Capabilities
Your administrator sets the firm's ceiling in Settings → AI Agent Access, and each person chooses what to allow within it when they connect. Permissions are per kind of data, and reading is separate from changing or acting:
| Data | Can read | Can change | Action |
|---|---|---|---|
| Clients | Yes | Yes | — |
| Matters | Yes | Yes | — |
| Time entries | Yes | Yes | — |
| Expenses | Yes | Yes | — |
| Fixed fees | Yes | Yes | — |
| Invoices | Yes | — | — |
| Account balances | Yes | — | — |
| Billing codes | Yes | — | — |
| Emails | — | — | Send reports |
Permission to change data is permission to change it at all, not permission to change anything without asking. A well-behaved assistant will tell you what it is about to do and wait for you to say yes.
For what an assistant cannot reach, see Limitations.
Send reports
Send reports lets an assistant email a report built from LeanLaw data, such as a weekly billable hours summary, to people in your firm. It can only send to LeanLaw users at your firm, never to clients or anyone outside it. The email comes from LeanLaw on the sender's behalf and replies go to them. A firm can send to at most 200 recipients a day, and each person can receive at most 10 report emails a day.
Skills
The connector gives an assistant access to your data. Skills tell it how to carry out a specific piece of your firm's workflow with that access. We publish ours at github.com/leanlaw/agent-skills, where you'll find the current list and what each skill does.
Use a skill when you do the same task regularly and want it done the same way each time, or when you want to run it as a scheduled task. For one-off questions you don't need one; just ask the assistant. Skills need the connector set up first, and they can only do what its permissions allow. The repository's README explains how to install them in Claude, Claude Code and Copilot.