Permission-aware
Field-level ACLs and record visibility apply to every AI read — the copilot literally cannot see what you can't.
Generic AI can draft an email. It can't tell you your DPI, watch a covenant, or file a quarter of financials — because it can't see your data, and you shouldn't want it to. mgmnt.work puts the AI inside the system of record: permission-aware, approval-gated, and audited like any other member of the team.
The copilot doesn't guess — it resolves. Filter, rank and aggregate run through the same query engine as every dashboard, so the AI's IRR is your IRR. Then the answer composes itself from the app's own building blocks.
Astra III has returned $1.28B (DPI 0.73×) — ahead of the 0.65× plan for this point in the fund's life. Two exits in diligence would lift DPI to ~0.9× by year-end.
Agents work with the same tools your team uses — reading records, running queries, drafting changes. The difference between reading and writing is a hard line: reads run instantly, writes wait for a human.
The builders turn unstructured input — a CIM, a cap table, a prompt — into drafts of the real thing, resolved against your live data and held for review before anything is committed.
Upload a board pack, a management report or a spreadsheet to any company's financials tab. Extraction streams live into a metric × period grid, and nothing lands without your review.
| Q1 | Q2 | ||
|---|---|---|---|
| Revenue | 48 | 52 | p.12 |
| Cost of sales → COGS | (17) | (18) | mapped |
| EBITDA | 8 | 10 | p.14 |
Put an AI summary on a record, a dashboard or inside a document template, written to your own prompt. When the underlying data changes, the summary knows it's stale and refreshes on the next view — never a stampede of background calls.
Helix delivered another quarter ahead of plan: revenue +11% QoQ with EBITDA margin at 20.3%, and the Q2 mark of $540M (3.0× MOIC) cleared IC on June 14. Watch item: payer concentration remains above the 35% covenant threshold.
Refreshed 2h ago · valuation changed — will refresh on next viewRegenerateField-level ACLs and record visibility apply to every AI read — the copilot literally cannot see what you can't.
AI writes are proposals until a human approves — with base-version guards so nothing stale ever applies.
Every AI run, answer and applied change lands in the same immutable audit trail as human activity.
Guardrails keep the AI on your workspace's data and workflows — it politely declines to be a general chatbot.
CIMs become linked records in minutes; pipeline data keeps itself current while your analysts do the thinking.
⌘K on live data replaces the "let me check with the team" email — and the tear sheet drafts itself from last quarter's.
Board packs file their own financials with page-level provenance; imports dedupe against your metric catalog.
Proposals, approvals and a complete audit trail mean the AI's work is as inspectable as anyone else's.
Bring a CIM and last quarter's board pack. In 30 minutes we'll turn one into linked records and the other into filed financials — in front of you, with every approval gate on.