Legal AI for contract review and research, where every answer points to the source text
Legal document AI uses language models to read contracts, filings and case law: extracting clauses, checking them against your playbook, answering questions and drafting summaries. Every output has to point to the exact source text, because the lawyer stays responsible for it and courts sanction fabricated citations. Sysvoy builds custom legal AI on retrieval and structured extraction, with citations, a review step for lawyers and testing against documents your team has already reviewed. We built midpage.ai, an AI research platform over case law for litigators and legal teams, end to end.
The path a question takes
Documents are parsed and split, embedded, searched, and the answer cites the passages it came from.
- Ingest
- Chunk
- Embed
- Retrieve
- Answer
Where legal AI projects stall
And what we do differently.
Citations that don't exist
General chatbots invent cases and quotes, and courts sanction lawyers who file them. Our tools answer only from retrieved text and link every claim to it.
Findings without evidence
A flag that says "non-standard liability clause" with no quote forces a full re-read. Ours show the clause, the page and the playbook rule it breaks.
Built without the lawyers
Engineers guessing at legal judgement build the wrong thing. We start from your playbook and past reviews, and lawyers set the pass mark.
What we build
Contract review against your playbook
Each clause compared with your preferred positions and fallbacks, with deviations flagged and suggested redlines drafted for a lawyer to accept or reject.
Clause and data extraction
Parties, dates, renewal terms, liability caps and governing law pulled from thousands of agreements into a spreadsheet or your contract system.
Due diligence review
A data room sorted by document type, with change-of-control, assignment and exclusivity clauses found and summarised for the deal team.
Case-law and research tools
Search and analysis over judgments where every proposition links to the passage that supports it, the approach behind midpage.ai.
Drafting from your precedents
First drafts built from your own templates and clause bank, not from the model's general knowledge.
How we deliver
From the first workshop to production. How we work
Start from your playbook
We turn your review guidelines and a set of past reviews into the rules and examples the system follows, and into the test it has to pass.
Extraction with evidence
Every field and finding carries the clause and page it came from. Where the model is unsure, it says so rather than guessing.
A lawyer signs off
The system drafts and flags; a lawyer accepts, edits or rejects, and those decisions feed the next round of improvements.
Measured on your documents
We report accuracy for each clause type against reviews your team has already done, and rerun the test after every change.
Custom legal AI compared with general chatbots and off-the-shelf legal AI platforms
| General chatbot | Off-the-shelf legal AI platform | Custom legal AI | |
|---|---|---|---|
| Answers tied to source text | Not reliably | Yes, within its own sources | Yes, every claim links to your documents |
| Fits your playbook and document types | Only through prompting | Partly, through configuration | Built around them |
| Where client data goes | Vendor's servers | Vendor's cloud | Your choice, including your own servers |
| Pricing | Per seat | Per user, often several hundred dollars a month | One-off build, plus usage at cost |
| Best for | Drafting help, never authority | General legal work across many matters | High-volume, repeatable review on your terms |
A good fit when
- Your team reviews NDAs, MSAs or leases at volume against a standard playbook.
- Off-the-shelf legal AI doesn't fit your document types, or your client data can't go to another vendor.
- You're a legal tech company building a product, and need engineers who have done it before.
What we build with
- Python
- TypeScript
- PDF and OCR layout parsing
- PostgreSQL
- Hybrid search
- Structured outputs
- OpenAI, Claude and open-weight models
- Labelled evaluation sets
- AWS and Google Cloud
Related work
midpage.ai
AI research platform that layers analysis over case law for litigators and legal teams.
midpage.ai ↗OtterSign
E-signature platform for a client's waiver and document workflow business.
ottersign.com ↗AI security platform
AI-driven security platform and SOC tooling for a European cybersecurity company.
Client under NDA
Planning a budget? Read our guides
- RAG development costWhat a retrieval-augmented generation system costs to build and run, from a studio that ships them. Price tiers, monthly infrastructure, build vs buy and two real examples.
- AI agent development costReal price ranges from a studio that builds AI agents, with the timelines, running costs and trade-offs behind them.
Common questions
Not answered here? Email contact@sysvoy.com.
How much does a custom legal AI tool cost?
Most projects for one document type, such as NDAs or leases reviewed against a playbook, cost $8,000–$20,000 and take 4–8 weeks. Tools that cover several document types, a data room or a research corpus usually cost $20,000–$45,000. Running costs for models and hosting are paid at cost. A 1–2 week discovery sprint turns this into a fixed quote.
How do you stop the AI from inventing citations?
The model only answers from passages retrieved from your documents or the corpus, every claim links to its source, and anything without a source is flagged rather than shown as fact. This matters: a public database maintained by Damien Charlotin had logged 2,266 court cases involving AI-hallucinated content by November 2026, and courts from New York to London have sanctioned lawyers for fake citations.
Is it safe to put client documents into an AI system?
It can be, with the right setup. ABA Formal Opinion 512 asks lawyers to assess the confidentiality risk before using a tool, and the SRA's August 2026 warning notice says confidential data should only go into approved tools with safeguards. We use model APIs that don't train on your data, can keep processing in your chosen region, or run open-weight models inside your own environment. This is general information, not legal advice; check your own professional obligations.
How is this different from Harvey, CoCounsel or Spellbook?
Those are ready-made platforms sold per user, and they're a good choice for general legal work. A custom tool is built around your playbook, document types and systems, can run where your data has to stay, and is a one-off build plus usage rather than a seat licence. We also build products for legal tech companies, including midpage.ai.
How accurate is AI contract review?
It depends on the clause type and the documents, so we measure it on yours. Before launch we test against agreements your team has already reviewed and report how often each clause type is found and classified correctly. Low-confidence results go to a lawyer, and we agree the accuracy bar with you before launch.
Do lawyers still need to review the output?
Yes. ABA Formal Opinion 512 and the SRA both make clear that lawyers stay responsible for work produced with AI and must verify it. Our tools are built for that: they show the source text next to every finding, so checking takes minutes rather than a re-read.
Can it run on our own servers?
Yes. We deploy to your cloud account or your own servers, and use open-weight models inside your environment when client data can't be sent to an outside provider.
Have something to build?
Tell us where things stand and what you need. A short message is enough to start.