AI agent development company in Zürich
Custom AI agents for Zürich companies, built with Swiss data protection in mind and human approval at every step. Meet us at Hardstrasse 201.
- 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 Hardstrasse 201, 8005 Zürich. We build custom AI agents, knowledge assistants and multi-step workflows for Zürich companies, with the Swiss FADP and, for financial firms, FINMA's expectations in mind, with human approval, logging and testing built in; agents can run in your own cloud account. You can meet us at our office or by video. 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 | Hardstrasse 201, 8005 Zürich, Switzerland |
| Meetings | In person at our Zürich office or by video |
| 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 Zürich businesses
Zürich is Switzerland's largest city and its main business and financial centre, with banks, insurers, ETH Zurich and a strong technology scene. Agents here often need to work in more than one language and meet strict expectations on confidentiality.
Zürich at a glance
- IndustriesBanking, insurance, wealth management, life sciences, tech
- Data lawRevised FADP (nDSG); FINMA rules for financial firms
- AI rulesNo AI-specific act; the FADP applies to AI
- LanguagesGerman, with French and Italian across Switzerland; English in business
- Time zoneCET in winter, CEST in summer
Banking, insurance and wealth
Agents that prepare client files, reports and checks for a person to approve, fully logged.
Health and life sciences
Document assistants that answer from approved sources and show where each answer comes from.
Tech and start-ups
Agent features for your product, tested on real cases before launch.
What our Zürich 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 Zürich
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.
- Regulatory reviewsFINMA-related documentation and security reviews for financial clients add time.
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.
Swiss data protection and AI
This is general information, not legal advice. Your legal adviser has the final word.
Revised FADP and AI
Switzerland has no AI-specific act. The FDPIC has stated that the Federal Act on Data Protection, in force since 1 September 2023, applies directly to AI-supported processing: purpose, functionality and data sources must be transparent, and people can ask for automated decisions to be reviewed by a person.
Official source: FDPIC statement (edoeb.admin.ch) ↗FINMA Guidance 08/2024
FINMA's guidance of 18 December 2024 on governance and risk management when using AI covers topics such as an inventory and risk classification of AI applications, data quality, testing, monitoring and documentation. We document our part for your review.
Official source: FINMA Guidance 08/2024 (finma.ch) ↗EU AI Act, if you serve the EU
The EU AI Act can also apply to companies outside the EU when their AI systems are placed on the EU market or their output is used in the EU. It applies in phases; most rules have applied since 2 August 2026. We help you classify your use case.
Official source: AI Act (European Commission) ↗Public standards we align our work to
Working with us from Zürich
One dedicated team; projects are led by UnFoldMart from our international offices.
Our Zürich office
Hardstrasse 201, 8005 Zürich
In person or by video
Meet the team at our office or talk 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 Frankfurt, Amsterdam, Paris, Madrid 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 Zürich: before you book a call.
Do you have an office in Zürich?
Yes. UnFoldMart's Zürich office is at Hardstrasse 201, 8005 Zürich. You can meet the team there in person or talk by video.
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 Zürich?
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.
Is there an AI law in Switzerland?
Switzerland has no AI-specific act. The revised FADP applies directly to AI that processes personal data, as the FDPIC has pointed out, and FINMA has published guidance for the firms it supervises. If you serve the EU, the EU AI Act can apply. Your legal adviser has the final word.
Do you build AI agents for banks and insurers in Zürich?
Yes. Agents start read-only, show their evidence and wait for a person's approval, and every action is logged. We document data handling, access and testing so your risk team can review it against FINMA's expectations.
Can the agent work in German, French and Italian?
Yes, modern language models handle German, French, Italian and English. We test the agent on real examples in each language you need. The free first call on this page is in English; our German page offers it in German.
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 Zürich?
Tell us the workflow. We'll say honestly if an agent fits.