Agent Mode
Agent is the AI Assistant's most capable mode — the one to reach for when you need an answer you can put your name on. Instead of replying in a single pass, Agent runs a real investigation: it plans the work, reads the right documents, searches case law and the web when outside authority is needed, checks its citations as it goes, and writes a structured legal memo. You watch the whole thing happen, and every citation in the result is something you can click and verify.
What an investigation looks like
When you submit a question in Agent mode, the Show Thinking panel opens and the investigation unfolds in front of you.
Key Issues
Agent first frames the question into Key Issues — the specific legal questions it intends to answer. This is the plan, in your terms, before any work starts.
Leads run in parallel
Each lead is one line of inquiry — a case to validate, a document section to read, a statute to pull. Leads run several at a time, each is an expandable card you can open to see what it found and how long it took, and they roll up into facts as they finish.
Facts accumulate
As it works, Agent extracts the facts that matter — parties, dates, dollar amounts, cause numbers, and the like — into a running list, and flags where a document was unreadable or a fact is missing. The investigation builds on these facts rather than re-reading everything each step.
The memo is written
Once Agent has enough, it writes a structured legal memorandum with inline citations. Each citation is anchored to the exact source it came from.
Citations you can check
Agent is built so you never have to take a citation on faith.
- Inline and clickable. Hover a footnote in the memo to see the citation; click it to open the Sources & Citations panel.
- Grouped by where they came from. The panel separates Case Law from Web sources, with court metadata, a search box, and a Copy All button.
- They link to the real thing. A case-law citation opens the original court opinion in a new tab; a document citation opens your source PDF at the cited page.
- Verified against source. As it works, Agent checks its case-law citations against the underlying authority — a step that happens during the investigation, so what reaches you has already been vetted. It is still AI: review holdings before you file.
Built for your matter — and your documents
Agent works from the case in front of you.
- Matter-aware by default. Inside a Matter, Agent already has every document in that matter's library — no re-uploading. Ask across the whole set ("what are the indemnification obligations across all the vendor contracts?") and it reads selectively to answer.
- Scales with the matter. Because Agent decides which documents to read rather than loading everything at once, it handles a 500-document matter as comfortably as a handful.
- It reads like a lawyer would. Agent recognizes the kind of work you're doing — contract review, litigation analysis, due diligence, a regulatory question, and the like — and shapes the investigation to fit, prioritizing the documents and issues that matter for that work.
- Faster the second time. Follow-up questions in the same matter reuse what Agent already found, so they come back faster than the first.
Working with the answer
- Download it. Export any reply as PDF or Word (DOCX).
- Keep it. Use Save this output to Notes in My Desk to file an answer alongside the matter for later.
- Copy sources. Copy All in the Sources & Citations panel pulls every reference at once.
When to use Agent
- Analyzing matter documents — "summarize the contract risks in this MSA."
- Validating case law — "pull and verify the controlling cases on X."
- Any question that combines your matter with outside legal authority.
- Anything you intend to hand to a partner, a client, or a court.
When not to use Agent
- A quick clarifying question or a one-line answer — Everyday is faster and just as accurate for those.
- Drafting a clause from a template, rewording a paragraph, or tightening prose — Everyday is built for fast text work.
- A general topic survey with no case-law dimension — Deep Research is the cleaner fit for memo-style coverage of statutes, regulations, and secondary sources.
Using Agent with the other modes
Agent works well alongside the other two modes in a single session:
- Triage in Everyday, then escalate. Start with Everyday to orient yourself or chase a fast question, and switch to Agent for the part where the citations have to hold.
- Deep Research for the landscape, Agent for the cases. Use Deep Research to map a topic's statutes, regulations, and secondary sources, then switch to Agent to pull, verify, and analyze the specific cases that drive your answer.
- Agent for the memo, Everyday for the cover note. Let Agent produce the cited memo, then drop into Everyday to draft the email, client update, or summary that goes with it.
Switching modes mid-conversation does not rewrite earlier answers — only the next message you send is handled by the new mode. See Switching Modes & Sessions.
Example prompts
- "Review the attached Purchase Agreement for enforceability risks under Texas law and cite the controlling cases."
- "Extract every case citation from this matter's documents and validate them."
- "Build a timeline of the payment disputes in this matter with supporting document citations."
For a quick answer, a fast draft, or rewording something you pasted in, Everyday is faster. For a single comprehensive research write-up, see Deep Research. See AI Modes for the full comparison.