AI Modes
The AI Assistant has three modes, in the selector below the input bar. They answer the same way you'd ask a colleague — the difference is how deeply Irys investigates and what the answer looks like when it comes back. You pick the depth; the platform picks the right machinery for that depth. Exactly one mode is active at a time, and Everyday is the default.
At a glance
| Everyday (default) | Deep Research | Agent | |
|---|---|---|---|
| Reach for it when | You want a fast answer, a draft, or a quick rewrite | You want one thorough, memo-quality write-up | The question turns on case law or your matter, and you'll stand behind the answer |
| Skip it when | The answer has to be cite-backed for a partner, client, or court | You need a quick reply, or the question hinges on case law | You just need a fast draft or off-the-cuff answer |
| How it works | A single, direct answer | One comprehensive pass across legal sources and your matter | A full investigation — plans, runs leads, checks citations, shows its thinking |
| Context & documents | Draws on your full matter context — everything available, answered directly | Pulls from your matter plus online legal sources and synthesizes across all of them | Reads selectively — opens only what each lead requires, so a 500-document matter is no harder than a handful |
| How citations come back | None — copy-ready text | Aggregated at the end — copy-ready | Inline, each linked to the source |
| Speed | Seconds | Slower — plan for it like a research memo | Scales with the question; you watch it work in real time |
How to choose
- Fast answer, draft, or quick question? → Everyday. Seconds. Clean text, no citations, nothing to wait for.
- Need solid coverage of a statute, regulation, or legal topic? → Deep Research. One comprehensive pass, citations gathered at the end — paste it straight into your work.
- The question turns on case law, or the answer goes to a client or court? → Agent. A full investigation with inline citations you can click and verify.
Not sure? Everyday is the quickest path to an answer. Move to Agent the moment the question touches case law or matter documents and the answer has to hold up.
Everyday
The default mode, and the right tool for most of the day. Ask a question, get a direct answer — draft a clause, summarize a document you paste in, reword something, or settle a quick point of law. It works from the full document context of your matter and stays fast even as a matter grows.
Citations. None. The answer comes back as clean, copy-ready text — ready to drop into an email, a draft, or a message.
Speed. Seconds. Comparable to a fast web search.
Skip Everyday when the answer has to be cite-backed, when the question turns on case law, or when the output is going to a client, a partner, or a court. Those are Agent questions. Full detail on the Everyday Mode page.
Deep Research
When a question deserves a real write-up, Deep Research makes one comprehensive pass — pulling from online legal resources and your matter and synthesizing across them into a structured, memo-quality report.
Citations. Aggregated at the end of the report. Sources are gathered into a reference section rather than threaded inline, so the body of the report stays clean and copy-ready.
Speed. Noticeably slower than Everyday — plan for it the way you'd plan for a research memo, not a quick reply. The wait is proportional to what you get.
Pair it with Cite Check. Because Deep Research gathers citations at the end rather than verifying each one as it writes, run Cite Check on the result to confirm every case is real and still good law before you rely on it. For case-law-heavy questions, Agent verifies citations against source as part of the investigation — see the full Deep Research page.
Skip Deep Research when you just need a quick answer (use Everyday) or when the question turns sharply on case-law verification (use Agent — its inline-cited memo is built for that).
Agent
The most capable mode — for the answers you put your name on. Agent plans an investigation, pursues several leads at once, reads the right documents, searches case law and the web, and writes a structured legal memorandum.
Citations. Inline and clickable. Hover a footnote to see the citation; click it to open the Sources & Citations panel. A case-law citation opens the original court opinion in a new tab; a document citation opens your PDF at the cited page. Agent verifies case-law citations against source as it works, so what reaches you has already been checked. It is still AI — review holdings before you file.
Speed. Scales with the question. A focused question may wrap up in under a minute; a complex, multi-document investigation may take several. Because the Thinking panel shows the investigation as it runs, you're following progress — not waiting blind.
Agent extracts the facts that matter into a running list, answers follow-ups faster by reusing what it already found, and reads selectively — a 500-document matter is no harder than a handful. Full detail on the Agent Mode page.
Skip Agent when you want a quick draft, a fast rewrite, or an off-the-cuff answer. Agent's depth is overkill for those — Everyday will be faster and just as useful.
Combining modes in one workflow
Mode is a per-question choice, so it's normal to move between them inside a single session. Common patterns:
- Triage in Everyday, dig in Agent. Start with Everyday to get oriented on a new matter or test a framing, then switch to Agent for the part that needs to hold up — the cite-backed analysis you'll actually use.
- Deep Research for the survey, Agent for the case law. Use Deep Research to lay out the legal landscape on a topic (statutes, regulations, secondary sources), then switch to Agent to validate and analyze the specific cases that drive the answer.
- Deep Research → Cite Check. Always run Cite Check on a Deep Research report before relying on its citations. Treat the two as a paired step.
- Agent for the memo, Everyday for the cover note. Once Agent has produced the memo, switch back to Everyday to draft the email or summary that goes with it.
Switching modes doesn't rewrite anything already in the chat — it changes how the next message is handled. If you want the same question answered with deeper investigation, switch the mode and re-ask.
Switching modes
Change mode any time from the selector below the input bar. The mode applies to the next message you send. Switching mid-conversation does not re-run earlier answers — a single chat can contain replies produced in different modes.
Mode persistence works on two levels:
- Per chat. Each existing chat reopens in the mode used for its last message — if your last reply in a chat was an Agent memo, that chat reopens in Agent.
- Globally. New chats start in whichever mode you last selected anywhere in the platform — not necessarily Everyday.
Everyday is the system default for every new account, but after you've selected Agent or Deep Research your last choice carries forward. Glance at the selector before you send, especially at the start of a session. See Switching Modes & Sessions for the full picture.
Working with any answer
Whatever mode you used, you can download the reply as PDF or Word (DOCX), Save this output to Notes in My Desk to keep it with the matter, or copy it straight out.