How it works

The whole brain runs inside your walls.

No jargon, no black box. Here is exactly where your data lives, where the AI thinks, and the one optional path that can ever reach outside your network.

Inside your private boundary

Everything below runs on hardware you control, or on dedicated hardware we run just for you. Nothing here touches the public internet.

Your systems & documents

Core banking
Loan files & PDFs
Fraud & CRM
AgentFinOS engineyour hardware or ours

Edge AI model

Post-trained open-weights model running locally. The brain that reads, decides, and writes.

Context graph

A private knowledge graph of members, accounts, loans, and policies the model reasons over.

Task runner

Carries out each step across your systems and pauses for a person when policy says so.

Finished work

Drafted & ready
Your team approves
Logged for audit

Optional: your approved cloud model

Already standardized on a vendor? Point the same engine at OpenAI, Anthropic, Azure, or Bedrock instead. The dotted line is the only path that ever leaves your network, and you decide whether to turn it on.

OpenAIAnthropicAzure OpenAIAWS Bedrock

What happens when a task runs

Four steps, start to finish. The interesting part is what never happens: your member data never leaves the building.

1

A task comes in

A new loan file, an alert to review, an account to open. The work starts exactly where it does today, in the systems your team already uses.

2

The engine reads your context

It pulls the right procedure from your context store, the SOPs and rules you taught it, so every decision follows your institution's own playbook.

3

The local model does the thinking

An open-weights model running on your hardware reads the documents, makes the call, and drafts the work. No data is sent anywhere to make this happen.

4

A person approves what matters

Anything that touches a member, a dollar, or a decision waits for your team's sign-off. Every step is logged in plain language for your examiners.

The short version

Where it runsYour hardware, or dedicated hardware we run for you
The modelOpen weights, self-hosted, fully inspectable
Your dataStays on your network. Zero egress by default
Your contextSOPs and knowledge stored on-site, never shared
CostFlat and predictable. No per-token metering
Cloud vendorOptional swap, on only if you choose it

Want to see it run on your own workflows?

We will walk your team through a task you pick, on your systems, and show you the audit trail afterward.