The pipeline
The engine behind Lore™ . It splits multi-document bundles into separately filed records, classifies them against your taxonomy, tags emails as they land, brokers external sharing with role-based access and compliance gating, and pushes live updates back to the case — without slowing the case system down.
For the architecture and deployment topology, see the infrastructure page.
Classify & split
One upload of a 30-page mixed bundle returns N classified, separately filed documents — against your taxonomy, with confidence scores, with suggested file names.
Drop a 30-page mixed bundle into the case workspace and watch it become N separately filed documents, each with a suggested name, description, tag, and folder placement. The AI detects boundaries (letterheads, page resets, signatures); splits are validated before anything is written.
Classifies against your tag list, not a vendor-fixed schema. Bring your own categories; swapping them is a config change, not a model retrain.
A fast preview classification (~2 seconds) populates the UI immediately; deeper enrichment runs in the background and updates the row when it finishes.
Anything below threshold is flagged for human-in-the-loop review before it is written to the case. Auditable, configurable, never silent.
Generated from the document content, in your naming style. The investigator gets a properly named, properly described file the moment it lands.
Tag every artifact
Documents, emails, and notes share the same tag set. One filter chip surfaces every relevant artifact across record types.
Every inbound and outbound email is tagged against the case taxonomy in roughly one second, at roughly $0.0002 per email. Fire-and-forget, no async queue, no UI block.
Documents, emails, and notes share the same tag set. One filter chip surfaces every relevant artifact across record types. No reconciliation between vocabularies.
Tag suggestions appear as the adjuster writes and refine as the note grows. The classifier sees what the writer sees.
Storage
Files go straight from the browser to your document store. The case system holds metadata and the document's URL — never the bytes. Per-row attachment footprint stays at zero.
Files upload from the browser straight to your document store. They never pass through the case system's attachment storage.
Per-row attachment footprint stays at zero. Storage costs and SN attachment limits stop being a scaling cliff. ServiceNow holds metadata and audit; SharePoint holds the bytes.
Once a file lands in your document store, its location is written back to the case record. Clicking the row opens the file in place.
External sharing
Role-based access, BAA gating, auto-grant on party-add, auto-revoke on party-remove, time-bounded by default. Every grant and revoke is audited.
Adjuster, attorney, expert, claimant — each role sees only what their role permits. The role-access matrix is enforced at the API, not in the UI.
Parties without an active Business Associate Agreement cannot receive PHI-bearing documents. The API blocks the share and writes an audit event — no manual gate.
Adding an attorney or expert to a case grants them access to in-scope documents automatically. No second step, no missed shares.
Removing a party expires every active share they hold. Document permissions are revoked at the same moment; an audit event is written for each revocation.
Every share has a configurable expiration. SharePoint enforces it; nothing lingers past its window.
Live updates
When AI finishes enriching a document, every open Lore session for that case updates in place. Per-user routing — no cross-noise.
When AI finishes enriching a document, every open Lore session for that case updates without a refresh. Updates route per user — your tabs see your uploads, not your colleague's.
Long-running classification doesn't block the upload UI. The user gets a row immediately; enrichment finishes when it finishes and the row updates in place.
Configure & guarantee
Per-customer rules, two-backend deploy, server-side cache, confidence-routed human review, idempotent installs, IP-stamped responses.
Per-customer, per-document-type rules for default folder placement. New files land where your operators expect them; no manual triage on every upload.
Preferred path runs in your cloud tenant. For environments that can't reach the cloud directly, Gather AI runs entirely inside the case system. Same product, two postures.
Claim Summary and other expensive outputs cache at the function tier, shared across every client viewing the same case. Models run once per case-state, not once per viewer.
Low-confidence classifications surface for adjuster confirmation before they're written. The model never silently mislabels; the human is the last word when the AI isn't sure.
Every backend install script is safe to re-run. Re-deploys, partial failures, and rollbacks don't leave orphaned config.
Every API response carries X-Copyright and X-Vergence-Product headers. The product asserts ownership on every byte it serves.
Upload a multi-doc PDF in our synthetic case and watch the splits and classifications appear.