Voice commands

The Paralegal agent listens to natural speech — there are no fixed phrases. This page documents the patterns that route to specific tools or shortcuts so you can learn the agent quickly. Anything you say outside these patterns the agent still understands; these are just the highest-confidence routes.

Inbox + Calendar

Say something like
What happens
Tool
"What's in my inbox from Sarah?"
Lists recent Outlook messages, optionally filtered by sender. Time-stamps appear as "2 hours ago" / "3 days ago".
list_recent_email
"Pull up Sarah's email about Acme."
Reads the full message body + attachments.
read_email_thread
"What's on the calendar Thursday?"
Lists upcoming Outlook events.
list_calendar
"Download that attachment into my library."
Pulls the attachment from Outlook into your Library; auto-attaches to the active matter.
download_attachment

Matters

You refer to matters by client name — never by ID. Say "the Acme matter" / "the Smith case" / "Tammy Bile's file" and the agent resolves it for you.

Say something like
What happens
Tool
"What matters do I have open?"
Lists active matters. Pass status="all" to include watching + closed.
list_matters
"Open the Acme matter."
Resolves the name via list_matters(search="Acme"), then loads the matter brief — client, counter-party, stage, deadlines, attached items, recent emails.
list_matters → read_matter
"Open a new matter for Acme on the NDA."
Drafts a create_matter card in the right pane. Approval gate before it lands in the database.
create_matter
"Advance this matter to discovery." / "Move us into redline."
Bumps the matter's stage. Practice-type triggers fire automatically — e.g. advancing a litigation matter to pleadings seeds "File complaint" + "Verify service of process" deadlines; closing a transactional matter archives open deadlines and flips status to watching.
advance_stage
"Pin this matter." / "Unpin the Acme one."
Toggles the attention flag. Pinned matters cluster at the top of the matters list.
pin_matter
"Close out the Acme matter." / "Park it for now."
Flips status to closed. Approval-gated.
archive_matter
"Change the response due to next Tuesday." / "Matter type is now NDA."
Edits a single field on the active matter (response_due / matter_type / posture / counter_party). Approval-gated.
set_matter_field
"Attach this redline to the matter." / "Pin that email to the file."
Links a vault item, library doc, chat, or email thread to the active matter.
attach_to_matter
"Detach the old draft from this matter."
Removes a linked item from the matter (the underlying doc is NOT deleted).
detach_item
"What's on my plate?"
Parallel-fetch: inbox + calendar + matters. Synthesizes a 3-sentence rundown.
shortcut
"Morning briefing."
Slightly longer — includes upcoming deadlines and addresses you by name (set in settings).
shortcut

Deadlines, time, notes

Every entry defaults to the active matter. Forward-relative dates ("next Friday", "in two weeks", "tomorrow") are resolved to absolute YYYY-MM-DD before the tool fires.

Say something like
What happens
Tool
"Set a deadline for next Friday — draft response brief."
Drafts a set_deadline card. Approval-gated by default; auto-fires when you've granted the deadline tool the "auto" policy in settings.
set_deadline
"Mark the discovery deadline done."
Flips a deadline's status to complete.
update_deadline_status
"Scrap that conflict-check deadline, we already did it."
Dismisses a deadline that no longer applies (separate from completion — used when the task became moot).
dismiss_deadline
"Log thirty minutes for the client call."
Adds a time entry. Auto-fires (no approval); rolls into the matter's hours total.
log_time
"Undo the last time entry — wrong matter."
Removes a time entry. Approval-gated.
delete_time_entry
"Note that the client wants to settle for under 50k."
Adds a matter note. Auto-fires.
create_matter_note
"Update the settlement note — they're now firm at 40k."
Revises an existing pinned note rather than creating a new one.
edit_matter_note

Document review

Say something like
What happens
Tool
"Find the Acme NDA."
Searches Library + Vault, picks best match by name. Use before any scan/redline so the agent has the right doc id.
find_document
"Run a quick scan on the Acme NDA."
Tabular review with 5 default columns (cap, indemnity, auto-renewal, termination, definitions). Findings stream into the transcript as cells complete. Accepts Library docs AND Vault items.
run_quick_scan
"Dive deeper on the indemnity cap."
After a quick scan, drills into one specific clause with deeper reasoning. Returns a markdown analysis the viewer can render.
dive_deeper
"Redline this against my standard playbook."
Runs a full redline. Uses your default playbook from settings if you don't name one. Track-changes .docx lands in your Library.
run_redline
"Compare the two versions of the MSA."
Clause-by-clause diff between two docs.
run_compare
"Show me that document."
Pops the most recently referenced doc into the in-session viewer. Works for PDF, DOCX, and MD vault notes.
display_document

Drafting

Say something like
What happens
Tool
"Draft me a response to Sarah's email about the indemnity carve-out."
Generates a draft, opens it in the viewer immediately. No approval gate — you'll see it right away.
draft_document
"Make it more formal" / "tighten paragraph 2" / "add a section on confidentiality."
Re-runs the LLM with your tweaks. Viewer auto-reloads with the new version.
revise_document
"Send this to john@acme.com."
Queues an email draft with the doc attached. Approval card lands in the right pane.
draft_email

Research

One unified search tool that fans out across case law, legislation, your vault, and the live web. The agent picks the scope from your phrasing; you don't have to name it.

