SOURCES
Comes in from your tools
Outlook, Lark and Google Drive sync into your knowledge base, with document-level permissions.
OUTLOOK · LARK · GOOGLE DRIVE
Files are redacted locally and only reach models through the AWS zero-data-retention (ZDR) channel. Your team and AI work side by side on pages, sheets, slides and meetings — every figure traceable, every external action approved.
Example run
How data moves
Data comes in from the tools you already use, is redacted before any model sees it, gets worked on by your team and AI in one Space, and leaves only with an approval.
SOURCES
Outlook, Lark and Google Drive sync into your knowledge base, with document-level permissions.
OUTLOOK · LARK · GOOGLE DRIVE
REDACT · ZDR
Sensitive fields are redacted locally inside your tenant; model calls go only through the AWS zero-data-retention channel.
LOCAL REDACTION · AWS ZDR
SPACE
Pages, sheets, slides and meetings live in one Space — your team and AI edit the same file.
PAGE · SHEET · SLIDES · MEETING
SIGN-OFF
Every figure links back to its source; anything sent outside waits for a human approval and is logged.
CITATIONS · APPROVAL · AUDIT
Every new connection goes through the same permissions and approvals.
Core use case
PER DEAL · IC
Investment committee memo with redlines from prior memos, DD checklist auto-populated from data room.
DOCX · A4 PDF
QUARTERLY · CLIENT
Per-client performance, attribution, narrative summary. PII redaction on all client identifiers.
Branded PDF
WEEKLY · EQUITY
Single-name deep dive: financials, valuation, catalysts, ESG flags, model debates. Slides + one-pager.
PPT · PDF · DOCX
DAILY · MACRO
Markets open, macro events, sector rotation, internal flags. PDF + Lark posting + email blast.
A4 · PDF · Lark
Templates ship with your branding. Mock data shown — real client data never leaves the tenant.
Governance is the product
Where data and models live
Why iMangrove
The team has architected, secured, and reconciled middle-and-back-office systems running multi-billion-dollar AUM in production. The same hands now design how AI plugs safely into your finance stack.
$10B+ AUM SYSTEMS
AUDIT-READY
PRODUCTION
BRIDGE
What it looks like in practice
The real moat in financial AI is trusted data, controlled execution, audit trail, and shipped deliverables. We've operated under that bar for years. iMangrove brings that DNA to your team's AI knowledge base and research workflows.
What changes during the pilot
Fixed scope. Fixed deliverables. No revenue promises — just range, acceptance criteria, and a signed artifact at the end.
| Stage | What you see | What we do |
|---|---|---|
| 01 · Discovery | One pilot workflow selected | Data source + permission assessment |
| 02 · Build | First Daily Macro Brief or research pack draft | Connector wiring + template fitting |
| 03 · Go-live | First signed-off delivery published | Approval flow + audit trail wired in |
| 04 · Scale | Expand to 3–5 workflows across the team | Operating metrics + governance policy |
Honest answers to the awkward questions
Tenant-isolated runtime. Self-hosted on your VPS or VPC. Field-level encryption with AES-256, keys generated and rotated by AWS KMS (or your own KMS). PII redacted locally before any model call; models are reached only through the AWS zero-data-retention (ZDR) channel. Audit log on every read.
No. Sensitive fields are redacted locally inside your tenant first, and model calls go only through the AWS zero-data-retention (ZDR) channel — prompts and outputs are not retained and not used for training.
Every figure is bound to a DataSlot — source, timestamp, confidence. Done Gate halts a workflow before completion. Multi-model cross-check surfaces conflicts. Provider chain with stale-flag fallback. Figures without a source are flagged for review.
Yes. Docker Compose deployment on your VPS, no Vercel dependency. Local Postgres, local object storage (or S3 if you prefer). Cron is OS-level crontab. KMS-managed keys optional.
iMangrove is the workflow runtime that sits on top of them, not a replacement. We connect market data, internal knowledge, models, approvals, and delivery into one auditable run — none of those tools does that end-to-end.
One workflow (for example an IC memo or DD pack from your data room) with fixed scope, signed-off deliverable, full audit trail, and a clear handoff path to your team. No revenue promises — just range and acceptance criteria.