For law firms
Your private legal AI, turnkey
A pre-configured box, shipped to your office. Plug it in — you're operational.
~70% of requests processed locally. ~30% routed to the best cloud LLMs (Claude, GPT, Mistral) — only anonymized excerpts. Always the best model, never locked in.
Extraction, Q&A, summaries, drafting
Anonymized before sending
The Problem
Three risks your firm takes every day
Cloud exposure
ChatGPT, Claude, Harvey — your queries pass through third-party cloud servers. Client data leaves your office, potentially waiving privilege.
Vendor lock-in
Harvey is locked to OpenAI. Claude for Legal, to Anthropic. The best model changes every month — your tool doesn't.
Privilege at risk
ABA Model Rules and state bar ethics opinions increasingly scrutinize AI use. Most legal AI tools weren't built with privilege preservation in mind.
The Solution
Private. Turnkey. Always the best model.
A physical box shipped to your office, pre-loaded with the RAGbase software. Plug it in — you're operational.
Your data stays at the firm
Local models for everyday tasks. Cloud LLMs (Claude, GPT, Mistral) for complex queries — only anonymized excerpts.
Turnkey
Hardware provided and pre-configured. No infrastructure to manage. No IT department needed. Plug and play.
Always the best model
Model-agnostic. When a better model releases, we deploy it on your box. Your system stays the same.
Hybrid private architecture
Firm documents
Emails, DMS, contracts, briefs
RAGbase Box
At your office
~70% Local
Extraction, Q&A, summaries
~30% Cloud LLMs
Complex reasoning
Only anonymized excerpts are sent to cloud LLMs. Full documents never leave the box.
In practice
40 legal tasks — where does your data go?
From client intake to final pleading, every task is assigned to the right sovereignty layer.
30 tasks never leave the building. 10 need a US-owned model.
Click any task to see difficulty, criticality, model tier, and verification.
Compliance
Built for attorney-client privilege
As state bar ethics opinions and ABA guidance on AI tighten, the RAGbase Box is architected for compliance by default.
On-premise processing
Client documents stay on the box at your office. 70% of tasks are processed locally — no external connection.
Anonymization pipeline
Built-in de-identification before any data reaches cloud LLMs. Only anonymized excerpts are transmitted.
Privilege preservation
Full client documents never leave the firm's premises. Reasoning traces and agent logs stay on the box.
Complete audit trail
Every query, every response, every source is logged. Full audit trail on the box for compliance review.
No training on your data
Neither local models nor cloud providers use your data for training. Contractually guaranteed via enterprise APIs.
ABA & state bar alignment
Architecture designed for ABA Model Rules 1.1, 1.6, 5.3. Documentation provided for your firm's compliance review.
Plans
All-inclusive per-user leasing
One monthly rate per user: box, software, updates, and support included. No upfront hardware cost. 24-month lease.
Essential
Solo / small team — 1–5 users
24-month lease — box + software included
From 3 users
Hardware included
Mac Mini M4 Pro — 48 GB RAM, 1 TB SSD
Qwen 3 30B, Mistral Small 24B, Llama 3 8B or similar (Q5–Q6)
- RAGbase Core: semantic index, chat, doc generation, email automation
- Q&A, summaries, translation
- Extraction and classification
- Model updates included
- Email support
Pro
Mid-size firm — 6–15 users
24-month lease — box + software included
Hardware included
Mac Studio M4 Max — 128 GB RAM, 1 TB SSD
Llama 3.3 70B (Q4–Q5), Qwen 3 30B (Q8), DeepSeek R1 32B or similar
- RAGbase Core: semantic index, chat, doc generation, email automation
- Native 70B models — no offloading, best quality
- Fine-tuning on your corpus
- Automated workflow agents
- Brief and memo drafting
- Priority support
Premium
Large firm — 16–30 users
24-month lease — box + software included
Hardware included
Linux dual-GPU station — 2× RTX 5090 (64 GB VRAM), 128 GB RAM, 2 TB NVMe
Llama 3.3 70B (Q5–Q8), Qwen 3 235B-A22B, multi-concurrent inference
- RAGbase Core: semantic index, chat, doc generation, email automation
- Dual-GPU for high-performance parallel processing
- Advanced fine-tuning on your corpus
- Custom client portal
- Dedicated SLA
- Strategic advisory
Enterprise
For firms with 30+ users
Volume-based per-user rate — 24-month lease
Hardware included
Multi-GPU custom — 4× RTX 5090+ or RTX PRO, Linux rack server
70B+ full-precision models, multi-concurrent, calibrated to your use cases
- Everything in Premium, plus:
- Volume-based per-user discount
- Custom AI agents for your workflows
- Multi-GPU sized for 20–50 concurrent inferences
- Dedicated integration project (2–3 weeks)
- "Fully managed" option by RAGbase
- Formalized SLAs (response time by severity)
- DMS, case management, and email integration
24-month lease, all-inclusive: hardware, software license, updates, monitoring, and support. Renews with current-generation hardware at term. Details below.
