Benefy

Built for San Francisco residents

Find every dollar you qualify for.

Tell us about your household in plain English. Benefy's intake agent builds your profile, a deterministic rules engine screens it against real SF benefit programs, and you get a dollar estimate plus a pre-filled application — in minutes, not hours.

See how it works
Benefy — new screening
I'm a single mom with two kids, work part-time at a bakery, and live in the Mission. Worried about my PG&E bill.
Got it — household of 3, part-time income. Checking CalFresh, PG&E CARE, and SFMTA Lifeline against 2026 FPL thresholds…
Intake agent building structured profile

Estimated annual value

$4,860

CalFreshLikely eligible
PG&E CARELikely eligible
SFMTA LifelineNeeds review

Programs

Tailored to your household

CalFreshCalFresh

Monthly grocery assistance based on household size and income.

PG&EPG&E CARE

A discounted rate on your monthly gas and electric bill.

SFMTA MuniSFMTA Lifeline

Reduced-fare Muni passes for qualifying San Francisco residents.

How it works

From a sentence to a screened case

1

Tell us about yourself

Type what's true about your household in plain English — no forms, no dropdowns, no waiting room.

2

Agent builds your profile

The Gradient AI intake agent extracts your household size, income, and situation into a structured profile.

3

Engine screens, not the AI

A tested, deterministic rules engine checks your profile against real SF/CA program thresholds — the model never decides eligibility.

4

Dollars and next steps

Get an annual dollar estimate, anything flagged for human review, and a pre-filled application ready to submit.

Trust & guardrails

The AI never decides eligibility.

A pure, unit-tested rules engine is the only code path that ever returns likely-eligible, likely-ineligible, or needs-review. The agent gathers information and explains results — the yes/no always comes from the function call, never the model.

Deterministic engine

Same input always yields the same output, checked against real 2026 program thresholds.

Human-in-the-loop review

Uncertain cases are flagged for review, never auto-approved.

Full reasoning trace

Every screening keeps a "view reasoning" trace so decisions stay auditable.

Guarantee-language guardrails

Agent output is checked for over-promising phrasing before you ever see it.

3

SF/CA benefit programs screened

12

Deterministic engine test cases

0

Eligibility calls made by the AI

Ready to find every
dollar you're owed?

No account, no auth, no personal data required to try the full demo.