Agent ModeCopy Anchor Link

Use Agent Mode when you need secure, general-purpose AI assistance inside Claira that is not tied to document coding templates.

What it doesCopy Anchor Link

• Creates persistent conversation threads per case - save as many threads as you need on every plan, and delete the ones you no longer need to keep your list tidy.
• Shares your chats with your case team by default so everyone can see the agent work done in the case: teammates can read (but not change) a shared chat, and you can switch any chat to private at any time. See Sharing, pins, and branching.
• Lets you reopen previous threads and continue where you left off.
• Titles each thread for you: moments after you send your first question, the thread's entry in the sidebar updates from the raw question text to a short AI-generated summary (about 5-7 words), written in the language of your question.
• Streams answers in real time from the configured model.
• Shows conversation history in a dedicated left sidebar for quick thread switching.
• Renders the Agent's responses with rich formatting (headings, lists, links, and code blocks).
• Gives the Agent read access (for the active case) to bulk scan task history, saved prompt history, and billing/plan summaries so it can explain status accurately.
• Knows Claira's own feature set: ask what any feature does (for example, "what are Insights?" or "what is a bulk scan?") and the Agent explains it in plain language first, then offers to find the full help articles for you.
• Field context (system only): Claira keeps an internal running log of how each connected Nuix field was used in bulk scan runs (which task, what role, who started it). This is not shown in field settings; the Agent can read it through tools to suggest the right field for a workflow.
• Can update a draft Case Context: the one you opened the chat about with Revise with Agent, one you name, or otherwise the case's default context. It reads the whole context first, and it can mark Missing information questions as answered. In Ask before running it waits for you to confirm in the conversation; in Give permissions it saves the edits you asked for without that step (see Ask before running or give permissions). An approved context cannot be changed. See Revising with the Agent.
• May offer a View plans button when you ask about capacity or hit plan limits; it opens the Settings plan manager, where all plan changes are made manually.
• May show a structured question card with tappable options when a fast clarification is needed before continuing (including Skip and Something else when applicable).
• Can format a ready-to-paste AI Review system prompt; use Insert into AI Review when that control appears. Clicking it takes you straight to the Prompt Lab with the prompt already in place, and the prompt box flashes green for a moment so you can see exactly where it landed.
• Can write a search term family for Nuix Discover when you ask for search terms or keywords: each term's query is checked against Nuix's search syntax, and the card's Download spreadsheet button saves an Excel file that imports into Nuix as it is. The card stays with the chat. See Generating Search Term Families.
• Can see and read the Insights already generated in the case: ask "what insights do we have?", or ask about a specific one by its IN reference and Claira summarises what it found. It reads the same insights your teammates created, and tells you when it has only read part of a long one.
• Can read the early case assessment of the case: ask "how many documents are in this case?", "who are the custodians?" or "where should I start?" and Claira answers from the assessment's figures and overview, always naming the date it was run. It knows whether this case has one without looking it up, lists the earlier ones so you can compare two runs, and offers a button that opens one here. It cannot run an assessment itself: the button brings you to the screen where you press Run assessment, and in a case with no assessment yet, one of the opening suggestions offers exactly that. See Early case assessment.
• Can set up an Insight with you and offer it on a card. Claira gathers the source (a finished bulk scan, or a field across the documents you have selected) and the instruction, then shows a Generate insight button. In Ask before running, nothing is generated until you press it; in Give permissions, the card counts down for 5 seconds and then generates unless you stop it. The same card handles revising an insight you already have.
• For an Insight over the documents you have selected, the card's header shows how many documents are selected, and Read AI results from names the field whose AI results the Insight will read. To read a different field, ask Claira in the chat before you press Generate insight.
• Can rename an insight, or regenerate its title, for free.
• Can read and add comments on bulk tasks and saved prompts: ask what the team noted on task #014, or ask Claira to note something on a task or a prompt. A comment it adds is saved in your name and marked Added by the Agent, and you can delete it like any comment of yours. It cannot edit or delete comments. See Comments on a task.
• Leaves a note on every bulk task it starts: why it started the task, what it is for and what the task does. The scan card shows the note under Comment on the task before the scan starts, and it becomes the task's first comment, in your name.
• Can offer to make a connected field this case's Log Field or Prompt Log field, on a card you approve with Set role. It cannot connect an existing Discover field for you. See Giving a field a logging role in a conversation.
• Can start a chat-native bulk scan: after your explicit confirmations (run intent, exact prompt, and document set confirmation) in Ask before running, or after a 5-second countdown you can stop in Give permissions.
• Lets you choose, for each chat, whether the Agent asks before running bulk scans, Insights and Case Context edits or whether you give it permissions to run them on its own. See Ask before running or give permissions.

