Private LLM hosting for legal aid and public defender organizations
A dedicated DGX Spark for reading client intake files, case notes, and filings with a language model that never phones home, at a rate legal aid budgets can absorb. The instance is yours alone, in EU-Central, at $0.79/hour for a single node or $1.79/hour for a linked 256 GB cluster.
Not legal advice. This page describes GPUwerk's infrastructure and contractual commitments. It is not an opinion on your duty of confidentiality to clients, your privilege obligations, or your bar association's rules on AI tools. Check with your supervising attorney, ethics counsel, or funder compliance office before deciding what can run where.
Why this is different from a law firm's problem
We publish a separate page for private LLM hosting for law firms, and if you're a commercial firm, that page covers your situation better than this one. Legal aid clinics and public defender offices share the confidentiality problem, case files and client communications shouldn't go through a consumer AI product, but the client population and the budget constraints are different enough to warrant their own treatment.
Clients of a legal aid clinic or a public defender's office are frequently people in an especially exposed position: an asylum seeker whose case file documents the persecution they fled, a domestic violence survivor whose intake notes include an abuser's name and a current address, a tenant facing eviction whose file includes income and immigration status. A leak or an over-broad AI vendor's terms of service isn't just a confidentiality breach in the abstract for these clients, it can mean a specific, identifiable person's safety or immigration status is exposed to someone who shouldn't see it. That's a materially higher bar than the reputational risk a commercial firm's client data carries, and it's worth treating as such when deciding what tools touch this data at all.
Budget is the other half of the picture. A commercial firm can absorb enterprise AI tooling costs; a legal aid clinic running on grant funding and pro bono hours generally can't. $0.79/hour for a single Spark is the same rate every GPUwerk customer pays, not a discount we're offering legal aid organizations specifically, but it's low enough that a clinic running document review a few hours a week can fit the cost inside an existing technology line item rather than needing a new grant.
What dedicated hardware changes
A DGX Spark from GPUwerk is single-tenant: no other customer's workload runs on the machine while it's yours, and the container and workspace are removed from the node before it's offered to anyone else, as described in our Privacy Policy, section 12. You choose the model, open-weight and free to run, or a commercial model you self-host under its own license, and client intake files, case notes, and draft filings stay on that instance. We don't access, read, copy, index, or analyse what runs on it.
128 GB of unified memory on a single Spark runs a 70B-class model well enough for the tasks most clinics need: summarizing an intake interview into a structured case note, drafting a first pass at a routine filing from a template, or searching case files for prior instances of a similar fact pattern. A statewide or regional legal aid organization running this across multiple offices would want the linked 256 GB cluster for the added capacity.
Where GDPR applies
Client intake and case files are personal data under the GDPR, and for the client populations described above, often special-category data, health information, or data revealing circumstances requiring extra care. Our Data Processing Agreement under Article 28 applies to how we handle the infrastructure it runs on; the DPA states plainly that GPUwerk hosts the machine but does not access the content of the instance, and that the controller decides what runs on it. Whether an LLM is appropriate for a given client's file at all, and what safeguards your intake process needs before that file goes anywhere near a model, are questions for your supervising attorney and your organization's ethics or compliance office, not something this page can answer for you.
Practical starting points
Clinics we'd point toward this setup generally start with the lowest-sensitivity task available, not case-level client work. A common starting point is legal research support: searching statutes, regulations, and prior case summaries that are already public or already de-identified, to help a caseworker find a relevant precedent faster. A 70B-class model on a single Spark handles that well.
A second starting point, once staff are comfortable with the tool, is drafting routine filings from a template using structured, already-reviewed case data, with an attorney reviewing every draft before it's filed. Case notes and client intake summaries involving identifying details of vulnerable clients are a further step we'd suggest approaching last, and only with sign-off from whoever holds your ethics or data protection responsibilities, given what's at stake if that data is handled wrong. None of this requires a perfect model; it requires one private enough that a resource-constrained organization can use it without a shared-cloud vendor's data practices becoming a client safety problem.
Getting started
Deploy a Spark directly from the console, or if you'd rather scope the workload first, an AI Opportunity Session covers your case management system, your document types, and what a first pilot should look like, starting from the lowest-risk task, before you commit to anything involving client-identifying data.
Back up before you terminate. Terminating deletes the workspace and GPUwerk's recovery copy of it. Export what you need before you stop paying for it.
Related pages
Run your case work on hardware that's only yours.
Single-node Spark from $0.79/hour, no shared GPUs, no egress fees.
Deploy a Spark Book an engagement