Frequently Asked Questions

We deploy full autonomous AI agent harnesses entirely on your premises — on your hardware, in your VPC, or in fully air-gapped environments. Not a chatbot. Not a question-and-answer tool. A complete agentic system that reads, reasons, acts, and coordinates work on your behalf. Read our complete guide to on-premises AI →

No internet connection is required. Your data never leaves your network or touches any external cloud provider. The agents run exclusively on your own documents, policies, contracts, procedures, and knowledge — they monitor channels, triage issues, draft responses, execute routine tasks, and handle complex workflows autonomously. All locally. Zero leak.

Here are examples we’ve seen with law firms running our agentic system:

• An always-on agent monitors incoming queries, retrieves context from your knowledge base, and drafts compliant responses overnight — escalating only when human review is needed.
• Upload new regulations or client contracts and have the system automatically highlight risks, inconsistencies, and next steps — then surface findings directly to the right team.
• Produce first-draft letters, NDAs, engagement letters, or other standard documents in your firm’s style, referencing past examples from your repository — autonomously.
• Handle due-diligence requests during transactions by having agents search, cross-reference, and summarise your full document set without manual prompting.
• Give junior lawyers and paralegals reliable, context-aware explanations of clauses, procedures, or points of law — without sending anything externally.

Want a deeper look? Read our analysis of what generative AI actually does for in-house legal teams →

Pricing depends on your setup: model size, number of users, storage volume, whether it’s air-gapped, and any professional services for data ingestion, customisation, or agentic extensions. Our complete guide breaks down the hardware costs, running costs, and payback period in detail. Investing in your own AI infrastructure early is one of the smartest decisions a business can make — here is why.

Under conservative assumptions, many clients see payback within 6 months through time saved and reduced risk exposure. Agentic features can accelerate ROI further by automating ongoing workflows. Try our ROI calculator to model the numbers for your firm.

Contact us for a tailored discussion and quote — no obligation.

We refresh the core models roughly every six months, timed to major capability jumps in the LLM field that justify the update effort. The performance gap between local open-weight models and cloud SOTA has effectively closed for most business workloads — a 27B-parameter model can now tie Claude 4.5 Opus on coding benchmarks while running entirely offline.

Updates are fully offline: we deliver new model weights via secure transfer (encrypted drives or similar), and your team — or ours during onboarding/support — applies them.

Since the entire system has no internet exposure, there are no ongoing security patches or vulnerability scans needed for the deployment itself — no external attack surface. This holds true even when running proactive agents.

Data integration is handled end-to-end by our team as part of deployment. We work with you to:

• Organise and securely export your documents and knowledge sources
• Convert formats as needed
• Build the searchable knowledge index that feeds the agents
• Configure the agentic harness — what tools each agent has, what workflows it runs, and how it escalates to human review

You don’t manage the technical pipeline — we set it up, test it with your data, and hand over a working system. Ongoing additions or tweaks are straightforward.

We can handle that. Through optional professional services, we coordinate secure scanning and OCR to turn physical files into searchable digital text ready for ingestion. This is quoted separately based on volume and complexity — get in touch for details and a realistic timeline. Digitised content then feeds seamlessly into both standard queries and any proactive agent behaviours.

Yes — archived email exports (PST, EML, or similar) are one of the richest sources of institutional knowledge we incorporate.

For compliance, we stick to non-live, historical exports only — no live mailbox connections. We can set up a completely local, agent-triggered refresh schedule (e.g., weekly ingestions) that keeps the knowledge base current without ongoing risk.

We fully support the most common business formats, including:

• PDF
• Microsoft Word (.doc, .docx)
• Microsoft Excel (.xls, .xlsx)
• Microsoft PowerPoint (.ppt, .pptx)
• Plain text, Markdown, emails, and scanned images via OCR

The pipeline manages both structured and unstructured content effectively — whether for on-demand queries or feeding into proactive agent tools.

Refresh cadence is entirely up to you and defined in your service agreement. Options include:

• Manual/on-demand (e.g., after major document updates)
• Scheduled automated syncs (daily, weekly, monthly) where feasible within your security rules

For agentic deployments, more frequent controlled refreshes can keep proactive behaviours highly relevant. We discuss the right frequency during setup.

In true air-gapped setups (common for our highest-security clients), the system runs isolated by design. Practical access options we help implement:

• Run the interface on a dedicated secure workstation or thin client connected via internal LAN (browser-based or approved app).
• Use controlled removable media (encrypted USB drives) to transfer queries in and responses out — sneakernet style.
• For agentic features: configure internal triggers (e.g., file drops, scheduled checks, or approved messaging channels on a segmented network) so the AI monitors and acts without needing constant manual input.

Many clients find this deliberate separation actually improves focus, auditability, and compliance — especially when agents handle routine monitoring autonomously.

Even when data resides in a highly secure cloud environment, sending that data to a third-party LLM provider creates a material risk: your confidential information temporarily leaves your perimeter and is processed by someone else’s infrastructure.

This isn’t hypothetical. Workers in Nairobi reported intimate footage from connected glasses streaming straight to the cloud, and researchers can deanonymise internet users for less than the cost of a coffee. The issue isn’t malice — it’s how cloud AI works by design.

With JD Fortress:

• Your data never leaves your controlled environment to reach an external LLM.
• You keep using your existing secure cloud storage as the source of truth.
• All AI inference happens locally — so answers, insights, and generated content remain entirely within your fortress.

This gives you the best of both worlds: the convenience of cloud-hosted data with the ironclad privacy of offline, on-premises AI.

Cloud AI services require your data to leave your network and be processed on someone else’s infrastructure. That creates attack surfaces you can’t control.

A hacker spent a month using Claude to breach the Mexican government — 195 million taxpayer records, voter data, government credentials. The AI refused at first, then it didn’t. Read what that means for enterprise AI →

With on-premises AI, there is no external attack surface. The model runs where your data already lives. No API calls to third parties. No data in transit. No shared multi-tenant infrastructure. Even the most sophisticated guardrail failures become irrelevant when the model never sees data it shouldn’t.

The performance gap has effectively closed for most business workloads. A 27B-parameter open-weight model now ties Claude 4.5 Opus on coding benchmarks and runs entirely offline on modest hardware.

For document analysis, contract review, legal research, policy Q&A, and internal knowledge retrieval — the quality is indistinguishable from cloud models in practice. The difference is your data stays where it belongs.

We select and tune models based on your specific use case, so you get the right capability without overpaying for capacity you don’t need.

You could attempt it, but most organisations find the gap between "we downloaded a model" and "our team actually uses it daily" wider than expected. OpenClaw showed the community what a real agentic harness looks like — but turning that into something your legal, compliance, or operations team can rely on requires more than downloading a tool.

We handle the hard parts: model selection, agent harness architecture, data ingestion at scale, retrieval tuning, air-gapped deployment, and ongoing support. You get a working system, not a research project. And you own the hardware — so when GPU capacity becomes scarce, you are not queueing behind a cloud provider.

Our clients are law firms, financial services, healthcare providers, and charities — not AI specialists. If they can run it, you can too.