Say something like
What happens
Tool
"Look up Daubert v. Merrell Dow." / "Case law on liquidated damages in Florida."
Searches CourtListener opinions (federal + state). Returns 3–5 cases with citation, court, year, and absolute URL.
research_search · case_law
"What does Fla. Stat. § 83.595 say?" / "Look up the FL statute on commercial leases." / "Pull up Title 17 of the US Code."
Reads the enacted statute code text from the state's official code site (with Cornell LII + Justia as fallback). Returns an excerpt + canonical citation format + source URL. Shares a 7-day cache with the chat surface.
lookup_statute
"Are there any pending Florida bills on AI disclosure?" / "Active legislation on data privacy in Texas."
Searches LegiScan for bills moving through state houses (different from enacted code — use lookup_statute for the latter).
research_search · statutes
"What's in my vault about Acme?" / "Do I have notes on indemnity caps?"
Hybrid semantic + keyword search over your vault items. Active-matter items get a relevance boost.
research_search · vault
"What's the market norm for indemnity caps in SaaS?" / "Recent news on the Acme CFO."
Live web search via the same LLM provider you're using — Anthropic web_search, OpenAI search-preview, xAI/Grok web_search, or Google Gemini grounding. Honors your BYOK key first; falls back to the server's key. No third-party search APIs.
research_search · web
"Find me everything on the new AI disclosure rules."
Fans out across case law, statutes, vault, AND web in parallel and returns the merged result set.
research_search · all
"Verify the citations in the brief."
Long-running citation check on a .docx; results land in Workspace › Citation Verifier.
run_citation_check

Context awareness

The agent watches what you're looking at — current page, active matter, focused row, filters, open modal — and resolves deictic references against that state. You almost never need to spell things out.

Say something like
What happens
Resolves via
"Add a deadline to this matter for next Friday."
"This" = whatever matter page you're on. No question asked.
screen context · matter_id
"Mark that deadline done."
"That" = the deadline row you just clicked or hovered. Tracked via focused_deadline_id.
screen context · focused_deadline_id
"Close out all of these." (on /matters/ with rows selected)
"All of these" = the matter ids in selected_matter_ids.
screen context · selected_matter_ids
"Where am I?"
Reads the current screen state aloud — page, matter, active tab, applied filters.
get_screen_context

Approval + control

Say something like
What happens
Effect
"Approve" / "send it" / "go ahead"
Fires the only pending action. Confidence ≥ 0.92 required (loud + clear).
approve
"Approve the email" / "approve the calendar invite"
Disambiguates when multiple pending actions are queued. Use "approve the first/second/third" too.
named approve
"Approve all"
Fires every pending action in the queue.
batch approve
"Discard" / "throw out" / "don't send"
Discards the pending action.
discard
"Cancel that" / "stop the redline"
Cancels the most recently started long-running job (quick scan / redline / compare / citation check).
cancel
"Take your time" / "no rush" / "be thorough"
Upgrades the NEXT tool call to the deep tier (Opus 4.5 by default) for stronger reasoning. Reverts after one call.
deep-mode (one-shot)
"Be quick" / "speed up"
Cancels a queued deep-mode upgrade if you changed your mind.
cancel deep-mode

Discoverability

Say something like
What happens
Effect
"What can you do?" / "Help me get started."
Agent speaks a short capability summary — under 25 seconds.
help
"What just happened?"
Recaps the last few tool calls + approvals from this session.
recent activity

How approvals work. The agent drafts emails, calendar holds, doc-shares, and new matters — but never sends until you voice-approve. Pending actions land in the right pane. Outbound goes through Microsoft Graph using your connected account. Per-action policy ("auto" / "ask" / "block") is configurable in settings.

Personalization. Set your Display name (first name) and Form of address (voice) in Settings → General. The agent uses them everywhere: "Morning, Counselor — three deadlines this week" in voice, "Hi Marcus," in drafted email salutations, and your name in the morning briefing, drafted memo signatures, and transcript labels. If only one field is set, the agent uses it for both purposes.

Session time tracking. Every voice session is logged with start + end timestamps under paralegal_audit_log. Each matter detail page shows the total minutes spent talking to the paralegal on that matter ("47 min · Across 3 sessions"). Closing the browser mid-call still records the duration via a keepalive ping.

BYOK. When you've added an Anthropic / OpenAI / xAI / Google API key in Account, the voice agent uses YOUR key for the realtime mint, drafting models, and web search — no server cost, no rate-limit contention. The agent falls back to the site's server key only when no BYOK is configured. Switch providers via the voice provider toggle in Settings → Voice.