---
title: "AI agents hosted in the EU · Build, schedule and approve"
description: "Build AI agents on open-weight models that call your tools, run on a schedule or a webhook, and wait for a person to approve risky actions. Hosted in the EU."
canonical: "https://inference.eu/agents/"
language: "en"
---

AI agent platform in the EU

# Build AI agents for your team that run in the EU and ask before they act

An agent is a model with tools and a goal. Give it access to your apps, tell it what to do, and it works through the steps and hands back a result. Anything that sends, pays or deletes waits for a person to approve it first.

[Contact sales](https://inference.eu/contact/)[How it works](#features)

## How it works

-   ### Tools and sub-agents

    Give an agent the tools it needs, from web search to your own APIs, or let it hand parts of a job to smaller agents.

-   ### Approval before risky actions

    Mark actions as needing approval. The agent pauses, asks the right person, and carries on once they say yes.

-   ### Schedules and webhooks

    Run an agent every weekday at 7:30, or start it when an email arrives, a form is submitted or a ticket changes.

-   ### A record of every run

    Each run keeps its steps, tool calls, outputs and cost, so you can see what the agent did and why.

-   ### Permissions per agent

    Choose which connections, data and skills each agent can reach, and nothing else.

-   ### Start agents from your product

    Trigger any agent through the API and get the result back as a response or a callback.

-   ### A model for each agent

    Run each agent on the open-weight model that suits the job, and switch later without rebuilding it.

-   ### Skills and connections built in

    Agents use the connections and skills in your workspace, so they reach your apps and follow your team's way of working.

## What teams build with it

-   ### Support triage

    Read new tickets, look up the customer, draft a reply and route anything unusual to a person.

-   ### Document processing

    Pull fields out of invoices, contracts and forms, check them against your rules and file them.

-   ### Morning briefings

    Collect yesterday's numbers, news and open tasks into one summary before the team starts work.

-   ### Finance follow-up

    Match payments to invoices, chase what's overdue and prepare entries for your accountant, with anything outgoing held for approval.

## Agents: common questions

The questions buyers, developers and legal teams ask us most. Something missing? Ask us directly.

An AI agent is a language model given a goal and a set of tools. It decides which step to take next, calls a tool, reads the result and repeats until the job is done. Unlike a chatbot, it acts in your apps, which is why approval matters.

Any model on the platform, all of them open-weight and processed in the EU. You can choose a different model for each agent, for example a small, fast one for triage and a larger one for drafting, and change it later without rebuilding the agent.

Yes. Agents run on infrastructure in the EU. Their steps, model calls and run records are processed in European data centres, and the credentials for your connections are stored encrypted in the EU. You receive the subprocessor list with the data processing agreement under GDPR Article 28.

Write the agent's job as plain-language instructions, then choose its model, tools and connections. Decide which actions need approval and when the agent should start, on a schedule, from a webhook or on demand. Developers can also start agents through the API and handle the results in their own code.

You decide per action. Most teams require approval for anything that sends a message, makes a payment, books something or deletes data. Reading, searching and drafting run freely, and only the final step waits for a person.

When an agent reaches an action marked for approval, it pauses and asks the person you named, showing what it wants to do and why. They approve, edit or reject it. The agent then carries on, or stops, and the decision is kept in the run record.

Yes. Every agent can be started through the API or a webhook, with input from your product, and returns its result as a response or a callback. That lets you add agent features to your own software without running the infrastructure yourself.

A schedule starts an agent at fixed times, such as every weekday at 7:30 or on the first of the month. A webhook starts it when something happens elsewhere, like a new email, a submitted form or a changed ticket, and passes that event to the agent as input.

Yes. An agent can hand part of a job to a smaller sub-agent with its own instructions and tools, for example one that only searches documents. This keeps each agent focused and lets you set narrower permissions for the parts that touch sensitive data.

Each run keeps its steps, tool calls, outputs, approvals and cost. When an agent gets something wrong, you can see exactly where it went off track, adjust the instructions or the tools, and run it again with confidence.

No. What agents send to a model and what they read from your apps is used to complete the run and nothing else. It is never used to train any model, and that commitment is part of the data processing agreement you sign with us.

By usage, mainly the model tokens each run consumes, on the same invoice in euros as the rest of the platform. Every run record shows its cost, so you always know what a given agent costs per day, per week or per month.

[Ask a question](https://inference.eu/contact/)

## Move your AI workloads to EU infrastructure

Tell us which models you use and roughly how much traffic you have. We reply within one working day with pricing and next steps.

-   Reply within one working day
-   Pricing for your volume
-   DPA ready to sign