Checked before it acts.
Every action is checked against your rules and your approval steps before it runs. Anything not explicitly allowed is denied, and the attempt is recorded with the reason.
DENIED UNLESS ALLOWEDYOUR APPROVERS
For operations leads at EU crypto and fintech companies
Put AI agents to work. Keep the accountability.
We deploy your first AI agent, live in 2–4 weeks. The agent never holds your credentials. Every action is checked against your rules before it runs. Every step is recorded with who approved it, and your auditor can verify the record without us.
Four questions, two minutes. A one-page plan for your first agent by email within two working days. No call unless you want one. Or watch one action go through the live demo
Live in production at an EU crypto company
Who answers for what it does? A 25 second answer.
01 / The problemWhy pilots stall
Agents are ready.
Accountability isn’t.
-
01
An agent needs the keys.
To do real work it needs access to payment, customer and core systems. You can’t hand those over.
-
02
You answer for every action.
Under DORA you own the risk of every system that acts for you, including an agent. Your auditor asks you who approved each action, not the vendor.
-
03
So pilots stall.
Or engineers spend months building controls in-house that an auditor still can’t check.
02 / What we doWhat happens to every action
We deploy your AI agents.
You keep control.
FeirAI deploys AI agents into the operations of regulated EU crypto and fintech companies, then runs the deployment. The software is FeirOS, our control layer: we install it between the agent and your systems, in your environment, and set it up for you. Every action goes through it: checked against your rules, run with credentials the agent never holds, recorded with who approved it. You keep your tools, your models and the rules you already have. The agent has no other way in.
ACTION / 7F-204
CHECKING
refund-assistant
Issue a €240 refund
acme-co · approved by ops lead
-
01
POLICYCHECKEDAGAINST YOUR RULES
-
02
CREDENTIALSWITHHELDNEVER SHOWN TO AGENT
-
03
RECORDSEALEDVERIFIABLE OFFLINE
-
04
BUDGETCHECKEDWITHIN LIMIT
STATUS
ALLOWED
RECORD 7F-204
It never holds the keys.
Passwords, API keys and wallet keys stay in FeirOS, in your environment. FeirOS uses them on the agent’s behalf once an action is approved. The agent never sees them.
CREDENTIALS OFF THE AGENTOPEN SOURCE
The part that holds the keys (Vultrino) is open source. View on GitHub
Every step on the record.
Each action is recorded with who approved it and signed, so any change to the record shows. Your auditor can verify it on their own machine, without us.
CHANGES SHOWOPEN SOURCE
The part that keeps the record (Averin) is open source. View on GitHub Try the verifier
It stops at the limit you set.
Each agent has a spend limit you set. At the limit it stops; the small overshoot is stated up front, not hidden. One switch stops the agent at any time. No key has to be changed, because it never had one.
SPEND LIMITONE-SWITCH STOP
03 / ProofOne deployment, in production
Already live in
regulated operations.
- In production
- Liveat an EU crypto company, running its client pipeline, reports and marketing. Not named publicly yet.
- What we commit to
- 2–4 wksfrom kickoff to the first live action, for the first workflow
- Who does the work
- FeirAIwrites the rules, ships the deployment and runs it, with no integrator in between
Your other options
Keep doing it by hand, or build it yourselves.
Both can work. Here’s the trade.
| Keep doing it by hand | Build it yourselves | With FeirAI | |
|---|---|---|---|
| First workflow live | It already runs, on your team’s time | After months of engineering | In 2–4 weeks |
| Who does the daily work | Your team, by hand | An AI agent your engineers built | AI we set up in your systems, run by your team or by us |
| Who approves | A second person signs off | Whatever approval step your engineers build | Your staff, on each step your rules say needs a person |
| Proof for the auditor | Your usual paper trail | Logs they take on trust | Records anyone can verify offline |
| When volume grows | Hire another analyst | More engineering | Add the next workflow |
| What it costs | A salary for every analyst you add | Three months of one engineer to build, then part of one to maintain | A fixed price for the first deployment, less than three months of one engineer, then a monthly fee |
You can also take Vultrino and Averin and build the rest yourself. What you buy from us is the deployment, the rules written into it, and someone who runs it.
