if the model doesn’t respond, the journey continues without getting stuck. Every call to the model is recorded, isolated by brand.
Frequently asked questions
Is Enxame an AI SDR?
Not in the market’s sense. It doesn’t decide on its own what goes into the pipeline and doesn’t touch the stage. It qualifies, writes, and triages — with evidence — and hands off to your salesperson.
What is an "anchored reply"?
A proposed response that shows, alongside it, the passage of the person’s message it relies on and the reason for the classification. It’s a proposal: it only goes out with approval within the operator’s approval limit.
Can I see why a lead was discarded?
Yes. Every discard keeps the fit, the evidence for the signals, and the justification.
What is an approval limit?
It’s how much a person approves on their own: how many sends per day without review and up to what lead value. It lives in their link to the brand. A new operator starts with a zero approval limit, and everything they approve becomes a draft for the manager.
How much approval time does this require from me?
It depends on the approval limit you grant. With a zero approval limit, every approval goes through the manager; as trust grows, review shrinks. A brand in shadow mode sends nothing until you take it out.