Single vs bulk behaviorCopy Anchor Link

• The document you have open: it goes with every question you send while it's open, and it goes whole: the document itself (PDF, image, audio or video) as well as its extracted text. That lets Agent Mode answer about scanned pages, handwriting, stamps or recordings that have little or no extracted text. You don't have to word the question a particular way. Files larger than 15 MB are read from their extracted text only. On a Claira Private case, only the extracted text goes, because your own model reads text: the document's own file is never sent. The document shows as a chip in the message box saying what goes with your next message, and the chip's × keeps it out of your messages entirely. See The document you have open.
• Saving an answer to a field: hover an answer and click its save icon to write it to a field of the open document. When your case has a Prompt Log field, the save dialog also offers Also write the prompt to the Prompt Log, which records the question that produced the answer in the document's Prompt Log. It starts on the same remembered choice as the checkbox in the Prompt Log section below the response in Prompt Lab and Multi-Code, and Claira remembers what you pick when you click Save. Without a Prompt Log field on the case, the dialog shows no checkbox. See Recording the prompt in the Prompt Log. Claira adds the answer without replacing what the field holds: a field that holds one value and already has a different one is left unchanged, and Claira tells you so. See When the destination field already holds a value.
• Bulk flow in Ask before running (strict): before any chat-started bulk scan, Agent Mode confirms:
  • that you want to run a bulk scan,
  • how the prompt is produced (basic prompt, generated prompt, or custom prompt text),
  • for generated prompts, a dedicated approval to generate/refine prompt text before generation runs,
  • the exact prompt text it will run (for non-trivial requests, Agent Mode can reuse Claira Prompt Generator before confirmation),
  • the document scope (including the exact active selection count from your current selection).
• Bulk flow in Give permissions: the Agent writes the prompt, uses your current selection and chooses the destination field itself, tells you what will run, and the card starts after a 5-second countdown you can stop. See Ask before running or give permissions.
• The in-chat Bulk scan proposal card is a single form: a prompt text box (with Save and Quickstart to open the same prompt and template experience used elsewhere in Claira), a destination field picker (connected fields), and only Cancel and Approve and start at the bottom. The card's header shows how many documents are selected, for example 3 documents selected, and updates as you change the selection in Nuix. Quickstart is the supported way to generate a prompt or pick a template. A template you pick there counts as that template, even after you edit the wording on the card: when the bulk task finishes, it offers the template's Insight, as a task you start from the template yourself does. Scope is normally the active selection. When the scan is to run on all documents in the case, the header shows All documents instead, and a line under it gives the document count from the case's latest early case assessment and its date, or suggests running one when the case has none. The scan searches the case again when it starts, so documents loaded after that assessment are included.
• The destination field picker is the same one you use in Prompt Lab and Bulk Scanning, so a long field name is shortened and shows in full when you hover it. It lists only the fields Claira can write results to: a field still waiting on admin access appears once that access is granted.
• The card's destination field dropdown opens on the field you are already working in on this case, the one your last scan or Insight used, so in the common case there is nothing to choose. It opens empty when you have not picked a field yet, or when the field it was holding has since been disconnected: in Ask before running, Agent Mode never picks a field for you, because a bulk scan writes to every document in the selection and a destination you did not choose is expensive to undo. Approve and start waits until a field is chosen. In Give permissions, the Agent chooses the field and tells you which one it chose.
• Approving a card makes that destination the field you are working in, so your next scan in Prompt Lab or Bulk Scanning opens on it. Browsing the dropdown does not: only approving does. See Bulk Scan for how that memory behaves across the rest of Claira.
• After Approve and start (or once the countdown reaches zero in Give permissions), that bulk scan proposal card becomes read-only: its header shows Queued, and the card tells you that the bulk scan started and points you to Bulk Tasks for progress and outcomes.
• Every card the Agent shows states where it stands in the same place, a small label in its header: Queued for an approved bulk scan, Needs details, Generating and Generated for an Insight, Created or Awaiting admin access for a new field, the role's name (Log Field or Prompt Log) once a logging role is set, and Failed or Ended once a card's work is over.
• In Ask before running, bulk start is only executed after these confirmations are captured in structured confirmation cards and validated through a one-time confirmation token step. In Give permissions, the countdown on the card takes the place of these confirmations and of your click, and the scan then starts exactly as if you had pressed Approve and start.
• Once run intent and prompt text are approved, Agent Mode keeps those approvals and does not ask for them again unless you change them.
• Run intent is captured from a structured confirmation card and must reference that explicit confirmation before queueing.
• For document scope confirmation, Agent Mode shows only:
  • "Run on the current selection (N documents)."
  • "Run on all documents in the database."
  • "Something else."
• If you choose "Something else", Agent Mode pauses for reselection and then asks only for document scope again before starting.
• Option picks and typed custom answers are treated equivalently when they represent the same approval decision.
• During an active chat-started bulk scan, an inline status card appears in chat with live progress and a shortcut to Bulk Tasks.
• If the page refreshes or your connection drops temporarily, open Bulk Tasks (or wait in Agent Mode) to re-sync active task status from saved bulk task history.
• If Agent Mode detects an in-progress task after refresh/reconnect, it posts a recovery hint in chat pointing you to Bulk Tasks.
• If a retry submits the same confirmation token, Agent Mode reports that the run was already started and reuses the same request ID (no duplicate run is queued).
• If generated prompt text was produced without approved Case Context, Agent Mode explicitly tells you before asking for prompt approval.
• Multi-code runs are still launched from the main Claira UI.

