Engineered, not prompted
Evaluation suites, shadow mode and an audit trail on every agent we ship.
Autonomous AI agents for operations-heavy businesses
Paid from measured results. If the numbers don't move, you owe nothing.
Why us
Making it safe to run inside a real operation is the work.
Evaluation suites, shadow mode and an audit trail on every agent we ship.
A baseline agreed in writing before we build. It's what we're paid on.
Working is the floor. We keep tuning for more gain, because that's our fee.
Paid from measured results, or not at all
One team end to end, not five vendors
In production in months, not years
Project ideas in your numbers, in 48 hours, free
Shadow mode before anything decides anything
Documented, monitored, and yours to keep
Where agents pay
Documents read, extracted and posted straight through. The uncertain ones stop at a person.
Forecasting and anomaly detection, backtested on seasons you have already lived through.
Agents that draft the quote and answer the routine questions — citing the document they came from.
Pricing, demand and lead-scoring models trained on what your business already knows.
The platform
Agent Stack
Trust Layer
PDFs, scans and spreadsheets go in; structured records come out. Every extracted field carries a confidence score — above the threshold it posts automatically, below it it queues for a person.
The agent reaches your operation through MCP connectors — phones, email, spreadsheets, charts, your databases and calendars. Every connection is scoped, read-only by default, and logged call by call.
An orchestrator plans the work and delegates each step to a specialist agent. Handoffs are explicit, so every step can be tested, swapped and audited on its own.
Answers are generated from retrieved records, never from memory. Each claim links to the exact passage it came from — no source, no answer.
Every change runs against a fixed test set before it can ship. Pass rates are tracked per category, and anything below the agreed gate is held back.
The agent runs beside your team on live work and decides nothing. It goes live only after its accuracy holds above the current process for a full cycle.
Every decision is logged with its inputs, confidence and outcome. Throughput, escalations and drift are monitored live, and any decision can be replayed end to end.
The current cost of the process is measured and fixed in writing before we build. Our fee is a share of the verified gain against that number — nothing else.
How you pay
A working session on your numbers. No fee, no commitment.
Built and run in shadow mode. No invoices while we prove it.
A share of the measured gain. No gain, no fee.
Baselines are agreed in writing and verified in your numbers, not ours.
A specialized practice
Your bank's margin is inside the rate it quotes you, and it never shows up as a fee. We spent five years pricing FX and derivatives at Bank of America, Safra and XP — we know how banks price you. We benchmark every transaction against timestamped central-bank rates and hand you the evidence. A flat monthly fee, and we are never paid on your flow.
About
We're a senior team of engineers and operators who build and run AI agents inside established businesses. Every recommendation arrives as ROI math, not a technology pitch.
Founded by Alex Moussalli — economics at Insper, derivatives at Bank of America, Safra and XP. We build and operate what we sell, for operations-heavy businesses where manual hours cost money. Built to outlive the engagement, and to hand over clean.
Tell us what you sell and what your team spends its days on. In 48 hours you'll have project ideas in writing, free.