Service — AI Agents
An AI agent isn't a chatbot. It's a piece of software that takes action — it reads your emails, fills in your forms, updates your systems and finishes tasks without you. Fly AI designs, builds and deploys AI agents for Belgian SMEs and public bodies. Live in a week. Starting from €2.500.
From €2.500
Live in a week, with human-in-the-loop launch and full training
Your team spends hours each week copy-pasting data between systems that were supposed to talk to each other. Your best people burn their creativity on email triage, invoice matching and the 14th version of the same quote. The boring work piles up, nobody enjoys it, and nobody seems able to stop it.
Meanwhile, large corporations are already deploying AI agents to handle exactly this kind of work — freeing their teams to think, create and serve clients better. The gap between "us" and "them" is widening by the month.
AI agents close that gap. And unlike enterprise AI, modern agents are cheap enough, fast enough and safe enough to be practical for a Belgian SME of 15 people or a municipality of 60 employees.
A chatbot answers questions. It's a conversation. You ask, it replies.
A classic automation (Zapier, Make, Power Automate) follows a rigid if-this-then-that recipe. If the recipe breaks, the automation breaks.
An AI agent does both — and more. It understands messy, unstructured input (emails, PDFs, voice), decides what to do, takes the action, checks the result, and corrects itself when needed. It's closer to a junior colleague than to a macro.
A real AI agent has four moving parts:
A large language model (Claude, GPT, Llama, Mistral) that does the reasoning.
The actions the agent is allowed to take (read an inbox, query a database, post to a CRM, send an email, call an API).
Short-term (the current task) and long-term (everything it has learned about your business, documents and preferences).
The rules that keep the agent inside the lines (what it can do, what it must ask before doing, what it must never do).
They are not magic. They won't replace your whole back office overnight. They won't think strategically for your business. And they are not a good fit for tasks that are truly creative, deeply emotional or legally high-stakes. We'll tell you honestly when an agent is the wrong tool for a given job.
We listen to your pain. By the end of the call, we'll have a shortlist of the processes where an agent would make the biggest impact — or we'll tell you you don't need one.
We sit with your team and walk through the targeted process(es). We identify the hand-offs, the data sources, the edge cases and the guardrails. You leave with a one-page blueprint of your future agent, a realistic time-to-live, and a fixed price.
We build the agent in your environment, connected to your real systems (on a sandbox first). You see it learn, fail, and get better in real time. We use Claude (hosted in the EU), custom tool integrations and retrieval over your own data.
Every new agent goes live in "supervised mode" first: it drafts, your team approves. As confidence grows, we dial down the supervision until the agent runs on its own for the parts that are safe to automate. Nothing goes to autopilot without your green light.
Agents drift. APIs change. Business rules change. Your Care plan keeps the agent healthy, fixes regressions, and adds new capabilities as your team learns what they really want.
Reads incoming requests, pulls the right info from your past quotes and price list, drafts a tailored offer in your tone of voice, parks it in the salesperson's draft folder. Saves 5+ hours per salesperson per week.
Watches the inbox, extracts invoice data, matches to POs, queues for one-click approval. Connects to Odoo, Exact, Yuki, Octopus and others. Saves 6+ hours of admin per week.
Handles tier-1 questions on your website, WhatsApp, or shared inbox in FR, NL, EN and DE. Uses your real documentation. Escalates only the hard ones to a human. Up to 70% of tickets automated.
Reads incoming CVs, scores them against the job description, writes a 5-line summary for the hiring manager, ranks them top to bottom. 80% time saved on screening.
Indexes your Drive, SharePoint, Notion and Slack. Answers your team's "where is that document" questions in plain language, with links to the source so they can verify.
The one that finally makes your CRM talk to your accounting tool. Or your HR tool to your project tool. Replaces the manual copy-paste with an agent that can handle the messy cases too.
Belgian SMEs (5–250 people) who are drowning in repetitive admin work and want the boring parts of their business handled automatically — without giving up control.
Public bodies (municipalities, agencies, public services) who need to do more with the same headcount, and who care deeply about GDPR, the EU AI Act and data sovereignty. We can host entirely in the EU, on-premise if needed.
Professional services (law firms, accountants, consultancies) who want to free their billable experts from non-billable work.
