Four of roughly twenty capabilities in production across ~42 surfaces — each one accountable, cited and reviewable. A representative set:
Upload a custodian statement. The platform extracts 22 standardised columns, enriches identifiers, screens for sanctions and sub-investment-grade exposure, and generates a brand-themed client deliverable. Hours of analyst work, in minutes.
KYC narratives drafted from structured client data and source documents, with citations back to source.
A daily news ingest, scored and surfaced as a sentiment signal inside the investment dashboard.
Assisted parsing for the long tail of statement formats — surfacing breaks before they reach the client.
Roughly twenty capabilities across ~42 surfaces, priced on value — not tokens. Every call routes through one governed gateway with per-tenant policy, confidence floors and audit-hashed logging.
One gateway over the model providers you approve. Provision, rotate or revoke a key; usage and spend metered by model and day; per-tenant AI governance enforced at the gateway.
Data-residency-locked routing, with a dedicated on-premises lane for sensitive and restricted data. Sold as control, not savings — for firms that must keep certain data inside their own walls.
Absent these, Connected is the default.
AI analyst over portfolios, a read-only family-scoped portal assistant, risk-lab Q&A and a daily risk briefing.
Passport/ID, custodian-statement, fund and private-asset extraction; LP look-through; a GP-document pipeline with review queues.
Health-check summaries, proposal narratives, factsheet and portfolio commentary, IC and legal memos — each validated before a client surface.
Adverse-media classification, screening adjudication (PII-minimised), regulatory obligation extraction and corpus PII scanning.
A morning brief for RMs, per-deal briefs, free-text activity structuring, support-ticket triage and a governed Support Autopilot in shadow evaluation.
KB embeddings with vector search, redaction-aware embeddings, RAG grounding with a confidence floor, and an audit-hashed AI-chat trail.
Every generated narrative links back to the source document it was drawn from.
AI reads only what the acting user is entitled to see — the same governed spine, the same rules.
Output is a draft for review, not an action. A person signs before anything reaches a client.
Why it survives a model-risk review: no autonomous approvals, payments or trades; models evaluated before deployment and on change; bounded retention; low-confidence outputs routed to human review; a per-tenant AI governance policy enforced at the gateway.