AI agents integrated with your ERP, CRM, WhatsApp Business or phone system. We start with one concrete use case, measure it and grow from there. No 12-month moonshots. No promises we can't deliver.
Companies that trust VisibleSoft
Tap a logo to read the testimonial
Three years ago an AI project for an SMB cost eight times as much. Today the entry ticket is under €8,000.
In 2026 a Spanish SMB can integrate an AI agent with its system in 4–6 weeks and an entry ticket below €8,000. Three years ago that cost eight times more. The difference is that APIs are mature, Spanish-language models work well and integration with existing systems is now standard.
What hasn't changed: most AI projects fail because the problem is poorly defined, not because of the technology. That's why we start with one concrete use case and clear metrics before we scale.
4–6 weeks
from diagnostic to production
From €5,500
entry ticket for a production pilot
2 out of 3
AI projects fail because the problem was badly defined, not because of the technology
Not theoretical examples. The four we get asked for most, with indicative pilot pricing.
Businesses with after-hours or repetitive calls (law firms, clinics, accountants, shops, workshops).
From €5,500 + €250–400/month depending on volume
Read the full caseBusinesses that live on WhatsApp with their customers (B2B, distribution, services).
From €4,500 + €200/month
Talk about this caseCompanies with high volume of invoices, delivery notes, contracts or expense receipts.
From €6,500 depending on volume
Talk about this caseTeams that waste time searching for information in SharePoint, Drive, Notion or internal docs.
From €7,500 + €300/month
Talk about this caseThree phases. If there's no case, we'll tell you at the diagnostic — neither of us wastes time.
Not sure what your case is? Start with the diagnostic — if there isn't one, we'll say so.
Built first for VisibleSoft. We adapt it for other companies with your voice, your services and your integrations.
Embedded in Cerrajería Carlos Rodríguez's management software. It polishes the text without data ever leaving the server.
Four rules and one constant: your data doesn't leave the EU if it doesn't have to.
We don't marry an AI vendor. We pick the model and the infrastructure according to the sensitivity of your data, the use case and the budget. Three rules we always honour: data in EU territory when possible, no vendor lock-in for the model, and per-use cost controlled from day one.
Anthropic Claude · OpenAI GPT · Mistral · open-weight on-prem when it applies
n8n · in-house .NET and Python APIs · Make
Telnyx · ElevenLabs · Whisper
AWS Europe · Hetzner · customer's own servers
Compliance by design from sprint one, not a last-minute scramble.
Personal data handled per GDPR from the first prototype. Documentation ready for your data protection officer.
The European AI Regulation becomes mandatory on 2 August 2026. Your AI systems are documented and risk-classified by design.
Every agent has a configurable spending cap. No surprises on the monthly AI provider bill.
The doubts that come up again and again before signing.
Every agent has a configurable monthly cap. When it gets close to the limit, we notify you. For high volumes we optimise with caching and cheaper models for intermediate steps. The initial proposal includes a per-use cost estimate based on your expected volume.
It depends on the model. With providers like Anthropic or OpenAI via API, data is not used for training and is deleted in short timeframes. For sensitive data (health, legal) we use models in EU territory or deployed in your own infrastructure. We decide this at the diagnostic.
The European AI Regulation becomes mandatory on 2 August 2026. Most SMB agents fall under limited or minimal risk categories. We document classification, purpose and traceability from the first sprint so compliance is automatic, not a last-minute scramble.
Yes. Most integrations go against a database, REST/SOAP APIs or files. If your system only allows manual exports, we handle that too — although that's usually when a small parallel modernisation project pops up.
A typical pilot is in production 4–6 weeks after signing. Simple cases (WhatsApp chatbot against existing FAQs) take 3 weeks. Complex cases with several critical integrations take 8–10 weeks.
All agents have configurable guardrails: when to escalate to a human, which topics to avoid, what never to claim. We also log every interaction for later audit. For critical cases (transactions, legal decisions) the agent never closes alone: it proposes and a human validates.
The qualification call is free. You leave with three things: whether your case is viable, an indicative budget and what the alternatives are if custom development doesn't make sense.