Customer-facing businesses (e-commerce, hospitality, SaaS) who want multilingual, always-on first-line support without outsourcing their brand.
| Package | What you get | Duration | Investment |
|---|---|---|---|
| Discovery call | 30-minute honest conversation. Go/no-go. | 30 min | Free |
| Agent audit & workshop | Process mapping, agent blueprint, realistic scoping. Counts as EU AI Act literacy. | 2–4 h | From €450 |
| Starter Agent | One focused agent. Fixed price. Live in a week. Includes human-in-the-loop launch. | 1–5 days | From €2.500 |
| Custom Agent | Multi-step agents, multi-system integrations, advanced logic. Scoped per project. | 2+ weeks | On request |
| Care & evolve | Monitoring, fixes, quarterly improvements, updates for API/model changes. | Monthly | From €250/mo |
We host in the EU, sign a DPA, minimise data collection, and document lawful basis. We can work inside your existing data processing framework.
We classify every agent we build according to the AI Act's risk categories (minimal, limited, high, prohibited) and we document it. For higher-risk use cases, we add human oversight, traceability and bias monitoring.
For sensitive cases — especially public sector — we deploy inside your environment, on EU infrastructure (AWS Frankfurt, Scaleway, OVH) or on-premise. Your data never leaves your boundaries.
We aren't tied to a single provider. We pick the right model per use case: Claude, GPT, Mistral, Llama, open-weights. EU-hosted where possible.
Every new agent starts in supervised mode. Nothing goes to autopilot without your explicit approval.
Every agent action is logged, reviewable and reversible.
We build and we care. Lots of firms will design you an agent. Fewer will still be there in 6 months when the OpenAI pricing changes or the Gmail API breaks. We build, we deploy, and we keep it running.
We're honest about what agents can and can't do. We'll tell you when a task is a great fit, when it's a so-so fit, and when you should stay away. That honesty is why our clients come back.
We're part of The Fly Group. When your agent project touches custom development, Agifly helps. Odoo integration? Flydoo. SEO? Flyeo. One group, four brands, every angle covered.
We speak Belgian. FR, NL, EN and DE. Tax incentives. GDPR. EU AI Act. Public procurement. You don't need to translate anything for us.
We ship fast. Most starter agents go live within a week of the workshop. We refuse to let AI projects drag on for months.
A chatbot answers questions. An AI agent takes actions. An agent can read your emails, fill in forms, update your CRM, send messages and complete multi-step tasks — the things a chatbot only talks about.
Our Starter Agent is a fixed price from €2.500. A scoping workshop starts at €450. Custom agents with complex integrations are quoted per project after the workshop.
Most Starter Agents are live within 5 working days of the workshop. More complex agents take 2–6 weeks depending on the integrations and edge cases.
Yes, when designed correctly — which is exactly what we do. We host in the EU, sign a DPA, minimise data, and document lawful basis. For sensitive cases we deploy on-premise or inside your EU cloud.
Yes. We classify every agent against the AI Act's risk tiers and we document the classification. For higher-risk use cases we add human oversight, bias monitoring and traceability.
It depends on the use case. Claude is our default for reasoning-heavy tasks. We also use GPT, Mistral, Llama and open-weight models — always preferring EU-hosted options where possible.
Yes — we regularly integrate with all the common Belgian ERPs, CRMs and accounting tools. If your tool has an API, we can talk to it.
Every new agent starts in supervised mode: it proposes, a human approves. As confidence grows we dial down the supervision. Every action is logged and reversible, so mistakes are easy to catch and roll back.
Yes. For sensitive public-sector or healthcare cases we deploy inside your environment, with open-weight models, so your data never leaves your walls.
That's a valid concern and we take it seriously. In every project we involve the team whose work is being automated. AI agents work best when they take over the parts people hate — the copy-paste, the triage, the data entry — and let the humans focus on the parts they actually enjoy.
Both. The build is one-off (Starter or Custom). The optional Care plan is monthly and covers monitoring, fixes and small improvements. Cancel any time.
That's exactly what the workshop is for. We look at your processes together and identify the 2–3 that combine the highest pain with the highest probability of success. We start small, prove the value, and expand.
Book a free 30-minute discovery call. We'll listen, we'll ask the right questions, and we'll tell you — honestly — whether an AI agent could help you or not.