PAIN 01
Our 12 analysts are each on a different AI tool. None of that work compounds.
Knowledge doesn't aggregate. Workflows aren't reused. Team productivity never compounds.
iFinclawThe Agent Workspace for Financial Teams
Built for
Family OfficesEAM / OCIOAsset ManagementVC / PEFinancial Platforms
Run research, reporting, client service, and operational workflows — on trusted data, controlled execution, and audit-ready outputs.
Connecting market data · Polygon · Tiingo
Why Finclaw
A team forged inside finance — not retrofitted for it.
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.
DOMAIN EXPERIENCE
$10B+ AUM
- Multi-billion-dollar AUM scale · 10+ years finance back-office.
- Cross-asset coverage: cash, bonds, FX, derivatives, alts, structured products.
SECURITY MODEL
AUDIT-READY
- AES-256 field-level encryption · AWS KMS key generation & rotation.
- Multi-tenant physical isolation · PII redaction at ingestion.
EXECUTION DISCIPLINE
PRODUCTION
- Daily reconciliation across providers · Approval gates before external action.
- Append-only audit log · Replayable runs · Done Gate before completion.
INTEGRATION CAPABILITY
BRIDGE
- Private bank feeds · Email extraction · Event-driven cash-flow engines.
- VPS / VPC / Singapore AWS hosting · Read-only by default, encrypted IDs.
What it looks like in practice
AES-256AWS KMSMulti-tenantPhysical isolationField-level encryptionPII redactionAppend-only auditDaily reconciliationCross-asset coverageSingapore AWS
The real moat in financial AI is trusted data, controlled execution, audit trail, and shipped deliverables. We've operated under that bar for years. Finclaw brings that DNA to the AI agent layer.
What we hear in every first call
Problems your team won't say out loud — but lives with every week.
PAIN 02
Claude for writing, GPT for code, Gemini for long context — and Slack DMs to ask 'which one did you use?'
No unified entry. No cross-check. No way to compare model outputs side-by-side.
PAIN 03
Our research notes live in 4 Google Drives, 2 Notion workspaces, and a stack of PDFs nobody can search.
Internal knowledge isn't searchable. No permission tiers. No audit trail.
PAIN 04
Half the price quotes our AI returns are stale. The other half it just made up.
Single data source fails silently. No fallback chain. No timestamps. No source labels.
PAIN 05
We tried generating a 60-page sector report. Page 23 had a P/E ratio that doesn't exist on any planet.
Numbers come from model memory. No source citations. Can't send to clients.
PAIN 06
We can't put KYC documents into any AI. So our team retypes paragraphs and pretends.
No redaction tool. AI is stuck on low-sensitivity Q&A. Real work doesn't get the lift.
PAIN 07 · THE ONE NOBODY WANTS TO ADMIT
Last quarter someone almost pasted client tax docs into ChatGPT. We don't know how many times it already happened.
Sensitive content leaks. No DLP. No audit log. Bans push usage underground — they don't stop it.
PUBLIC RECORD
- 2023Samsung semiconductor source code leaked via ChatGPT
- 20236 Wall Street banks ban ChatGPT internally
- 2024EU AI Act high-risk classification for finance
Bans don't fix this. Governance does.
Your team isn't trying to leak data — they're trying to get work done.
Not just for leadership — for everyone touching the workflow
Find your scenario, see the answer in 30 seconds.
Five scenarios Finclaw is built to win. Each one spans analysts, operators, compliance, and leaders — pick the work that looks like yours.
Research & Analysis
Who's involved:Investment analystsFund analystsPortfolio managersResearch headsCIO
How do I trust a number that came out of a model?
- Every figure bound to a DataSlot with source and timestamp.
- Multi-model cross-check; conflicts surfaced, not hidden.
- Done Gate before completion; recovery from any failure.
Investment Mode · 3-Agent
- Resolve DataSlots14 / 14
- Cross-check Claude · GPT · Gemini3 / 3
- Done Gate · Awaiting CIOPending
SOURCE-LINKED · NUMBERS BOUND
Four capabilities, built for finance
Not just chat — the Agent layer that sits on top of your finance stack.
01 · FINANCIAL DATA
Financial data plane
Multi-provider quote chain with stale-flag fallback. Macro calendar (events · actual · consensus · prior). Curated news streams. Per-tenant watchlists.
5+ PROVIDERS · STALE FLAGS
02 · KNOWLEDGE
Compliant knowledge base
pgvector RAG with document-level ACL. Lark Wiki sync. Page-level inline citations. PII auto-redaction at ingestion.
ACL · CITATIONS · LARK SYNC
03 · RESEARCH
Auditable research agent
Investment Mode with 3-Agent cross-check. DataSlot binding for every figure. Done Gate before completion. RunState recovery on failure.
