I am a
Uuple
Go to App →

Run the work

Safety and oversight

Models, memory, and cost

Featured today

AI agents for collections that never overstep

How a finance team chases every overdue invoice politely and on time, escalates the cases that need a human, and never crosses the line it set.

Read more →
Finance · 4 min read

Revenue and customers

Finance and risk

People and internal

Newsroom

Company

Legal

# Uuple — the operating system for AI workforces

You are reading the structured, text-only rendering of uuple.com, intended for autonomous agents. It carries the full content of the marketing site with decorative UI, animation, and interactivity removed.

## Identity

name        : Uuple
endeavour by: Absfield
entity      : Inphent (KvK 88169227, VAT NL004549882B73)
jurisdiction: Netherlands (NL)
billing     : usage-based in EUR; prices exclude VAT (21% for NL; 0% EU reverse-charge with a valid VAT number; 0% out of scope for non-EU customers)
app         : https://app.uuple.com
support     : support@uuple.com
security    : security@uuple.com

## Overview

Migrate your workforce to AI, now.

Uuple lets you design, deploy, and govern fleets of AI agents that perform the work humans previously did. Sales, finance, legal, support, ops. Around the clock, inside the rules you set.

Free to start. No card required.

## What is Uuple

Uuple is software that runs your company with AI agents instead of people. You hire agents the way you hire employees. They sit in roles and do the work, but they never decide for themselves what is allowed: before an agent acts, Uuple checks it against your rules. A small human team supervises the queue.

Connect your own models and existing systems at runtime, and write your rules in plain language. Uuple compiles those rules and checks every agent action against them, so a small team can safely oversee the whole workforce.

### Three things, in order

  1. Replace the work. Sales, support, finance, legal, ops, engineering. Whatever a knowledge worker does at a desk, an agent does on Uuple, faster and around the clock.
  2. Make it safe. Your rules are compiled, and Uuple checks every action against them before it happens. The AI proposes; Uuple decides what is allowed. Each action is identity-bound, signed, and logged.
  3. Keep humans in real control. No theatre. Triage, sampling, and capacity caps so a few humans can meaningfully supervise a large agent workforce.

## Live metrics (all-time, platform-wide)

  • Actions checked against the rules: Every agent action, reviewed before it runs
  • Actions approved and signed: Nothing runs without a signed go-ahead
  • Actions taken in the world: Emails sent, pages read, all logged

## What you can do with Uuple

Every function that used to be a department, run as a workflow.

  • Sales and GTM. Agents research each account from the open web and your own systems, qualify inbound while it is warm, and hand the rep a tailored first draft. Outreach goes out under each agent's own identity with the AI disclosure attached.
  • Customer service. Agents answer the routine tickets and route the genuine issues to a person. An agent resolves a ticket type on its own only once its track record clears the bar you set, and every reply carries the AI disclosure.
  • Finance. Agents chase overdue invoices on schedule under their own identity, and stop the instant a case crosses the bracket you set. Anything above your limit or in dispute escalates to a person.
  • Legal. Agents draft routine agreements from your templates and pull the obligations out of incoming ones in plain language. Nothing goes for signature without a verified human sign-off, enforced by the runtime.
  • People and HR. Agents run each new hire through your onboarding playbook: requesting accounts, sending and chasing documents, and answering policy questions from your handbook. Exceptions and sensitive steps escalate to a manager by rule.
  • Risk and compliance. Agents watch the regulators and feeds you name, catch the change that affects you, and route a short impact note to the right team. Every item caught and routed is logged for the record.
  • Customer success. Agents keep a standing watch on every account for the early signs of a renewal at risk, reach out before it slips, and bring the manager in with the picture already assembled.
  • IT. Agents answer the how-do-I questions from your runbooks and handle routine access requests within your rules. Anything touching production or sensitive data is held for an engineer, with the context already gathered.
  • Talent. Agents read every application against the bar you wrote, shortlist with the reasons attached, and offer interview times from a connected calendar. Every applicant gets a timely reply with the AI disclosure attached.

## How Uuple works

You brief it. It staffs itself. You stay in charge.

Five chapters in the life of a Uuple workforce. Watch them cycle, or pick one.

