The standard business case for legal AI assumes documents leave your network and travel to a vendor's cloud. For sworn statements, witness evidence, privileged advice and regulator submissions, that assumption is the problem. This article makes the business case for local-first legal AI: tools that run on hardware your organisation controls, so the document never crosses the perimeter.
Every legal AI product is either cloud-hosted or local-first. Cloud-hosted tools upload documents to vendor infrastructure for processing. Local-first tools run the model on the user's device or inside your network. The cloud model is strong for contract review at scale, clause libraries and deal velocity. The local-first model is required for affidavits, witness statements, criminal defence, family matters and any document with evidential weight.
If the work is sensitive enough to be regulated, why upload it? If the risks are real, who carries them after a vendor breach? If your prior matters are your competitive edge, why feed them into a shared system? If the tool has to meet teams where they work, what about teams that work offline, in court rooms, custody suites, site visits and cross-border travel? The standard business case for cloud legal AI does not answer these questions.
No data in transit and no third-party processor simplifies DPIAs and data processing agreements. Predictable behaviour at outage: a vendor going down does not stop a hearing. No model drift you did not authorise: the model is the one you signed off on. Audit trails belong to you, not to a vendor's retrieval system.
Client trust is a feature: "Your draft never leaves our perimeter" is a sentence that converts. Jurisdictional flexibility: the data is where the user is, sidestepping most data residency questions. Cost flatness at scale: local-first runs on hardware you already own, so the marginal cost of the ten-thousandth affidavit equals the first.
Three questions sort any legal AI use case. Would you be uncomfortable explaining the document's path to a regulator after a vendor breach? Does the document have evidential weight in itself, such as a sworn statement or signed exhibit? Would a network failure on the day stop the work? Any yes means local-first. Everything else can sit in a cloud-hosted tool. The two architectures are complements, not rivals.
myAffidavit is a 100% offline AI affidavit and witness statement generator. The model runs on the user's device. Drafts, exhibits and signatures never leave the machine without an explicit, audited share. The user is the source of every fact: no case retrieval, no citation generation, no hallucination surface. Court-ready format, jurisdiction-aware, with an audit trail embedded in the PDF/A-3 export. Pre-order at £99.
Pre-order myAffidavit - Offline AI Affidavit Generator