Skip to main content
The caller · Agents

Agents can reason.
Not the person.

Models generate language. None of them gets the financial state of the person, or whether this moment fits spend, save, invest, or silence.

01The gap

Reasoning is common.
Understanding is not.

More banks will ship agents. More fintechs will ship copilots. Reasoning gets cheaper. What stays scarce is a durable view of the actual human on the other side — and gating the turn before the model speaks.

An agent that re-reads the last thirty transactions will sound fluent and still miss the person. It assumes an average user. Almost no one is average.

Neumetria is a layer an agent can read before it plans: one financial state for the person, plus readiness gates, each versioned and confidence-weighted. Your policies still bound the action.

And when you run more than one agent — a copilot, a lending agent, a wealth assistant — they read the same state rather than each inferring a different person from the same transactions.

03Gates

Then ask: does this
action fit now?

Understanding alone is not enough for a trustworthy turn. The agent also needs readiness — whether this moment fits spend, save, invest, or silence — before your product allows a prompt.

SpendSpend readiness

Capacity for a spend-facing suggestion. Your agent policy decides whether that becomes a message, a tool call, or nothing.

SaveSave readiness

Whether a buffer or savings suggestion fits current load — or would add pressure the person cannot take.

InvestInvest readiness

Whether a funding or allocation turn matches holding posture and room, not a stale risk label.

QuietStay quiet

When drift or strain says wait. “Not now” is a valid agent outcome when your policy binds the turn.

»» Gates are readiness for your runtime to read. They are not advice to the end user. Your product and policies own the action.

04How agents use it

Query. Gate.
Your product decides.

01
Ground
The agent reads one financial state for this person — summary plus the objects it came from — instead of inventing context from a short window of raw events.
02
Gate
Readiness for spend, save, invest, or silence tells your runtime whether a turn is in bounds before the model plans language.
03
Bound
Your product applies policy: what the agent may suggest, escalate, or withhold. Neumetria does not instruct consumers.
04
Audit
Objects carry confidence and as-of. You can show why the agent was allowed to speak — or why it stayed quiet.
  • ReadAgent runtime fetches state and gates before planning a turn.
  • BoundYour product applies policy on top of readiness — allow, tone, escalate, or withhold.
  • LogOptional outcome feedback sharpens the view over time. You still own the duty of care.
Agent turn · grounded
// illustrative grounding bundle
{
  "state": { "posture": "stable_improving", "confidence": 0.94 },
  "basis": {
    "income_rhythm": "biweekly",
    "behavior_drift": "cooling_19d"
  },
  "gates": {
    "spend_readiness": "moderate",
    "save_readiness": "hold",
    "stay_quiet": false
  },
  "policy": { "allow_prompt": true, "tone": "quiet" }
}
05The line we hold

Understanding for agents.
Your product decides.

Neumetria is not the agent. It is the state and gate layer agents read before they act — it reports what is, and does not instruct consumers or approve turns.

Your agent runtime and product policies own what happens next.

That line matters for trust, for oversight, and for every person on the other side of the chat.

Agents

Neumetria gives AI the financial understanding your products decide with.