04 / Who it’s forCrypto and fintech first
AI agents for regulated operations: crypto, payments, fintech.
AI agents for crypto operations AI agents for payment operations Banking, insurance and other regulated operations
05 / QuestionsBefore you fill in the form
Questions a COO asks first.
01What is FeirAI, and what is FeirOS?
FeirAI is the company and the service: we deploy AI agents into the operations of regulated EU crypto and fintech companies, then run and extend the deployment. FeirOS is our software, the control layer the agent runs through. It sits between the agent and your systems, in your environment. The parts that hold credentials (Vultrino) and records (Averin) are open source; the rule engine and operator console are ours.
02Do we have to change our agents or tools?
No. Agents built on LangChain, Hermes or your own code keep running as they are, with the model vendor you already use. If you don’t have agents yet, we build the first one with you.
03Which regulations does this help with?
DORA has applied since January 2025. It makes you own the ICT risk of every system that acts for you, including us. It covers crypto-asset service providers under MiCA and payment institutions. The EU AI Act’s transparency rule already applies: since 2 August 2026, people must be told when they are dealing with an AI system unless that is obvious. Its high-risk duties follow in December 2027 for the uses it covers. FeirOS gives you rules that are enforced and a record you can verify without us. It doesn’t replace your compliance team. Read what DORA requires you to show for an AI agent and what changed in the EU AI Act in 2026.
04Where does it run, and what if FeirAI goes away?
In your environment, on infrastructure you control. Credentials and records never leave it. Prompts and outputs go only to the model vendor you already use. We have no standing access: during the deployment we work through access you grant and can revoke, and what we do is on the same record. If FeirAI goes away, you keep running. The software is in your environment, the parts that hold keys and records are open source, the rules are yours, and every record verifies without us.
05Are you SOC 2 or ISO 27001 certified?
Not yet. Today your security team can read the two components that handle your secrets and your records, Vultrino and Averin, verify any record on their own machine (verify.feir.ai is the same tool hosted), and review the full rule set. We share the architecture, data flows and access model in a security pack, and we answer your vendor questionnaire and your DORA register-of-information request directly. Certification is on the roadmap.
06What does it cost?
The readiness check is free. The first deployment is a fixed-price project: one workflow, live in 2–4 weeks. We agree the price after the readiness check, and it stays below three months of one engineer building it in-house. After that, a monthly fee to run the deployment, keep the rules current and add the next workflow. We don’t publish list prices yet.
07Can we stop an agent immediately?
Yes. One switch stops the agent at any time. No key has to be changed, because the agent never had one: FeirOS holds the credentials and stops using them for the agent. Each agent also has a spend limit, with the small overshoot stated up front. Read how agent spend limits work and why the agent never holds the keys.
08Which actions need a human approval?
You decide, per action and per amount. Every action is checked against your rules before it runs, and anything not explicitly allowed is denied. A small refund can run on its own, a larger one can wait for one approver, and a payout or a limit change can need two. Each approval is recorded with the action. Read what an AI agent governance layer does.
09How long does a first deployment take?
Two to four weeks from kickoff to the first live action, for one workflow. The readiness check comes first: four questions, then a one-page plan by email within two working days. If it fits, we need a named owner, access to the one system the workflow touches, and your rules in whatever form they exist today.
Free readiness check
Get the plan for
your first agent.
Four questions, about two minutes. Within two working days you get a one-page plan by email: which workflow can go live first, what blocks it today, and what it takes. A call only if you want one. If it fits, we need from you a named owner, access to the one system the workflow touches, and your rules in whatever form they exist today.
Dominik Zachar, founder
Previously CTO of Cypher Enterprises and solution architect of the Summon Platform, Cardano’s first DAO platform: treasury and governance infrastructure for votes of up to about 100,000 concurrent participants. I read and answer every readiness check personally.
A young company is a risk you should price in. What happens if FeirAI goes away: the software is in your environment, the parts that hold keys and records are open source, and every record verifies without us.