Ask before running or give permissionsCopy Anchor Link

Each chat has a permission setting that decides whether the Agent waits for you before it starts work that spends tokens, writes to your documents or changes a Case Context. Choose it from the small control under the message box, to the left of the line reminding you that AI can make mistakes.

Ask before running is how Agent Mode has always worked: a bulk scan card waits for Approve and start, an Insight card waits for Generate insight, and a Case Context edit waits for you to say yes in the chat.

What happens in Give permissionsCopy Anchor Link

• The Agent sets up the scan. When you ask for work across your selected documents, or when scanning your selection is the best way to answer your question (for example "which of these are privileged?"), the Agent writes the prompt, uses your current selection and chooses the destination field itself. It tells you what will run, on how many selected documents, and into which field.
• How the destination field is chosen. The Agent picks a connected field Claira can write to that is neither your case's Prompt Log field nor its Log Field, preferring one whose name fits the task, and tells you which field it chose. If no field fits, it asks you.
• The card counts down. The bulk scan card shows Starting in 5s, counting down, with a Don't start button. At zero the scan starts exactly as if you had pressed Approve and start: it costs the same tokens (per successful document, by scan type), it is recorded in Prompt History, and it appears in Bulk Tasks.
• Changed your mind? Press Don't start. The card stays open as a normal card you can edit and approve yourself, and says "Not started. Approve the card when you're ready."
• Only your current selection. A run started without asking covers only the documents you have selected. With nothing selected, the Agent asks you to select documents first. A scan of every document in the case always waits for your click.
• Insights count down too. An Insight card counts down the same way and then generates. Insights are free, but generating one tags its source documents with the Insight's IN reference.
• Case Context edits are saved straight away. The Agent saves the draft edits you asked for without asking you to confirm first. It still reads the whole context and double-checks its changes, and an approved context still cannot be changed.

What always waits for your clickCopy Anchor Link

Creating a field, and giving a field the Prompt Log or Log Field role, always wait for you to press the button on the card, whichever setting the chat uses.

How the setting is chosenCopy Anchor Link

• The setting belongs to the chat. Changing it applies from your next message.
• A new chat starts from the setting you last chose in that case. Until you have made a choice there, new chats start from your organization's default.
• Teammates reading a chat you shared do not see the control, because they cannot send messages in it.
• If your organization's admin has turned Give permissions off, the control shows Ask before running with a lock, and its tooltip says "Your organization's admin has turned off Give permissions." If that happens while a card is counting down, the card does not start and asks for your click instead: "This chat now asks before running. Approve the card to start."
• Organization admins decide whether Give permissions is allowed, and where new chats start, in the Admin Portal. See Settings and Discover Portals.

