Nutshell AI FAQs
4 min read
Common questions about Nutshell AI — the document-summarisation feature inside Squirrel. Each section is a question with a short, practical answer.
Does Nutshell change how Squirrel archives files?
No. Nutshell runs after Squirrel's normal archive process and only affects what is left behind in the SharePoint stub. Archive scheduling, lifecycle policy, restore behaviour, and where the archived file physically lives are all unchanged. See How Nutshell works for the pipeline detail.
Can I turn Nutshell on for archives that were made before I enabled it?
Yes. Retrospective summarisation queues previously archived documents for bulk summarisation, so historical archives become just as discoverable as anything archived going forward. Watch the queue drain on the Active AI Summarization Jobs page.
Does Microsoft Copilot actually use the Nutshell summary?
Yes — indirectly. Copilot queries the same SharePoint Search index that any other Microsoft 365 search relies on. When Nutshell writes a summary into the stub, SharePoint Search indexes that content, and Copilot then reads it when answering questions. From Copilot's perspective, the stub looks like an ordinary document with content that matches the query.
Which mode should I start with?
Standard for most tenants. It reads the start, middle, and end of the document — enough coverage for a useful summary without the throughput cost of Detailed. Move to Detailed selectively when a subset of your content (contracts, policies, records) needs full-fidelity summaries. See Summarisation modes for the trade-offs and per-worker throughput.
Which temperature is safest for archival content?
0.1. Low temperature keeps the summary close to source wording and consistent across runs — which is what you want when the summary is often the only representation of a document that a search or Copilot query will ever see. See Temperature settings for the fuller ranges.
How is Nutshell licensed?
Per worker. A worker is one GPU-backed processing engine, and throughput is reported per worker. Scaling capacity means adding workers. Contact sales@smikar.com for pricing and worker-count sizing.
Where does the summary content live?
Two places, both in your own Azure subscription:
- Inside the SharePoint stub (this is what SharePoint Search indexes and what Copilot reads).
- As a companion copy alongside the archived original in Azure Blob Storage.
Nothing about Nutshell requires customer data to leave your tenant. See Stub files: before and after.
Can I stop names and financial values from appearing in the summary?
Yes. Redaction has toggles for personal names, financial amounts, contact details, and dates / timestamps. Redaction is applied during summary generation, so the redacted values never reach the SharePoint index in the first place.
What happens if the file type is not on the supported list?
Squirrel still archives it. Nutshell just skips summarisation for that file — the SharePoint stub will not carry an embedded summary, and the file falls back to filename-and-site-only discoverability (the same as archiving without Nutshell). See Supported file types for the current list.
Does restore behaviour change when Nutshell is on?
No. A user clicking a stub still triggers Squirrel's normal restore flow. The summary is a discoverability aid; the restore path is unchanged.
Can I change mode or temperature later without breaking existing summaries?
Yes. Mode and temperature apply to new summary writes. Existing summaries stay as they were until you re-run summarisation over them using retrospective summarisation. This makes it safe to try a new setting on a small sample before rolling it tenant-wide.
How long does a bulk pass take?
Depends on mode, worker count, and total file volume. As a rough guide, a single Standard-mode worker processes ~2,400 files per hour. See Summarisation modes for the per-mode throughput and how scaling adds capacity.
Need help? support@smikar.com.