Monthly lease
What's included in your lease
RAGbase software license (agents, RAG, traceability engine)
Local model updates as soon as they release
Intelligent routing to cloud LLMs (~30%) — Claude, GPT, Mistral based on task, anonymized excerpts
Box monitoring, security patches, encrypted backups
Leased hardware — maintenance, replacement, and refresh at term
Technical support (level depends on plan)
Fine-tuning on your firm's corpus
Lease billed monthly per user. 24-month term, renewable with current-generation hardware.
Comparison
RAGbase Box vs. cloud legal AI
| RAGbase Box | Harvey | Claude for Legal | Generic SaaS | |
|---|---|---|---|---|
| Where your data runs | Box at your office (local) + anonymized cloud LLMs | Cloud (Azure OpenAI) | Cloud (Anthropic) | Shared cloud |
| Data privacy | 70% local + anonymized cloud via enterprise APIs | Full documents in cloud | Full documents in cloud | Variable, often unencrypted |
| AI model | Model-agnostic — always the best | Locked to OpenAI | Locked to Claude | Locked to one vendor |
| Pricing | All-inclusive per user — hardware + software leased, no CapEx | ~$500–1,500/user/mo, software only | ~$200–400/user/mo, software only | Per-seat, software only |
| Your internal docs | 15+ years indexed, stay at the firm | Manual upload, external cloud | No permanent indexing | Limited, external cloud |
| Attorney-client privilege | Documents never leave the box | Third-party cloud access | Third-party cloud access | Third-party cloud access |
| If you leave | Your data and indexes are returned — never sat in a third-party cloud | Everything stops, data in vendor cloud | Everything stops, data in vendor cloud | Everything stops |
Frequently asked questions
FAQ
Routine tasks (extraction, classification, Q&A, summaries, translation) run directly on the box at your office with no external connection. Complex requests (advanced legal reasoning, long-form drafting) are routed to the best cloud LLMs — Claude, GPT, or Mistral depending on the task. Only anonymized and de-identified excerpts are transmitted, never your full documents.
Full client documents never leave your office. The RAGbase Box processes 70% of tasks locally. For the 30% routed to cloud LLMs, only anonymized excerpts are sent via enterprise APIs that contractually guarantee zero retention and zero training on your data. This architecture was designed to preserve privilege by default.
You choose: renew the lease — we then swap the hardware for a current-generation configuration — or stop. If you stop, the box is collected and we hand back all of your data and indexes. Because your data never left your office, there is nothing to retrieve from a third-party cloud.
For routine tasks (70% of volume), local models deliver excellent results with near-instant response times. For complex tasks, routing to the best cloud LLMs (Claude, GPT, Mistral) gives you access to frontier models. The hybrid system combines the best of both worlds.
The RAGbase Box is not locked to any model vendor. When a better local model releases (Llama, Mistral, Qwen, etc.), we deploy it on your box. For cloud, we route to the best model available — Claude, GPT, or Mistral depending on the task. Your interface and workflows stay the same — only the engine improves.
Before any data is sent, it passes through an anonymization and de-identification pipeline built into the box. Only the excerpts necessary for the request are transmitted — never full documents. We use enterprise APIs from each provider (Anthropic, OpenAI, Mistral) that contractually guarantee zero retention and zero training use.
Yes. The per-user rate includes the pre-configured box (Mac or Linux station depending on plan) on lease, the RAGbase software license, updates, monitoring, and support. No upfront hardware cost. 24-month term, renewable with current-generation hardware.
The RAGbase Box architecture is designed for compliance with ABA Model Rules and state bar ethics opinions on AI use. Client documents stay on-premise, cloud data is anonymized, and full audit trails are maintained. We provide documentation to support your firm's compliance review.
Order
How to order your RAGbase Box
A simple three-step process. You're operational within days.
Free consultation
30 minutes
Book a 30-min call with our team. We analyze your needs, query volume, and constraints (number of attorneys, DMS, security requirements).
Custom configuration
Quote in 24h
We size the right hardware, select the optimal AI models for your firm, and prepare a detailed quote.
Delivery & setup
Shipped to your office
The box ships pre-configured to your office. Plug it in, our team completes setup remotely — you're operational.
Ready to order?
Book your free 30-minute consultation. No commitment — we answer all your questions.
No commitment · Quote within 24h · Delivered in days
Ready to take control of your data?
Book 30 minutes with our team. We analyze your needs, size the right box, and ship it to your office.
Free consultation · No commitment