Agent Mode limitations
See in FrenchWhat Agent Mode cannot do, why, and how to get the same outcome safely—Nuix search, bulk scan summaries, internet access, and more.
Agent Mode limitations
Agent Mode is built for consistent, auditable review: a clear prompt, an explicit document set, and tools that respect case boundaries. Users often ask for shortcuts that would bypass those safeguards. This page explains what the assistant cannot do, why, and what to do instead.
If you use Agent Mode in chat, the assistant should answer in line with this page when these topics come up.
Asking the agent to run a Nuix search or pick documents for you
Users may ask the assistant to search the case, find every email from X, pull documents about topic Y, or run a query in Nuix on their behalf.
Why Claira limits this. Bulk review is designed around a single, user-approved prompt applied to an explicit document set you have already chosen. Letting the assistant run ad hoc searches or silently define the item set could process matter data outside a scope you have reviewed, and Nuix integration constraints mean the assistant is not a substitute for your search and selection workflow.
What to do instead.
- Run the search yourself in Nuix Discover (or your case tool).
- Review the hit list and refine the query or filters until the set is what you intend.
- Select the items you want (for example, select all in the current view).
- Start a bulk scan in Claira on that selection with your approved prompt.
Suggested reply (for the assistant). I cannot run Nuix searches or choose the document set for you—that stays with you so every bulk run uses an explicit selection and a prompt you have approved. Run your search in Nuix, select the items, then start a bulk scan on that selection.
Summarizing or analyzing across many bulk scan results in chat
Users may ask to summarize all bulk scan outputs, aggregate answers across the case, search my Nuix fields for what the model wrote, or analyze thousands of prior results inside Agent Mode.
Why Claira limits this. Claira does not maintain a separate, assistant-queryable warehouse of every document body paired with every AI answer. For privacy, privilege, and security, the assistant also cannot freely search or mine arbitrary Nuix fields for those outputs as a general database.
What to do instead.
- Export bulk results (for example to a spreadsheet) from Claira or your agreed export path.
- Re-import the columns you need, or paste representative rows or summaries into Agent Mode so the assistant can work on the text you supply.
- For large-scale analytics, use offline workflows or another trusted tool under your firm’s policies.
Suggested reply (for the assistant). I do not have a queryable archive of every scan answer tied to every item, and I cannot search Nuix fields broadly for those outputs. Export results to a spreadsheet and share the part you want analyzed, paste summaries here, or analyze offline with a tool your team approves.
Live internet access, web search, or opening URLs
Users may ask to Google something, look up a statute, fetch a web page, or use live web search while working in the same session as case documents.
Why Claira limits this. Workflows may include sensitive, privileged, or otherwise protected matter material. Combining open-ended network access with that context increases accidental disclosure risk (for example through prompts, logs, or vendor-side processing). Agent Mode therefore uses offline grounding: the model’s training knowledge, shipped Claira help, and approved in-app tools—not live browsing or URL fetching.
Suggested reply (for the assistant). Agent Mode does not use live web search or fetch URLs, to keep network access separate from matter context. I can use general knowledge and Claira’s help content and tools; for current or source-specific web research, do that outside this chat under your security policies.
Reading arbitrary documents or “browsing the case” from chat
Users may ask the assistant to open document 12345, read every attachment, or quote text from items they have not made active without running a scan.
Why Claira limits this. Outside explicit reference to the current active document (and tool-backed summaries such as bulk task history), the assistant does not have a general browse-the-case view of full item text. That boundary keeps chat aligned with explicit scope (single-document flow or bulk scan).
What to do instead. Open the item in the single-review workflow, or include it in a selection and run a bulk scan with the prompt you need.
Suggested reply (for the assistant). I cannot pull full text for arbitrary items on demand. Make the document active for a single-review question, or select the set and run a bulk scan so the scope and prompt are explicit.
Operating Nuix Discover or other case systems for you
Users may expect the assistant to click through Nuix, apply tags, run productions, or change case settings on their behalf.
Why Claira limits this. The assistant guides and explains; it does not drive Nuix Discover or other case systems as the logged-in user. Those actions stay under your control for auditability and permissions.
Suggested reply (for the assistant). I cannot operate Nuix or change case data for you—I can walk through the steps. Run searches, tagging, and exports yourself in Discover with your normal permissions.
Related
- Agent Mode — full capability overview.
- Bulk Scan — running scans at scale on a defined set.
Was this page helpful?