Runtime as featured inForbesRead the article

The AI agent harness for payment teams

Build and run AI agents across payments, risk, compliance, and support, in your cloud with human approvals.

Book a demoRequest access

Trusted by teams running mission-critical operations.

RainNixtlaMonoLineLeap
Search
risk-ops
MessagesCanvasFiles
Today
MR

Maya 9:12 AM

re-review Harbor Vape Supply, dispute ratio spiked overnight

APP 9:13 AM

Dispute ratio is 1.4%, above the 0.9% VAMP threshold. KYB still matches, but the website changed owners on Sep 22 and 61% of disputes are "not received".

Recommend a 25% rolling reserve. 14 queries, 3 sources.

Approval needed

Apply 25% reserve to Harbor Vape Supply

👀 2🙌 1

DK2 repliesLast reply today

DK

Dev 9:15 AM

Can you check their two sibling MIDs too?

Message #risk-ops

Aa

Expand your ops team with digital teammates

Call an agent from Slack, Teams, email, text, or a phone call, and it asks before it acts.

How Runtime works

1

Describe the job

Describe the work in plain English, attach the SOP, and Runtime builds the agent.

New agent

Create a banking ops agent that handles RFIs, hold-harmless requests, and returns of funds from our partner bank, verifies against our ledger, and drafts replies for my approval.

Describe the agent you want
Agent ready

bank-ops-agent

Tools picked from your prompt

EmailLedgerSFTPMastercard
2

Connect your tools, even legacy ones

APIs, databases, MCP servers, and a browser for bank portals with no API.

Tools and access · bank-ops-agent
Bank emailRead requests
LedgerRead-only
Returns of fundsNeeds approval
3

Run it

Call it from Slack, Teams, email, or phone, or trigger it from an alert or schedule.

Triggers · bank-ops-agent
Slack@Bank Ops can you take the Column RFI?
Teams@Bank Ops in #exceptions
Email[email protected]
SMS+1 415 555 0142
VoiceAnswers the merchant support line
AlertsSplunk alert · daily 7:00 schedule
4

Keep a human in the loop

Money movement, reserves, and replies to your bank wait for approval.

Approvals
Money movementNeeds approval

Return $12,400 to the originating bank for the flagged transfer, with a drafted reply to Column and the evidence file attached.

ApproveView run
Run stored with every step, query, and cost
Evidence packet filed for compliance

One harness for every team that touches a transaction

Start with one team. The same context and controls carry to the next.

Getting customers live

Underwrite, onboard, and launch customers in days, not weeks.

Keeping money moving

Trace every exception before it turns into a backlog.

Protecting the business

Work every alert, with the evidence on record.

Supporting customers

Answer “where's my money” with real payment data.

See how Runtime powers one of the fastest-growing fintechs

Rain

Running at the speed of a $1.95B stablecoin company

$750k

Vendor and salary spend saved

Rain runs Runtime for

Built for mission-critical payment operations

When agents work your money movement, they can’t go down and they can’t go rogue.

AuditedSOC 2Compliant

Your cloud, your models

Run in your own cloud or self-host, and we never train on your data.

Built to stay up

Isolated sandboxes and failover keep investigations running through outages.

PCI and PII stay private

Card numbers, SSNs, and secrets are masked before any prompt or log.

Access and approvals

SSO, role-based access, and a person approving money movement.

Measure every agent’s risk

Every run scored, with evidence ready for your auditors and partner bank.

See it on your agents
Agent riskLive

  • payout-investigator

    42Medium

  • dispute-responder

    18Low

  • kyb-reviewer

    71High
Release $12,400 held for approval

A system of intelligence for payment teams

Every investigation becomes shared memory, so the next case is faster for every team.

Every query, tool call, approval, and cost, ready when an auditor or your partner bank asks.

Run #1047

Re-review Harbor Vape Supply against VAMP thresholds

Tool calls

14

Duration

4m 02s

Cost

$0.31

Pulled 30 days of disputes and KYB records
Found an ownership change on Sep 22
Reserve recommended and approved by Maya

Agents learn from resolved cases and propose updates to their own skills.

merchant-re-review · skill

Learned from the last 40 resolved cases:

Run #983Ownership change preceded dispute spike
Run #1012Reserve approved at 25% for vape MCC
Run #1029Checked sibling MIDs before closing

Suggested skill update

Add step 4: check sibling MIDs before closing a re-review.

