AI agent development company in Madrid
Custom AI agents for Madrid companies, connected to your tools, kept under human control and built with the EU AI Act and GDPR in mind.
- Custom quote after a brief
- No sales pitch
Company-wide:★★★★★ 4.9+ on Clutch·250+ brands·10 offices
UnFoldMart is an AI agent development company with an office at P.º de la Castellana, 259C, Planta 19, 28046 Madrid. We build custom AI agents, knowledge assistants and multi-step workflows for Madrid companies, with the EU AI Act and GDPR in mind, with human approval, logging and testing built in; agents can run in your own cloud account. Meetings are by video, in English. Pricing is a custom quote after a short brief and a free 30-minute call.
Last updated:
Key facts
| Service | Custom AI agents: task agents, knowledge assistants, integrations (CRM, helpdesk, databases, APIs, MCP), multi-step workflows, testing and monitoring |
|---|---|
| Company | UnFoldMart |
| Office | P.º de la Castellana, 259C, Planta 19, 28046 Madrid, Spain |
| Meetings | By video (in-person meetings only at our Frankfurt and Zürich offices) |
| Languages | First call in English; project calls by video in English |
| Time zone | Central European Time (CET/CEST); projects are led from our international offices |
| Control | Human approval before actions, every step logged, test sets from your own cases; agents can run in your own cloud account |
| Typical timeline | Use-case and data review 1–2 weeks · prototype 2–4 weeks · pilot 3–6 weeks · production rollout 2–6 weeks (ranges, not guarantees) |
| Pricing | Custom quote after a short brief: prototype, fixed-scope pilot or ongoing agent team. Model and hosting costs are billed by your providers. No price list. |
| Booking | Free 30-minute video call ↗ |
| Company-wide | 10 offices in 10 markets · 250+ brands · 500+ projects · 4.9+ on Clutch · 8 working languages |
AI agents for Madrid businesses
Madrid is Spain's capital and largest economy, the base of many Spanish banks, telecoms and retailers, and a gateway to Spanish-speaking markets in Latin America. A Spanish-language agent can serve customers well beyond Spain.
Madrid at a glance
- IndustriesBanking and insurance, retail, telecoms, travel, real estate, SaaS
- Data lawGDPR and Spain's LOPDGDD, supervised by the AEPD
- AI rulesEU AI Act, applying in phases since 2025
- LanguagesSpanish and English
- Time zoneCET in winter, CEST in summer
Banking and insurance
Agents that prepare checks and case files for a person to approve, fully logged.
Retail and consumer brands
Support triage and drafted replies in Spanish and English.
Travel and hospitality
Booking and service assistants connected to your systems, with a person in the loop.
What our Madrid AI developers build
AI agent development and AI automation that work inside the tools your team already uses.
Task agents
Triage, reports, leads, documents.
Knowledge assistants
Answers from your documents, source shown.
Integrations
CRM, helpdesk, databases, APIs, MCP.
Multi-step workflows
Plan, hand off, pause for approval.
Testing
Test sets from your real cases.
Guardrails and monitoring
Permission limits, checks, alerts.
Automation, chatbot or AI agent?
Not every task needs an agent. Use-case discovery shows where an agent helps, and where simpler automation wins.
| Rule-based automation | AI agent | Chatbot | |
|---|---|---|---|
| What it does | Runs fixed steps you define | Plans steps and uses your tools to finish a task | Answers questions in a chat |
| Messy input (emails, documents) | No: needs structured data | Yes | Partly: answers, doesn't act |
| Takes actions | Yes, exactly as scripted | Yes, within the permissions you set, after approval | Usually not |
| Best for | Stable, repetitive rules | Multi-step work across tools | FAQs and simple support |
How an agent earns trust
Each step unlocks only when the one before has proved itself.
Read-only
Looks, changes nothing.
Shows its evidence
Data and sources behind every suggestion.
A person approves
Nothing happens without a yes.
Automatic, if approved
Only for actions you list.
Everything logged
Suggestions, approvals, actions.
Agents we can build
AI automation for everyday work. Illustrative, not client work.
Support triage
Sort, tag and draft replies.
Lead qualification
Score and route new leads.
Document processing
Read forms, fill fields.
Report preparation
Weekly numbers, summarised.
Knowledge assistant
Answers, source linked.
Back-office workflows
Multi-step, with approval gates.
- Your cloud account
- Your regions
- Your access rules
- Your logs
How we work
Start small, prove it on your own examples, then scale.
- 1ReviewUse cases and data
- 2PrototypeTested on your examples
- 3PilotYour team approves
- 4RolloutStaged, in your setup
Timeline
| Phase | Typical duration | You get |
|---|---|---|
| Use-case and data review | 1 to 2 weeks | Ranked use cases; data and access checked |
| Prototype | 2 to 4 weeks | Working prototype, tested on your examples |
| Pilot | 3 to 6 weeks | Pilot with your team; every action approved |
| Production rollout | 2 to 6 weeks | Staged rollout in your setup, monitoring on |
| Ongoing | Ongoing, as agreed | Reviews, test runs, new tasks |
Typical ranges in weeks, not a guarantee.
How to choose an AI agent development company in Madrid
Six checks before you sign, whoever you hire.
Proof on your own examples
Ask for a prototype tested on your real cases before you commit to a pilot.
UnFoldMart: A prototype tested on your real examples first.
Human approval and logging
Who approves which actions, and is every suggestion, approval and action logged?
UnFoldMart: Read-only first, approval before actions, everything logged.
Testing before and after launch
How the agent is tested, and how prompt or model changes are checked before release.
UnFoldMart: Test sets from your own cases; changes pass release checks.
