The best-known legal AI products are built for large firms with large budgets and a knowledge management team. A firm of five or twenty lawyers has different problems: intake calls that interrupt billable work, research that starts from scratch every time, and documents assembled by hand from the same templates. We build AI agents for those problems, inside the tools your firm already uses, with a lawyer approving anything that leaves the firm.
What is the best legal AI tool?
The best legal AI tool depends on the size of the firm and the job: research platforms such as CoCounsel or Harvey suit large firms with heavy research volume, while practice management suites such as Clio add AI features to the tools small firms already run. When the need is specific to your practice, a custom agent connected to your own templates and sources often fits better than another subscription.
| Option | Good for | Limits for a small firm |
|---|---|---|
| Enterprise legal AI platforms | Large-scale research and drafting | Price, onboarding, built for big-firm workflows |
| AI features in practice management software | Firms already on that software | Limited to what the vendor ships |
| General assistants (Claude, ChatGPT) with firm rules | Drafting, summarizing, first-pass review | No access to your matters or sources without integration |
| Custom agents (what we build) | Your intake, your templates, your sources | Needs a scoped project and maintenance |
What an AI agent can do in a small law firm
- Client intake. Answer the first contact on your website or by email, collect the facts with the questions your lawyers would ask, check for conflicts against your client list, and hand a structured summary to a lawyer.
- Research connected to official sources. Search statutes, regulations and case law through official databases, and return citations a lawyer can verify, rather than answers from memory.
- Document preparation. Fill your engagement letters, standard motions or contract templates from the matter data, and flag what needs a lawyer's judgment.
- Matter follow-up. Summarize new documents and emails on a matter and prepare deadline reminders for review.
What we have built: an AI connector to official legal databases
We built and host a connector (an MCP server) that gives Claude direct access to official French legal sources: legislation and codes, case law from the French supreme court's open database, the official company-announcement bulletin and the tax administration's published doctrine. It exposes nine search and read tools, protected by an access token, so a lawyer can ask Claude a question and get answers grounded in the official texts, with references.
To be clear about scope: this connector covers French law. It started from an open-source project, which we extended with two additional official sources and deployed as a hosted service. The same architecture plugs into the sources a US or UK firm relies on: public legal databases with an API, your research subscriptions where their terms allow it, and your firm's own document management system.
Can AI handle client intake for a law firm?
Yes, for the first steps: an AI agent can answer a prospective client, ask the intake questions, check basic eligibility and conflicts, and book a consultation. It should not give legal advice or accept a matter; a lawyer reviews every intake before anything is promised.
Is it safe to use AI with client data?
It can be, if the setup is designed for confidentiality: business-grade AI plans that do not train on your data, access limited to the matters a user may see, logs of what the AI read and did, and a lawyer reviewing outputs. Bar guidance in the US, such as the ABA's 2024 formal opinion on generative AI, puts the duty of confidentiality and supervision on the lawyer, so the system must make supervision easy.
How we design for it:
- Read-only access by default, write access only where needed, nothing sent to a client without approval.
- Per-user identity and an audit log of every action the agent takes.
- Data kept in the region you choose, and no client data used to train models.
- A written description of what the agent does alone and what it submits for review.
How a law firm project runs
- A 30-minute call to pick the first workflow: usually intake or document preparation.
- A discovery session with a lawyer and a staff member who does the work today.
- A written scope with approval points, and a fixed quote. What drives the price is in our AI agent cost guide.
- Build and test on anonymized or sample matters, then acceptance by the firm.
- Launch with monitoring and a monthly maintenance plan.
For the agent offer in general, see AI agent development; for connecting AI to your existing software, AI integration services.