Every finding lands in one governed memory, so no team investigates a case twice.

memory · one paymentlive
Support
Payment ops
Finance
Compliance
MID 4410 · memory
✓Account changed Sep 18
✓Release approved
✓Sibling MIDs clean

Every run gets faster and cheaper

Agents start on frontier models to get the work right, then Runtime drives cost per run down.

123$0.42 / run$0.03 / run10,000 runs / weekWeek 1Week 24Frontier modelOpen modelDeterministic scriptCost per run

10×

more runs

−93%

cost per run

−29%

total weekly spend

1

Route to the smallest model that passes

−86% model cost on routine enrichment

2

Sharper skills, fewer steps

−40% steps per case

3

Proven runs become reviewed scripts

$0 model spend on the routine path

Figures are illustrative.

Forward-deployed AI engineers for your AI strategy

An engineer who has built payment and risk systems at companies like these builds your first agents with your team.

Plan your AI rollout
Modern Treasury
Finix
Marqeta
Adyen
BlackRock
Meta
Hulu
Disney
Yahoo
Y Combinator
Berkeley
University of Michigan

Frequently asked questions

Who is Runtime for?
Payment companies on both sides of the transaction: issuers and program managers, acquirers, processors, gateways, ISOs, and payfacs. Teams across the operation use it, from underwriting, onboarding, and implementation to payment ops, disputes, risk, compliance, and support.
Which workflows do payment teams start with?
Usually the queue that is growing fastest: transaction-monitoring alerts, ACH returns and notices of change, proof-of-authorization requests, chargeback evidence, merchant re-reviews, partner-bank RFIs, or "where is my payout" escalations. Once the first agent works, the next team builds on the same harness.
Can an agent move money on its own?
Only if you let it. Agents start read-only, and you choose the steps that need a person, like releasing a held payout, applying a reserve, booking a ledger entry, or replying to your sponsor bank. The agent pauses and asks in Slack, Teams, or the dashboard, and resumes once someone approves.
What will our sponsor bank or examiner see?
Every run is recorded end to end: the trigger, every query and tool call, what each returned, the approvals, the cost, and the result. When your sponsor bank, an auditor, or an examiner asks how a case was handled, you export the run.
Does card data or PII ever reach the model provider?
It does not have to. Agent computers can run inside your own cloud account, so PCI data stays in your environment. Agents use masked credentials and never see raw secrets, hooks and filters strip card numbers and SSNs from prompts and logs, and you can serve open-weight models from your own cloud for the most sensitive work.
Can we run Runtime in our own AWS, GCP, or Azure account?
Yes. Most payment teams run agent computers in their own cloud account, set up with Terraform or CloudFormation, while we manage the control plane. Teams that cannot send anything outside their network can self-host the whole platform with a Helm chart.
Will it work with our processor, ledger, and bank portals?
Yes. Agents reach tools through APIs, databases, CLIs, and MCP servers, and use a browser for processor, network, and bank portals that have no API. Common connections include your processor and ledger, Salesforce, Zendesk, Jira, Snowflake, Postgres, Splunk, and Datadog.
We already built an internal agent. Why not build the rest ourselves?
The first agent is the easy part. Running agents on payment data takes isolated computers, scoped credentials, role-based access, approvals, an audit trail, evals, model routing, and fallbacks when a provider goes down. An engineering team can spend two quarters on that before the first agent touches real data. Runtime gives you all of it on day one, and works with the agents you already have.
What does a pilot look like?
We start with one SOP and agree on the success measure up front, such as cases per analyst or time to resolution. A forward-deployed AI engineer builds the first agent with your team, sets up guardrails with security, and trains your admins so your team can build the next ones. Many teams start on non-PCI work, then expand.

Do more with the team you have

See agents work your payment, risk, and compliance queues in your own cloud.

Book a demoRequest access