DATASLOT · DONE GATE · RUNSTATE
04 · REPORT FACTORY
Report Factory & delivery
Smart Slides + A4 finance templates. IC memo, one-pager, client quarterly. Export PPT / PDF / DOCX. Batch queue with idempotency. Approval before delivery.
PPT · PDF · DOCX · BATCH QUEUE
How a request flows through Finclaw
From your question to a signed-off answer.
Sensitive data is masked before any model sees it. Permissions and audit run on every step.
01 · Chat
Chat input & mask
Your message and any uploaded files enter a tenant-isolated lane and are sanitized before leaving storage.
- 1
Chat input
User questions and files arrive through the agent shell.
- 2
Parse & mask PII
Documents are parsed; names, IDs, accounts are tokenized.
- 3
Encrypted DataSlot
Vectors and metadata land in tenant-scoped, AES-256 encrypted storage.
02 · Reasoning
Multi-model AI
Only the masked prompt and context reach the model layer. Top providers run in parallel and cross-check each other.
- 1
Masked prompt
Sanitized text + retrieved context is forwarded to the model layer.
- 2
Claude · GPT · Gemini
Top providers run the same task in parallel; you pick the winner.
- 3
Draft answer
Numbers stay bound to their DataSlot source — nothing is invented.
03 · Delivery
Verify & deliver
A policy engine checks permissions, masked fields are rehydrated only for entitled viewers, and every action is logged.
- 1
Policy engine
Role, tenant and data classification are verified before any response is shown.
- 2
Rehydrate
Tokenized PII is restored only for users who are entitled to see it.
- 3
Deliver & audit
Final answer is sent to chat, Lark or email — full trace is logged.
Tenant-isolated storagePII masked end-to-endAppend-only auditRole-based delivery
What ships, not just what runs
Four deliverables your team can hand off today.
DAILY · MACRO
Daily Macro Brief
Markets open, macro events, sector rotation, internal flags. PDF + Lark posting + email blast.
A4 · PDF · Lark
WEEKLY · EQUITY
Equity Research Pack
Single-name deep dive: financials, valuation, catalysts, ESG flags, model debates. Slides + one-pager.
PPT · PDF · DOCX
PER DEAL · IC
IC Memo / DD Checklist
Investment committee memo with redlines from prior memos, DD checklist auto-populated from data room.
DOCX · A4 PDF
QUARTERLY · CLIENT
Client Quarterly Report
Per-client performance, attribution, narrative summary. PII redaction on all client identifiers.
Branded PDF
Templates ship with your branding. Mock data shown — real client data never leaves the tenant.
Governance is the product
Every line below is a default — not a sales promise.
- Every external action requires approval.
- Every sensitive read is permissioned.
- Every financial number is sourced.
- Every generated report leaves an audit trail.
ACLPII RedactionTenant IsolationApproval GateAudit Log
Not a deck. A platform that runs today.
Everything above is already shipped.
Here is how it shows up — today.
Multi-model entry
One inbox for Claude, GPT, Gemini, frontier open-source. Parallel runs, voting, win-rate, fork at any turn.
4 PARALLEL · WIN-RATE
Expert Agents
Legal · Risk · Research · Compliance experts. Each with its own prompt, KB, read-only tool whitelist.
PARALLEL · INDEPENDENT
Compliant knowledge base
pgvector RAG · Lark Wiki sync · document-level ACL · page-level citation · PII redaction.
ACL · RAG · LARK SYNC
Deep document intelligence
MineRU-grade PDF parsing: tables, formulas, footnotes, OCR, bbox citation. Plus Office and scanned docs.
PDF · OCR · OFFICE · BBOX
Financial data plane
Watchlist · multi-provider quotes · macro calendar · curated news · market summary boards.
5+ PROVIDERS · STALE FLAGS
Auditable research agent
Investment Mode · 3-Agent cross-check · DataSlot binding · Done Gate · RunState recovery.
DATASLOT · DONE GATE
Report Factory & delivery
Smart Slides · A4 finance templates · IC memo · one-pager · batch queue · multi-format export.
PPT · PDF · DOCX · BATCH
Admin · tenants · governance
Users · roles · models · API keys · cost dashboards · feature flags · audit logs · multi-tenant.
MULTI-TENANT · AUDIT
80+
Tools in runtime
23
Lark integrations
5
Specialized agents
3
Memory tiers
3
IM channels
5+
Financial data providers
What changes during the pilot
From kickoff to first signed-off workflow.
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
What compliance and CIOs ask in the first call.
Where does our sensitive data go?
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 auto-redacted at ingestion. Audit log on every read.
How do you prevent hallucinated numbers?
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. No silent fills.
Can we self-host or run on our own VPS?
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.
We already have Bloomberg / FactSet / ChatGPT Enterprise — why Finclaw?
Finclaw 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.
What does a pilot actually include?
One workflow (Daily Macro Brief or a research pack) 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.