Bring Your Own Provider
Supply your own AI provider API key and model choice. There is no bundled model and no vendor lock-in on which provider or model your bank runs.
Native AI
Bring your own AI, grounded in your own compliance data.
Connect the AI provider and model your bank already trusts, then let analysts hold a contextual conversation over your own customers, transactions, and cases, with confirmation-gated actions and no delete access at all.
Bring your own provider
Your model. Your data. Your existing access boundaries.
A conversation over your compliance data is only useful if it respects the same boundaries your team already operates under. OneView's AI assistant is built on your provider, your model, and your existing permissions.
Grounded In Your Own Data
Conversations stay contextual to the client's own compliance records, not a generic model with no visibility into your customers, transactions, or cases.
Permission- & Licence-Gated Read Tools
Every read tool available to a conversation is bounded by the requesting user's own permissions and the modules your bank is licensed for, the same as the rest of OneView.
Persisted Transaction Monitoring Investigations
Continue an AI-assisted Transaction Monitoring investigation exactly where it left off. Prior conversation context is retained, not rebuilt from scratch each session.
Confirmation-gated write actions
Nothing changes your data without explicit confirmation.
Read access already follows the requesting user's permissions and licence. Write actions add a second, explicit safeguard on top of that: nothing changes compliance data without the user confirming the exact action first.
Confirmation Required Before Any Write
A write action is never executed on the strength of a single conversational request. The requesting user must explicitly confirm the specific action before it runs.
Delete Is Never An AI Action
Delete actions are not implemented as callable tools at all, so no conversation, prompt, or confirmation can remove a record through the AI assistant.
Persistent memory
Context that carries forward, entirely on your terms.
The assistant retains key facts and preferences across separate conversations, only when you tell it to, giving you full control over everything it stores.
Remembers What You Instruct It To
The assistant can remember what you instruct it to, so you never have to repeat yourself.
Carries Across Conversations
What's remembered stays available in later, separate conversations, not just for the rest of the current chat.
Always Visible, Always Removable
Everything remembered can be viewed and deleted at any time from a dedicated page, so you stay in control of what the assistant carries forward.
Transaction Monitoring rule authoring
Turn a plain-language description into a working detection rule.
Build detection logic that would otherwise need an engineer, described the way a compliance officer already thinks about it.
- 01
Start From A Plain-Language Description
Describe the detection logic you want in plain English. No code and no engineering ticket required to get started.
- 02
Compares Calculated Values
The assistant can build logic that compares two calculated values, for example comparing how much money moves out of an account against how much moves in over a chosen time window.
- 03
Spots Patterns Across Multiple Transactions
It can also detect patterns spread across several transactions rather than just one, the kind of logic needed to catch typologies like structuring.
- 04
Asks When It Needs More Detail
If more information is needed to complete the rule, the assistant asks a structured question with clear answer choices, single-select, multi-select, or free text, and waits for a real answer before continuing.
Transaction Monitoring
AI investigates every flagged transaction.
The moment a transaction comes back Flag, Hold, or Block, AI builds a structured investigation from non-identifying, server-calculated evidence only, never raw personal data.
- 01
A Summary, And Why It Was Flagged
A plain summary of the transaction and why it was flagged.
- 02
Customer And Counterparty History
Customer history and counterparty analysis.
- 03
Device And Location Signals
Device and location signals.
- 04
A Risk Assessment, With Its Reasoning
A Low, Medium, High, or Critical risk assessment, with its reasoning.
- 05
Next Steps And Data Limitations
Recommended next steps, and any limitations in the data available.
- 06
An Analyst Still Decides
AI never approves, releases, or changes the transaction. An analyst still makes and acts on that decision, with the investigation saved to come back to.
Reporting
AI explains exactly what to fix before you submit.
When a report fails validation against goAML, the report format Nigeria's financial intelligence regulator (NFIU) requires, an AI guidance panel groups the related errors together and explains exactly which tab and field to open and what to change.
- 01
Errors Grouped By What's Actually Related
Errors grouped by what's actually related, not a flat list.
- 02
Names The Exact Tab And Field To Open
Names the exact tab, record, and field to open, for example: open the Transactions tab, find transaction TXN-..., and check the source and destination country fields.
- 03
Explains What To Change, And Why
Explains what to change and why.
- 04
AI Never Submits The Report
AI never validates, approves, rejects, or submits the report itself. It only explains what a person should edit.
- 05
Party Details Are Redacted First
Party details such as names, addresses, identification numbers, dates of birth, and account numbers are redacted to fixed placeholders before anything reaches the AI provider, only the document structure and validation errors are shared.
Start a conversation
Request an AI DemoReady to put AI to work inside your own boundaries?
See how a bring-your-own-provider AI assistant can work over your own compliance data without giving up permission, licence, or delete boundaries.