Legal answers in minutes.
Proof in every sentence.
Contract review, compliance answers, and risk flags — grounded in your company’s own approved legal knowledge, powered by a private AI model that never sends a single document outside your walls.
Install in 2 Steps See the Story
GPL-3.0 is a strong copyleft license — shipping it inside your product requires releasing your source code under the same license GPL-3.0 §5 §10.
Sources: gpl-3.0-only.txt · sha-256 verified · approved by Legal
From a pile of documents to an answer you can trust
Watch how one contract travels through the system — and why the answer at the end is different from anything a public chatbot can give you.
Your documents
Contracts, policies, regulations, licenses — dropped into nine organized vaults
AI builds knowledge
Classifies, extracts clauses, summarizes, generates Q&A, flags risk — 7 automated steps
Your lawyer approves
Nothing goes live without a named reviewer’s sign-off
human gate
Indexed twice
Approved knowledge and the original text — searched together, always in sync
Cited answer
Verdict + clickable citations + checksum trail back to the exact source file
And if a source document ever changes? Its knowledge is automatically pulled from the index and sent back for re-approval. Stale answers can’t survive here.
Not another chatbot. A governed legal system.
The difference isn’t the AI — it’s everything wrapped around it.
| Public AI chatbot | External AI legal SaaS | Enterprise Legal AI Solution | |
|---|---|---|---|
| Your documents stay in-house | ✕ sent to public cloud | ✕ sent to vendor cloud | ✓ never leave your network |
| Answers cite verifiable sources | ✕ answers from memory | △ vendor’s corpus | ✓ your corpus, checksum-traceable |
| Human approval before knowledge goes live | ✕ | ✕ | ✓ named-reviewer gate |
| Knows your playbook & red lines | ✕ generic | △ limited config | ✓ per-team practice profiles |
| Choose & own the AI model | ✕ theirs | ✕ theirs | ✓ any private model — your pick |
| Escalates to your lawyers | ✕ | △ | ✓ built into every skill |
Pick a private model. We’ll size the workstation.
The solution is model-agnostic — any OpenAI-compatible private model plugs in. Our reference standard was verified on a 32B-class on-premise model; choose your tier and see the ABS workstation class that runs it.
- Single-team pilots
- Skill testing & light Q&A
- Fastest responses
- Department-grade assistant
- Knowledge-base builds
- The tier this solution was proven on
- Company-wide endpoint
- Batch contract processing
- Highest reasoning depth
Starter Workstation — pilot ready
Runs compact 7B–14B private models (quantized) for single-user pilots and skill testing.
Pre-built to this GPU class — pick a current model from Newegg’s AI hardware hub.
Team Workstation — the sweet spot
Runs 14B–32B private models comfortably — the same class our end-to-end legal verification ran on. One box serves a whole department.
Pre-built to this GPU class — pick a current model from Newegg’s AI hardware hub.
Max Workstation — company scale
Runs 70B-class flagship models with high concurrency — a company-wide private endpoint and batch contract processing.
Pre-built to this GPU class — pick a current model from Newegg’s AI hardware hub.
GPU names indicate hardware classes, not specific SKUs — see the live ABS lineup for current-generation models. Everyday users need no GPU at all; they connect to the shared endpoint.
Built for everyone who touches a legal question
Procurement, product, marketing, engineering — and the legal team itself.
Contract triage in minutes
GREEN / YELLOW / RED verdicts against your own playbook — deal-breakers flagged, redlines drafted, escalation routed.
Compliance answers with receipts
CCPA, COPPA, FTC shipping rules — answered from the actual regulation text with clickable citations.
Open-source guardrails
“Can we ship with this license?” answered from official SPDX texts before the code ships.
Risk flags before you sign
Auto-renewal traps, one-way indemnity, unlimited liability, hidden IP assignments — checked on every document.
Discipline built in
Research for attorney review — it pauses before legal consequences and routes decisions to your lawyers.
Your knowledge, your control
Every entry reviewed by a named person, version-tracked, and auto-invalidated when its source changes.
Two steps for users. One server for the company.
For every user
- Get
AI-Solution-Legal-Setup-v1.0.zipfrom the internal distribution point, unzip, openclient, double-click install.bat. - In Open Cowork: Settings → Skills → Refresh Skills.
Then run each skill’s one-time setup interview, e.g. Run commercial-legal:cold-start-interview.
For the solution admin
- Deploy the knowledge-base service from the package’s
serverfolder — runbook included. - Connect your private model endpoint and load company legal documents.
- Approve knowledge via the named-reviewer workflow; grant access by group.
Frequently asked
Is our data safe?
By architecture, not by promise: the model, the knowledge base, and the search index all run inside your network. No contract or question ever reaches an external AI service — and retrieval only uses knowledge a named reviewer approved.
Which AI model does it use?
Your choice. Any OpenAI-compatible private model works — Qwen, Llama, Mistral, DeepSeek, Gemma and others. The verified reference is the 32B balanced class.
Is this legal advice?
No — every output is research material for attorney review. The assistant labels it as such, pauses before consequential actions, and routes decisions to your legal team.
What if a source document changes?
Every file is tracked by checksum. A change automatically pulls its derived knowledge from the index and back into the review queue — stale knowledge can’t linger.