Grants · Catalonia · 2026 call
ACCIÓ, the Catalan government's business agency, pays 75% of a technology exploration project: a proof of concept or pilot that answers a question you genuinely don't know the answer to yet. This page is my own reading of the 2026 call, and what Clutch Developer can build inside it.
Applications close on 21 September 2026 at 14:00. A solid application takes about two weeks to prepare, so if you're reading this in September, the call to make is this week.
The numbers
Official source
Everything on this page is a summary. The binding text is ACCIÓ's own page and the two DOGC publications behind it: the regulatory bases in DOGC 9708 (15 July 2026, Resolution EMT/2403/2026) and the 2026 call in DOGC 9718 (29 July 2026). Applications go through Canal Empresa with your company's digital certificate.
I'm a technology partner, not a grant consultancy. Figures here come from the 2026 call as published and may be updated by ACCIÓ. Check the bases before you commit a budget.
The technology exploration line exists for one thing: reducing uncertainty. ACCIÓ wants companies to test whether an advanced technology works for them before they invest seriously in it. You put a hypothesis on paper, build the smallest thing that can prove or disprove it on your real data and processes, and finish with objective evidence either way.
That framing matters more than the technology. A project that says "we will implement an AI chatbot" is a deployment and doesn't fit. A project that says "we will find out whether an AI agent can close 60% of our support tickets without a human, and measure where it fails" is exploration and fits perfectly. Same work, different question.
For 2026 the agency has singled out agentic AI — autonomous agents that execute a task end to end, where the outcome carries measurable uncertainty — as a priority area. That is exactly the kind of work an AI-first product studio does every week, which is why this page exists.
What we can build inside it
Each one is a genuine open question, sized for a €20,000–€40,000 budget, and built to finish with evidence — a number, not a slide. All run in fixed-price sprints, so the external-provider line in your application is exact from day one.
Can an AI agent process X% of our orders, invoices or tickets end to end, with an error rate below a human's?
An agent that reads incoming work from your real systems, takes the actions a person would take, and logs every decision. We measure accuracy, coverage and cost per case against your current process, on your live data.
Can a model running locally on iOS and Android do this task well enough that no data leaves the phone?
Twelve years of native mobile, applied to the newest on-device models. A proof of concept inside your app or a standalone one, validated on real users' devices for accuracy, latency and battery.
Can an assistant grounded in our documents, catalogue or history answer customers or staff correctly enough to trust?
Retrieval on your data, an assistant that cites its sources, and a validation protocol with real users — precision, hallucination rate, deflection. Web, mobile or inside the tools you already use.
Can a phone camera plus a model spot defects, read documents or count stock reliably in our conditions?
Mobile capture, a vision model tuned to your environment, and a pilot with the people who'd use it. We report detection rates in your lighting, your products, your workflow — not the vendor's lab.
Can a voice agent handle our inbound calls — bookings, status, FAQs — to the point where customers complete the task?
A phone or in-app voice agent connected to your calendar, CRM or order system, piloted on a slice of real traffic with human fallback. Task completion, handover rate and satisfaction, measured.
Is there a new digital product or service line here — and will real users pay attention to it?
An MVP of a product that only became possible with current models, put in front of real users to validate demand and behaviour. Native or Flutter, shipped to TestFlight and internal testing, instrumented from day one.
How we'd do it
I handle the technical side: hypothesis, scope, budget structure and the technical memory. The administrative submission is done by your company with its digital certificate, on your own or with your usual gestor or grant consultancy. That split keeps each party doing what it's good at.
That's the minimum eligible budget, and 75% of it comes back. Your own staff hours count toward it, so a €7,500 sprint plus internal time plus the audit line reaches the floor for many small companies. We size it honestly on the call.
That is a valid outcome for this line. The call funds exploration, and a well-documented negative result — with the evidence of why — is a completed project. You still get the grant, and you didn't spend a real rollout budget finding out.
Only if there's still a genuine open question to test. Costs before approval aren't eligible, and a project that reads as "finish and ship what we have" is a deployment. Reframe it around what you don't yet know, or pick a different line.
Whatever the hypothesis needs. Usually Claude for agents and reasoning, open models on-device when privacy or cost demands it, and your existing stack for integration. The choice is part of what the pilot validates.
Not for the same costs — the call is incompatible with any other aid that would exceed the intensity limits for identical expenses. Different costs, different projects, yes. Your grant advisor should confirm your specific case.
Next step
Tell me what you'd want to find out. If it fits the call, we'll have a hypothesis and a budget on paper within days — well before the deadline.
Contact
Tell me the process, the data you have, and the question you'd like answered. Remote-first from Barcelona — we start with a short call.
Skip the form and grab a slot straight away.
Book a 30-min call