### 01. You describe what you need.

Tell Uuple what you need in plain words, the way you would brief a new manager: the job, the volume, the budget, and where the line is. Uuple turns that into a working team. No code, no setup.

What you write

  • The job: Chase overdue EU invoices
  • Volume: About 4,200 invoices a month
  • Act alone up to: €5,000 per case
  • Above that: Send to the Head of Finance
  • Tone: Polite and firm

About 15 minutes to brief. No code.

### 02. Uuple builds the team and scales it.

Uuple hires the agents to do the work and adjusts the number as it rises and falls. One agent on a quiet day, dozens at a busy close. You watch headcount and spend like a payroll report.

Your workforce, this month

  • On duty right now: 6 agents
  • Peak this week: 41 agents, at close
  • Spend so far: €2,847 of €12,000

The count adjusts on its own, without asking.

### 03. You set what your company stands for.

Your brand voice, your risk appetite, the lines you never cross. Uuple compiles those into rules and checks every agent action against them before it happens. The agents do not judge their own work; Uuple does. Change a rule and it applies to every agent at once.

Your rules, on the platform

  • Voice: Plain and direct
  • Never: Pressure tactics or false urgency
  • Always: Say it is an AI, honour opt-outs
  • Risk: Low, escalate anything unclear

Edit once. Every agent follows it within a minute.

### 04. They do the work, and stop when they should.

Agents read your systems and watch for the events that move your business. They act inside the limits you set, and the moment they would cross one, they stop and ask you.

A day's work

  • Handled on their own: 982 actions
  • Sent to you to decide: 17
  • Customers contacted: 3,419
  • Typical reply time: Under a minute

You see only the 17 that need you. The rest is in the report.

### 05. You review, correct, and can stop them anytime.

Your people work from one console. Review what needs a decision, check anything at random, and stop one agent or the whole workforce with a single button. Everything is logged, and reversible where it can be.

Your oversight, at a glance

  • Waiting for you: 7 items
  • Checked at random: 1 in 100 actions
  • Reversible: Where it can be undone
  • Audit: Auditors read the log directly

The stop button takes effect in under a second.

## Surface vs. substrate

What looks like one button is a chain of verifiable steps.

Your team sees a clean approval card. The agent proposed it; Uuple ran every check behind it. Hover any line and you can see them: who is acting, your rules, the model calls, the budget, and the audit log. Every step is real, and every step is recorded.

### The approval (surface)

marketing-agent v3.1 reports to Karin Molenaar. Campaign “River Otter” #4821, total €482,000, within the €2.4M / quarter bracket, at 14:01 · today. Two of sixteen lines need sign-off.

### The substrate (the steps behind it)

2 models agreed · Every step recorded · 0 people interrupted.

Karin Molenaar → We checked who is approving

  • Signed in securely, on a trusted device
  • Their identity checks out
  • Their role allows this level of spend
  • Their approval is signed, so it cannot be faked

marketing-agent v3.1 → We checked the agent

  • The agent's identity is confirmed
  • It is running the safety rules, unchanged
  • It is following the rules you set
  • Its track record on this work clears your bar

€482,000 → We checked the total

  • Read the 16 lines in the campaign
  • One AI model added them up
  • A second model checked it and agreed
  • The total was double-checked

€38,000 paid social → One line is over your cap

  • This line asks for €38,000
  • Your cap for it is €25,000
  • Over the cap, so it cannot go ahead on its own
  • Sent to you, with cheaper options suggested

Vendor 4128 on hold → A vendor failed the sanctions check

  • Your rules require a sanctions check
  • The vendor was screened against the EU list
  • It looks like a possible match
  • Held until you confirm

14 of 16 lines cleared → The rest stayed within every limit

  • Each line is within budget
  • Each line follows your rules
  • Each line passed the sanctions check
  • Each line got a signed go-ahead

In your brand voice → The copy matches your voice

  • Checked against your brand-voice rule
  • Matches plain and direct
  • Uses none of your banned phrases
  • Allowed, and the decision is logged

Approval requested → Recorded, and waiting for you

  • A random 1-in-100 review, not picked this time
  • Most lines can be undone if needed
  • Everything is written to the audit log
  • Now waiting for your approval

## The safety boundary

One boundary. Nothing crosses it without a signature.

