Skip to content
Case studies

AI agents act on financial data, inside a mandate.

A financial platform wanted its customers' AI agents to query its data and initiate changes, without exposing the transactional core to them.

Client withheld under NDA

Challenge

An agent that only reads changes nothing. One that writes has to prove what it changed, under whose authority, and inside limits the client set in advance.

What we built

  • A prepared data layer, kept apart from the transactional core, so an agent's questions never run through the payment path.
  • A semantic layer over it, so an agent asks in business terms and gets an answer in them.
  • MCP for reading, and a separate Change MCP for writes, each write confirmed before it lands.
  • Agent identity, mandates and policies: who an agent acts for, which actions, what amounts, over what period.
  • Limits and escalation rules that stop a run and hand the decision back to a person.
  • An audit trail carrying every request and every action, with the mandate it happened under.

Outcome

The platform's customers run agents against their own data, and every change one makes is confirmed, bounded by a mandate, and recorded.

Capabilities

  • MCP
  • Change MCP
  • Semantic data layer
  • Agent identity
  • Mandates and policies
  • Audit trail

Related work

Describe your challenge.

Early idea or system already in production. Both are worth a conversation.

  • Built and run by our senior team.
  • Confidential from the first call, NDA as standard.

We couldn't send your message.

Your message is still in the form. Try again, or send it to [email protected] instead.

Open email draft

Message sent.

Thank you. We'll reply within one business day.

Prefer email?

[email protected]