Data and hosting
Where the agent, its logs and your data run, and whether providers may train on your data.
UnFoldMart: Agents can run in your own cloud account.
Running costs
How model and hosting costs are estimated, limited and tracked.
UnFoldMart: A cost estimate during the prototype; usage limits per agent.
Ownership and handover
Who owns the code, prompts and tests, and how a handover works.
UnFoldMart: Code, prompts, tests and docs are yours.
Off-the-shelf AI tool, custom agent or in-house team?
| Custom agent (with us) | Off-the-shelf AI tool | In-house team | |
|---|---|---|---|
| Fit to your process | Built around your tools and rules | You adapt to how the tool works | Built around your process |
| Start | After scope and quote | Quick to try | After hiring |
| Data and hosting | Can run in your own cloud account | On the vendor's platform | Your choice |
| Approval and logging | Built to your rules | As far as the tool allows | What your team builds |
| Cost model | Project quote or monthly team, plus model and hosting costs | Per-seat or usage subscription | Salaries, hiring and tools |
| Best for | Multi-step work across your systems | Common tasks in standard tools | AI as a long-term core capability |
What affects the cost
No price list. Your quote depends on these factors.
- TasksHow many, how complex.
- SystemsTools and data to connect.
- DataDocuments and access.
- Risk levelApprovals and checks.
- HostingYour cloud or managed.
- LanguagesEach extra language, such as Spanish and English, needs its own test cases.
Ways to work with us
Prototype first
Prove it on real examples.
Fixed-scope pilot
Scope and price in writing.
Ongoing agent team
New tasks, monthly.
Model and hosting costs are billed by your providers; we estimate them early.
How quoting works
- Short briefThe task and the tools.
- Free 30-minute callBy video, in English.
- Written estimateScope, timeline, price.
- You decideNo obligation.
EU AI Act and GDPR
This is general information, not legal advice. Your legal adviser has the final word.
EU AI Act
Regulation (EU) 2024/1689 applies in phases: bans on certain practices and AI-literacy duties since 2 February 2025, rules for general-purpose AI models since 2 August 2025 and most other rules since 2 August 2026; some high-risk rules apply later. Many business agents are not high-risk, but transparency duties, such as telling people they are dealing with an AI, can apply.
Official source: AI Act (European Commission) ↗GDPR and the LOPDGDD
The GDPR applies whenever an agent processes personal data; Spain's Organic Law 3/2018 (LOPDGDD) adds the national rules. We plan lawful basis, data minimisation and access limits into every agent.
Official source: LOPDGDD (BOE) ↗AEPD
The Spanish Data Protection Agency (AEPD) supervises how organisations in Spain handle personal data and publishes guidance for them.
Official source: aepd.es ↗Public standards we align our work to
Working with us from Madrid
One dedicated team; projects are led by UnFoldMart from our international offices.
Our Madrid office
P.º de la Castellana, 259C, Planta 19, 28046 Madrid
By video
Workshops, demos and weekly check-ins by video.
Language
Calls and project work in English.
Time zone: Central European Time (CET/CEST)
Office-hours overlap: –
09:00–18:00 local time in both offices. Same time zone as our Amsterdam, Frankfurt, Zürich, Paris and Warsaw offices.
Company-wide figures across all our services, not AI projects.
- Zürich · in person
- Frankfurt · in person
- London
- Amsterdam
- Paris
- Madrid
- Warsaw
- Singapore
- Vancouver
- Gurugram
Frequently asked questions
AI agent development in Madrid: before you book a call.
Do you have an office in Madrid?
Yes. UnFoldMart's Madrid office is at P.º de la Castellana, 259C, Planta 19, 28046 Madrid. Workshops, demos and weekly check-ins with Madrid clients take place by video, in English.
What is an AI agent, and how is it different from a chatbot?
A chatbot answers questions. An agent also acts: it uses your tools to finish multi-step tasks, like updating a CRM record, within the permissions you set.
How much does AI agent development cost in Madrid?
Pricing is custom. Cost depends on the number and complexity of tasks, the systems to connect, your data, the risk level and hosting. After a short brief and a free 30-minute call, you get a written estimate. Model and hosting costs are billed by your providers; we estimate them early.
How long does it take to build an AI agent?
Typically: use-case and data review 1 to 2 weeks, prototype 2 to 4 weeks, pilot 3 to 6 weeks, production rollout 2 to 6 weeks. These are ranges, not guarantees.
How do you stop an agent from making costly mistakes?
It starts read-only, shows its evidence and waits for a person's approval. Only actions you list can run automatically, and everything is logged. Test sets built from your own cases run before every release.
Can the agent run in our own cloud account?
Yes. The agent, its logs and, if needed, an open-weight model can all run in your own cloud account. Otherwise we agree regions and providers that don't train on your data.
Does the EU AI Act apply to our agent?
It applies in phases, and most rules have applied since 2 August 2026. Many business agents aren't high-risk, but transparency duties can still apply. We help you classify your use case; legal sign-off stays with your counsel.
Can the agent serve customers in Spain and Latin America?
Yes. Modern language models handle Spanish well, including regional wording. We test the agent on real examples from each market you serve, and a person approves replies until the agent has proved itself.
Are you an agencia IA, or do you only advise?
We review your use cases and data, then build, test and run the agent with you, so you get working software, not only advice. Calls are by video, in English.
Can we see AI agent projects you have built?
Our published projects are web, brand and marketing work, so we don't present them as AI agent references. Instead, you judge the work on your own data: the prototype is tested on your real examples before you decide on a pilot.
Building an AI agent in Madrid?
Tell us the workflow. We'll say honestly if an agent fits.