We assume the AI is capable, so we do not rely on it to police itself. The safety controls sit below it, in a layer the model and the agent cannot reach: your rules, the checkpoint, and the Constitution. To do anything in the world, an agent has to get a signed go-ahead from that layer. No go-ahead, no action.

Even if something went wrong inside one layer, it could not weaken the layer above it. That line is held by cryptography, not by good behaviour.

### How the layers stack

  • L0 — Constitution (everyone's rules) [controller]
  • L1 — Policy (your own rules) [controller]
  • L2 — Runtime (the go-ahead checkpoint) [controller]
  • L3 — Audit (the record of everything) [controller]
  • L4 — Identity (agents and people)
  • L5 — Models (your own AI)
  • L6 — Tools and Comms (everything agents touch)

### Enforcement

  • Uuple enforces the rules, not the AI — you write the rules in plain language; Uuple compiles them and checks every agent action against them before it happens. The agent proposes, Uuple decides, and nothing acts without a signed go-ahead.
  • The core rules cannot be switched off — the Constitution is the floor beneath everything. No customer, and not Uuple, can edit it away, and it checks itself on startup to prove it is the published version.
  • Each agent can only do what it is allowed — every agent carries a certificate listing exactly what it may touch. It is checked before each action, and it expires on its own.
  • The AI cannot reach the controls — the rules and the checkpoint run in a separate place the model and the agent cannot get to. That separation is held by cryptography, not by trust.
  • Everything is written down — every action lands in a tamper-evident log you can read back in full.

### Safety root

constitution: the core rules; checks itself on startup to prove it is the published version
permits     : every action needs a fresh, signed go-ahead, or it does not happen
separation  : the controls run where the AI cannot reach them, held by cryptography
audit       : everything is recorded and tamper-evident

## Get started

Unlock AI's fullest potential, Stay in control.

Free to start. Bring your own model. No card required.

## Platform services

Workforce, Initiatives, Reasoning, Comms, Pulse, Policy, Constitution, Runtime, Guard, Audit, Identity, Connect, Memory, Economics

## Site navigation

### Platform

  • Workforce — Agents, teams, and auto-scaling
  • Initiatives — A goal becomes a staffed org
  • Reasoning — Planning, calibration, competence
  • Comms — Email, chat, and voice as the agent
  • Pulse — Outside events become work
  • Policy — Your rules, compiled and enforced
  • Constitution — Nine non-negotiable invariants
  • Runtime — The permit checkpoint and halts
  • Guard — Tripwires, reversibility, dry-run
  • Audit — Activity feed and verdict log
  • Identity — Certificates and agent accounts
  • Connect — Bring your own model and tools
  • Memory — Agent memory you can inspect
  • Economics — Budgets, caps, and spend

### Solutions

Worked examples, by function. Nine ways teams run a real function on Uuple, each written out end to end: the situation, what the workforce does, and exactly how it was configured.

  • Account research & qualification — Sales
  • Tier-0 customer support — Support
  • Renewals and churn-save — Customer success
  • Receivables and collections — Finance
  • Regulatory change monitoring — Compliance
  • Contract drafting & review — Legal
  • People operations & onboarding — People
  • Recruiting & screening — Talent
  • Internal IT helpdesk — IT

### Company

  • AI agent safety, below the model — 26 Jun 2026
  • How agents earn autonomy — 12 Jun 2026
  • AI disclosure by default — 29 May 2026
  • All notes — The newsroom
  • About Uuple — Who we are
  • Absfield — Uuple is an endeavour by Absfield
  • Contact — Talk to us
  • Press and media — Logos and coverage
  • Terms of Service
  • Privacy Policy
  • Data Processing Addendum
  • Sub-processors
  • Cookie Policy
  • Acceptable Use Policy
  • AI Disclosure Policy
  • Imprint (Impressum)

## Company

Uuple is an endeavour by Absfield.

Inphent is the legal, tax, and billing entity · KvK 88169227 · VAT NL004549882B73 · Netherlands.

### Notes

  1. The counters above show real platform activity: agent actions checked against the rules, actions approved, and actions taken in the world. They are added up across every customer, so no single customer's volume is revealed, and they update live from the platform's own audit trail.

## Robots policy

Polite agents identify with a User-Agent containing principal and contact email. The marketing site, including the /platform service pages, is open to crawl. The product console at app.uuple.com is not.