Knowledge and privacyCopy Anchor Link

• Agent Mode can answer general-knowledge questions (for example definitions, history, and arts) using the model's built-in knowledge.
• Ambiguous terms are read the Claira way first: words like "scan", "fields", "tokens", or "insights" are treated as Claira features unless your question is clearly unrelated to the product or your review.
• For Claira-specific product facts, status, plans, and workflows, answers are grounded in Claira's shipped help content and first-party tools (billing, history, bulk scan and field context, Case Context).
• Replies are typically conversational first (a direct answer). For product topics, the Agent may add a short Learn more line with help links that come from shipped documentation (it does not browse the web for new pages).
• The Agent can also search the help site for you: accept when it offers to find out more (or simply ask for documentation) and it searches Claira's shipped help articles and replies with the most relevant links.
• When a request matches something the Agent can run in chat (for example setting up a bulk scan), it may offer the next step in text and can show a subtle in-chat hint alongside the streamed reply.
• Live web search (including Google Search grounding) is not used for Agent Mode, so responses about recent events may be out of date. See Agent Mode limitations for why and how to work around it.

Plan and token usageCopy Anchor Link

• Agent Mode is available on every plan, with unlimited saved threads per user per case.
• Messages in Agent Mode are free: chatting with Claira costs no tokens on any plan.
• An Insight you approve on a card is free as well. A bulk scan you approve on a card is charged exactly like the same bulk scan started from Bulk Scanning: per successful document, by scan type. Claira shows the cost on the card before you press the button. A bulk scan that starts after the countdown in Give permissions is charged the same way. See Plans and Billing.
• If the case does not have enough tokens for a bulk scan you approve, Claira shows what is blocking it and offers the fastest ways to continue.

Scope and privacyCopy Anchor Link

• Threads are scoped to your active case.
• Your list shows your own chats plus any chats teammates have shared with the case team. Chats marked Private to me are visible only to their owner. See Sharing, pins, and branching.
• Agent Mode does not depend on your case's token balance.

How to use itCopy Anchor Link

1. Open the top-right Modes menu (hamburger icon), then select Agent Mode.
2. Click New chat to start a thread.
3. In an empty conversation, use a suggested prompt chip or type your own question.
4. Send your message from the composer docked at the bottom of the conversation panel.
5. Select any saved thread in the left sidebar to continue it later.

Conversation layoutCopy Anchor Link

• Top bar: the row across the top of the chat window. It reads Agent Mode until the chat has a title of its own, and shows that title from then on; click the title to rename a chat you own. The chat's own controls sit at the right end of the same row, with New chat.
• Left sidebar: thread history with active-thread highlighting and quick delete, a Pinned section at the top, owner names on teammates' shared chats, and a filter button (the funnel icon) next to the search box, which turns green and shows a count while a filter is on. When history is collapsed on tablet- or desktop-width layouts, open it from the rounded tab on the left edge (hover near the edge on tablet width until you pin it open; click on desktop widths).
• Main panel: a light grey conversation area inside a solid white chat window with a clear outline, so the chat stands out from the page around it. It shows a centered empty state before the first message, then a scrolling transcript.
• Message alignment: your messages appear on the right, the Agent's responses on the left.
• Bottom composer: stays anchored to the bottom of the conversation panel for fast follow-ups. It starts one line tall, with the paperclip, your text, and the send button lined up in a row, and grows as you type. Above your text it shows the document you have open and any files waiting to go with your next message. Once your message runs past two lines, the paperclip and send button move to the bottom of the box, next to the line you're typing.
• Disclaimer: small italic text under the composer reminds you that AI can make mistakes and to verify important information.
• Permission setting: to the left of the disclaimer, a small control shows whether the chat will Ask before running or Give permissions, and lets you switch. See Ask before running or give permissions.
• Message length: one message can be up to 100,000 characters, which is a long document's worth of text. As you approach the limit the composer shows a character count, and past it says how much to remove and keeps Send unavailable until the message fits. Nothing is lost: your text stays in the box while you trim it, or you can split it across two messages.

TipsCopy Anchor Link

• Start each thread with a clear first question: the AI-generated title shown in the sidebar summarizes that question in a few words, so a clear question yields a meaningful title.
• Suggested starter chips highlight what Agent Mode can do in this case (Claira capabilities overview, explain the open document, bulk scan task history). You can also ask general-knowledge questions, plus Case Context, plans, or turning notes into an AI Review prompt.
• Keep one thread per workstream (for example: billing questions, bulk task status, prompt history).
• Delete threads you no longer need to keep your list focused.

LimitationsCopy Anchor Link

Agent Mode deliberately does not replace Nuix search-and-select, live web research, or a queryable archive of every bulk-scan answer. For common “why can’t you…?” requests (running case searches for you, summarizing entire bulk outputs in chat, browsing arbitrary documents, or driving Nuix for you), see Agent Mode limitations, including suggested replies the Agent can use for consistency.