The situation in Barcelona’s law firms
AI has already made its way into law firms, but through the back door. According to the Barómetro Sector Legal Digital 2026, which surveyed 6,512 professionals, 69% already use it, yet half of organisations have given no instructions at all and 27.4% admit to having entered real client data into ChatGPT. In a Barcelona law firm or advisory firm, that means claims, payslips, deeds and income tax returns passing through third-party servers with no contract and no record.
The profession is moving. The Consejo General de la Abogacía Española (Spain’s General Council of Lawyers) has already published a circular on documents drafted with AI and has announced another on confidentiality. And outside Spain, a US federal judge has already ruled that conversations with an AI are not covered by attorney-client privilege. The answer is not to ban AI, but to keep it in-house.
What private AI looks like day to day
At nine o’clock, a lawyer asks for a summary of the 400-page case file going to trial this week: timeline, parties, deadlines and key documents, each point with its citation. An adviser at the firm asks which internal circular explained how to handle expense allowances and gets the exact document. In the afternoon, a client meeting is transcribed on your own server, identifying who is speaking, and produces minutes with the outstanding tasks.
Everything runs on a machine inside your office, with individual accounts, usage logs and no per-token fees. We work this way every day: in our lab we transcribe meetings with speaker identification locally, and our private gateway has served around 11.15 billion tokens without paying for each one.
How we deploy it
- Assessment. We look at the documents you handle, which tasks take up the most hours and which data is sensitive. You come away with prioritised use cases, an architecture and a fixed proposal.
- Pilot of 2 to 4 weeks. One specific case (for example, case file search) with a small group and your real documents. We measure time saved and quality.
- Production. Final installation, accounts for the whole team, an AI use policy, training and maintenance so the models are updated without interruptions.
Catalan and Catalan law
A large share of case files in Catalonia are in Catalan, and the Codi civil de Catalunya has its own terminology. Today’s open models, together with resources from the AINA project and Softcatalà, make it possible to work in Catalan to a high standard. We have gone a step further: we built a Catalan vocabulary pipeline that raised the coverage of the ‘draft’ model that speeds up responses (a technique known as speculative decoding) from 75.2% to 99.6%. Catalan is not a second-class language for us.
When you don’t need it
If your use is limited to drafting generic texts or searching public information, with no client data, a cloud business subscription with a data processing agreement may be enough. We will tell you so in the assessment, even if it means not selling you a server.