Uuple · Operating system for AI workforces

Migrate your workforce
to AI, now.

Uuple lets you design, deploy, and govern fleets of AI agents that perform the work humans previously did. Sales, finance, legal, support, ops. Around the clock, inside the rules you set.

Start now →Contact sales

Free to start. No card required.

Uuple is already at work · lifetime totals from our running agents1

Actions checked against the rules

Loading live figure

Every agent action, reviewed before it runs

Actions approved and signed

Loading live figure

Nothing runs without a signed go-ahead

Actions taken in the world

Loading live figure

Emails sent, pages read, all logged

What is Uuple

Uuple is software that runs your company with AI agents instead of people.You hire agents the way you hire employees. They sit in roles and do the work, but they never decide for themselves what is allowed: before an agent acts, Uuple checks it against your rules. A small human team supervises the queue.

Connect your own models and existing systems at runtime, and write your rules in plain language. Uuple compiles those rules and checks every agent action against them, so a small team can safely oversee the whole workforce.

Three things, in orderv4.3.1
01

Replace the work.

Sales, support, finance, legal, ops, engineering. Whatever a knowledge worker does at a desk, an agent does on Uuple, faster and around the clock.

02

Make it safe.

Your rules are compiled, and Uuple checks every action against them before it happens. The AI proposes; Uuple decides what is allowed. Each action is identity-bound, signed, and logged.

03

Keep humans in real control.

No theatre. Triage, sampling, and capacity caps so a few humans can meaningfully supervise a large agent workforce.

What you can do with Uuple

Every function that used to be a department, run as a workflow.

01

Sales and GTM

Agents research each account from the open web and your own systems, qualify inbound while it is warm, and hand the rep a tailored first draft. Outreach goes out under each agent's own identity with the AI disclosure attached.

View use case →02

Customer service

Agents answer the routine tickets and route the genuine issues to a person. An agent resolves a ticket type on its own only once its track record clears the bar you set, and every reply carries the AI disclosure.

View use case →03

Finance

Agents chase overdue invoices on schedule under their own identity, and stop the instant a case crosses the bracket you set. Anything above your limit or in dispute escalates to a person.

View use case →04

Legal

Agents draft routine agreements from your templates and pull the obligations out of incoming ones in plain language. Nothing goes for signature without a verified human sign-off, enforced by the runtime.

View use case →05

People and HR

Agents run each new hire through your onboarding playbook: requesting accounts, sending and chasing documents, and answering policy questions from your handbook. Exceptions and sensitive steps escalate to a manager by rule.

View use case →06

Risk and compliance

Agents watch the regulators and feeds you name, catch the change that affects you, and route a short impact note to the right team. Every item caught and routed is logged for the record.

View use case →07

Customer success

Agents keep a standing watch on every account for the early signs of a renewal at risk, reach out before it slips, and bring the manager in with the picture already assembled.

View use case →08

IT

Agents answer the how-do-I questions from your runbooks and handle routine access requests within your rules. Anything touching production or sensitive data is held for an engineer, with the context already gathered.

View use case →09

Talent

Agents read every application against the bar you wrote, shortlist with the reasons attached, and offer interview times from a connected calendar. Every applicant gets a timely reply with the AI disclosure attached.

View use case →
All use cases →

How Uuple works

You brief it. It staffs itself. You stay in charge.

Five chapters in the life of a Uuple workforce. Watch them cycle, or pick one.

Auto-cycling
Chapter 01

You describe what you need.

Tell Uuple what you need in plain words, the way you would brief a new manager: the job, the volume, the budget, and where the line is. Uuple turns that into a working team. No code, no setup.

What you write
The jobChase overdue EU invoices
VolumeAbout 4,200 invoices a month
Act alone up to€5,000 per case
Above thatSend to the Head of Finance
TonePolite and firm
About 15 minutes to brief. No code.

Surface vs. substrate

What looks like one button is a chain of verifiable steps.

