AI agents
Agents that work with your tools and your rules: they read, draft, update systems and send alerts. With local models and human approval where it matters.
We build AI agents that carry out real tasks in your business (email, documents, ERP, CRM) with local models and human approval where it matters. For SMEs and law firms with repetitive processes. They go into production in 4 to 8 weeks, after a pilot with real users. We use n8n, Hermes Agent or OpenCode and log every action.
Almost every business has work that nobody wants to do and that never ends: reading emails and routing them, copying data from a PDF into the ERP, preparing the same report every Monday, answering the same ten questions. Cloud automation platforms only half solve it: every AI step sends your data to a third party and, if it is client or health data, the GDPR requires a contract and safeguards; if you are subject to NIS2, it is one more supplier in the chain. And if the agent talks to clients, since 2 August 2026 the EU AI Act (Art. 50) has required that they know they are talking to an AI.
A local AI agent does that work with your tools, with models running on your server and with clear rules about what it can do on its own.
Controls aren’t left to the model’s goodwill: step limits, verified approvals, no deleting without a backup and isolation by client or department. If you don’t have a server yet, we start with local AI for business.
If the process changes every week or nobody can explain how it is done, it needs sorting out first. If the volume is small, just a few tasks a month, automation doesn’t pay off. If a simple rule in n8n solves it without AI, we’ll tell you: it will be simpler and easier to maintain. And we don’t automate decisions with legal or medical effects on people without human review: it isn’t a good idea, and the GDPR (Art. 22) restricts it.
Dozens of agents work with local models in our Barcelona lab. A team of agents builds WordPress Gutenberg blocks for a Barcelona digital agency, following its methodology and coding standards. On a family estate in the Penedès, an agent keeps the field logbook from voice notes in Catalan and diagnoses vine pests from a photo. We know where agents go wrong because we have watched them go wrong, which is why every safety rule we apply comes from a real case.
An assistant that knows your case files and templates, runs inside the firm and never exposes a single piece of client data.
Your technical know-how, searchable in seconds and without leaving the company network.
AI in Catalan for administrative work, installed inside your security perimeter.
A field agent that speaks Catalan, knows your farm and keeps the logbook for you.
Agents that work with your standards and your files, inside the studio and respecting every NDA.
Yes, because the agent works with models that run on your server, through the LiteLLM gateway, so your data never leaves the office. If you don’t have it yet, we start with the base local AI installation and build the agent on top; if you already have a machine with a GPU, we make use of it.
Automation follows fixed rules; an agent understands text, decides the next step and uses tools. We combine them: fixed rules where they are enough, and the language model only where something has to be read, drafted or decided.
Only the things you decide. We define in writing what it can do on its own, what requires your approval and what is off limits, and we enforce it with technical controls, not just with instructions to the model.
Yes. Much of the work is remote. For the discovery workshop we prefer to meet in person if you are in or near Barcelona; if not, we do it by video call.
We build AI agents that carry out real tasks in your business (email, documents, ERP, CRM) with local models and human approval where it matters. For SMEs and law firms with repetitive processes. They go into production in 4 to 8 weeks, after a pilot with real users. We use n8n, Hermes Agent or OpenCode and log every action.