Phainly Website

Phainly comes from the Greek phainō, to bring to light.

Every machine decision, brought to light.

Models already decide who gets credit. Next, AI agents will pay on your customers' behalf. Phainly explains each automated decision, proves who authorised it and keeps it within limits, so you can automate and still answer for every outcome.

EU Consumer Credit Directive applies in 15 d 10 h 04 m

Every bank now has to show its work.

For decades a person decided, and a person could explain. Now models decide in milliseconds and agents are learning to pay. Supervisors in Brussels, London and Abu Dhabi are asking banks the same question.

Credit is decided by models.

From 20 November 2026, EU lenders must explain automated credit decisions and offer a human review on request. UK and UAE rules point the same way.

Customers will send agents.

Visa and Mastercard opened agentic payments to dozens of European banks in 2026. Banks still can't see who sent an agent, or what it is allowed to do.

Machines will move money.

Gartner expects AI agents to handle more than $15 trillion of business purchasing by 2028. Many spend controls for AI only send an alert.

Three products, timed to the market.

LUCID solves the problem lenders face this month. SIGIL and KEEL are built for the next wave, and we are designing them with partner banks now.

Pilots openLU 2026 000001

LUCID

Explained credit decisions and human review

Every “no”, explained. Every review, real.

LUCID sits beside any scoring model. It turns each automated credit decision into a clear explanation, a path to yes, a one-tap human review and a sealed record. Those are the things CCD2 asks of lenders from 20 November 2026.

  • Keeps your scoring model and explains whatever it decides
  • Customer screens ship as a software kit for your app
  • Evidence packs ready for complaints, audits and supervisors
Design partnersSG 2027 000002

SIGIL

Mandates and proof for AI agents

Know which agent. Know who sent it. Know what it may do.

Your customers are about to send AI agents to their accounts. SIGIL lets them give each agent a signed mandate with a scope, limits and an end date. It checks every agent request against that mandate and records the result in a tamper-evident black box.

  • Signed, revocable mandates set by the customer
  • Allow, ask or deny on every request, with the reason
  • Evidence for disputes, complaints and supervisors
Early accessKL 2028 000003

KEEL

Treasury and brakes for agents' money

Agents can spend. KEEL decides how much.

KEEL gives every agent an allowance worked out from the business's real cash forecast. It checks each spend before it happens and stops a runaway agent, a team or the whole company in one move. Every agent payment lands in a ledger auditors can trust.

  • Limits that move with the cash forecast
  • A kill switch with three levels
  • A ledger of every agent payment and its reason

One ledger underneath.

All three products write to the same decision ledger. Each entry records what was decided, by whom or by what, on which data and under which rule or mandate. Entries are chained by hash and time-stamped, so nobody can quietly rewrite history.

Read it as

Was each decision explained, and did a person review it when asked?

Example decision ledger with synthetic entries
Time Who decided What On what basis Explanation Human review Within limits Seal
09:14:03 Scorecard v4.2Credit model Decline: pay in 6, €680 R-17: commitments 52% of incomeR-09: returned direct debits ShownVersion E-14, English RequestedBy the customer at 09:16 — 7f3c…a1
09:41:12 Trip plannerAgent acting for A. Novak Pay €186: rail ticket Mandate M-0193Travel, €900 per booking RecordedWithin mandate Not needed Yes€606 of €2,000 this month 9c0d…4e
10:15:30 Research agentAgent for Acme Ltd API spend €18.40 Budget B-12Research tools RecordedNormal rate Not needed Yes€212 of €600 today e04a…7b
10:17:02 KEEL gateAutomatic control Pause Research agent 31 calls a minuteNormal rate is 1.2 RecordedAnomaly rule K-04 Owner askedApprove or keep paused StoppedSpending 26× normal 55d2…c9
10:20:47 Unsigned botNo verified identity Balance request No mandateNo signature RecordedBlocked, customer told Not needed — c3e9…12
11:31:18 M. KovalenkoCredit reviewer Approve €400 on review Case R-2611-0417Refund confirmed in transactions ShownVersion E-15, English Done2 h 15 m after the request — 1b77…0c

Explains any model

Works beside your scoring and decision engine. There is nothing to replace.

Answers in plain language

Teams ask questions in words. Every answer shows the SQL that produced it.

Keeps personal data minimal

Stores references and hashes where it can, rather than copies of customer data.

Runs where your data lives

Deploy in your cloud region or on your own infrastructure.

The rules arrive on a schedule.

The dates that matter to lenders and payment firms in the EU, the UK and the UAE.

  1. UAE

    Central Bank guidance on AI and machine learning

    Banks are expected to explain AI-driven decisions and keep human oversight of credit decisions.

    Published
  2. UK

    Data (Use and Access) Act safeguards

    New rules for significant automated decisions, including human intervention and the right to contest.

    In force
  3. EU

    Distance contracts for financial services

    Online tools must come with adequate explanations, and a person on request.

    In force
  4. EU

    Consumer Credit Directive (CCD2) applies

    Borrowers gain rights to an explanation, human intervention, to give their view and to a review of automated credit decisions.

    In 16 days
  5. EU

    Anti-Money Laundering Regulation applies

    Automated decisions to accept or refuse customers need meaningful human intervention.

    In about 8 months
  6. EU

    AI Act duties for credit scoring

    High-risk obligations apply to creditworthiness assessment: human oversight, logging and explanations.

    In about 13 months
  7. EU

    Payment Services Regulation

    Expected to apply, with new rules on fraud and on how payments are authorised.

    Expected 2028

What we hold ourselves to.

Show the evidence.

Every sentence we generate points to the data and the rule behind it.

Leave the decision with you.

We explain, route and record decisions. Your people and your models make them.

Make human review real.

We measure review quality, including test cases that show whether reviewers truly check.

Collect less.

Minimal personal data, clear retention and deployment in your region.

CCD2 applies on 20 November 2026.

Time left

15days
10hours
04minutes
00seconds

Book a readiness diagnostic. We map your automated credit decisions against the new consumer rights and show you what is missing. For SIGIL and KEEL, ask about the design-partner programme.

hello@phainly.com

Tell us what you are working on