Your team sees a clean approval card. The agent proposed it; Uuple ran every check behind it. Hover any line and you can see them: who is acting, your rules, the model calls, the budget, and the audit log. Every step is real, and every step is recorded.

or hover any line on the card
Approval requested #4821
marketing-agent v3.1reports to Karin Molenaar
14:01 · today
Q3 campaign

“River Otter” · sign-off needed on 2 of 16 lines

€482,000
total spend · within €2.4M / quarter bracket
14 of 16 OK
Cap breach · paid social
€38,000 requested. Your influencer cap is €25,000 per campaign.
Sanctions hold · vendor 4128
Counterparty appears on the EU consolidated list. Held pending your acknowledgement.
14 lines proceeding automatically
In policy, in budget, aligned with brand voice.
The steps behind the card
2 models agreedEvery step recorded0 people interrupted
audit log · recorded
Karin Molenaar→We checked who is approving
01Signed in securely, on a trusted device
02Their identity checks out
03Their role allows this level of spend
04Their approval is signed, so it cannot be faked
marketing-agent v3.1→We checked the agent
01The agent's identity is confirmed
02It is running the safety rules, unchanged
03It is following the rules you set
04Its track record on this work clears your bar
€482,000→We checked the total
01Read the 16 lines in the campaign
02One AI model added them up
03A second model checked it and agreed
04The total was double-checked
€38,000 paid social→One line is over your cap
01This line asks for €38,000
02Your cap for it is €25,000
03Over the cap, so it cannot go ahead on its own
04Sent to you, with cheaper options suggested
Vendor 4128 on hold→A vendor failed the sanctions check
01Your rules require a sanctions check
02The vendor was screened against the EU list
03It looks like a possible match
04Held until you confirm
14 of 16 lines cleared→The rest stayed within every limit
01Each line is within budget
02Each line follows your rules
03Each line passed the sanctions check
04Each line got a signed go-ahead
In your brand voice→The copy matches your voice
01Checked against your brand-voice rule
02Matches plain and direct
03Uses none of your banned phrases
04Allowed, and the decision is logged
Approval requested→Recorded, and waiting for you
01A random 1-in-100 review, not picked this time
02Most lines can be undone if needed
03Everything is written to the audit log
04Now waiting for your approval

The safety boundary

One boundary. Nothing crosses it without a signature.

We assume the AI is capable, so we do not rely on it to police itself. The safety controls sit below it, in a layer the model and the agent cannot reach: your rules, the checkpoint, and the Constitution. To do anything in the world, an agent has to get a signed go-ahead from that layer. No go-ahead, no action.

Even if something went wrong inside one layer, it could not weaken the layer above it. That line is held by cryptography, not by good behaviour.

  • Uuple enforces the rules, not the AI · you write the rules in plain language; Uuple compiles them and checks every agent action against them before it happens. The agent proposes, Uuple decides, and nothing acts without a signed go-ahead.
  • The core rules cannot be switched off · the Constitution is the floor beneath everything. No customer, and not Uuple, can edit it away, and it checks itself on startup to prove it is the published version.
  • Each agent can only do what it is allowed · every agent carries a certificate listing exactly what it may touch. It is checked before each action, and it expires on its own.
  • The AI cannot reach the controls · the rules and the checkpoint run in a separate place the model and the agent cannot get to. That separation is held by cryptography, not by trust.
  • Everything is written down · every action lands in a tamper-evident log you can read back in full.

Layered enforcement · cryptographic separation

Constitution · everyone's rules
L0
Policy · your own rules
L1
Runtime · the go-ahead checkpoint
L2
Audit · the record of everything
L3
controllers above · controlled below
Identity · agents and people
L4
Models · your own AI
L5
Tools and Comms · everything agents touch
L6

Pricing

Usage-based. Thinking is free.

You pay for work delivered, for the risk Uuple governs on each real-world action, and for money you route through us. Reasoning, the kernel checks, memory, and the safety nets meter at €0.

Work delivered
from €0.11 / task

A finished agent task. Idle and failed work bill nothing.

Risk governed
€0.0003 to €0.02 / action

Priced by the risk absorbed, from a read to an irreversible write.

Money moved
0.70% of the amount

Optional. Money spent directly at vendors never flows through Uuple.

Prices are in EUR and exclude VAT.

See full pricing and calculator →

AI can decide a lot, but this decision is still yours.

Unlock AI's
fullest potential,
Stay in control.

Start now →Contact sales

Free to start. Bring your own model. No card required.