Working with LegalClaw

Once LegalClaw is deployed, you work with it the way you'd work with a junior associate: you brief it in plain language, hand it the matter documents, and it produces review-ready work product. This page explains how to drive it day to day.

Open a chat session

How you reach the agent depends on how you deployed it:

• Desktop: open the chat UI in Hermes Desktop.
• Cloud (Docker): start an interactive session from your terminal.

docker exec -it legalclaw hermes chat

(If you added the host alias from the cloud setup, just type hermes chat.)

Brief the worker

LegalClaw does its best work when you give it the matter context up front. Before it starts, tell it:

• Client or matter — who you're acting for and on what.
• Jurisdiction — the governing law and forum that apply.
• Practice area — for example M&A, funds, employment, or data privacy (it can usually infer this, but saying it removes ambiguity).
• Source documents — the agreements, filings, correspondence, or diligence materials it should work from.

Then describe the task in your own words — "review this credit agreement markup and flag off-market changes" or "draft a merger agreement from this term sheet and diligence file." LegalClaw picks the matching skill from its 1,251-skill library, asks for any documents it needs, and runs it. To see everything it can do, ask it directly or run hermes skills list; for the library grouped by practice area, see the LegalClaw skills catalog.

What a skill gives it

Every LegalClaw skill encodes how to work one class of legal matter:

• The failure modes it corrects — the mistakes a generic draft or review typically makes on that task.
• The legal frameworks and market conventions that apply — the doctrines, statutes, and drafting norms the work must track.
• The shape of the deliverable — what a partner expects the memo, markup, or analysis to contain.
• A self-audit — checks it runs on its own output before handing it back.

What you get back

LegalClaw produces attorney-reviewable work product, not just chat replies — issue lists, clause-by-clause markups, drafts, comparison tables, and memos, with open items flagged in bracketed comments for your review. Each deliverable follows a consistent structure:

1. Matter, documents, and jurisdiction reviewed.
2. Legal frameworks and conventions applied.
3. Key findings and identified issues.
4. Open items and risks requiring attorney follow-up.
5. Recommended next actions.

Guardrails: an attorney stays in control

LegalClaw is built to assist a legal professional, not to replace one. By design it:

• Never invents statutes, case citations, regulatory provisions, filing deadlines, contract terms, or facts not present in the source documents — if support is missing, it says so.
• Separates observed facts, assumptions, identified issues, and open items, so you always know what's grounded and what's judgment.
• Escalates material uncertainty, missing documents, unusual terms, and judgment-heavy questions of law or strategy back to you.
• Treats everything as privileged and confidential — contracts, filings, correspondence, diligence materials, and client documents.

LegalClaw does not provide legal advice, does not create an attorney-client relationship, and does not replace attorney review, partner sign-off, or required filings review. Use it to do the heavy lifting; keep the professional judgment and final sign-off with the human.

Next, browse everything LegalClaw can do in the LegalClaw skills catalog.