DPO Radio

An AI advisor reads DPIAs, contracts, vendor records and breach notes. Before it does, procurement, security and the data protection officer want to know which AI provider will process that material. Many organisations have already reviewed and contracted one provider. They should not have to approve a second one just to use Forseti.
Forseti runs on the provider your organisation configures. An administrator enters the organisation's own API key for Anthropic, OpenAI or any endpoint that implements the OpenAI API, such as a model gateway or a self-hosted model server. The prompts and documents Forseti sends go to that provider, under your organisation's own agreement with it. You record the provider in your vendor register and your processing records, the same way you record any other processor.
Cost follows the same decision. There are no prepaid Forseti tokens and nothing to top up. Your provider invoices your organisation for what Forseti uses, and a run is never refused because a Forseti allowance ran out. The Usage page reports what was consumed, so a budget owner can set it beside the provider's invoice.
Forseti itself does not change with the provider. The personas, citation grounding, human sign-off on every draft and the tool policy are the same whichever provider answers.
The AI Model page has a card for Anthropic, one for OpenAI and one for an OpenAI-compatible endpoint. Each card holds its own API key, model and effort, so you can prepare more than one provider and switch between them.
You paste the API key once. It is stored encrypted and never shown again: the card shows only its last four characters. Leaving the field blank on a later save keeps the saved key, and removing a key asks you to confirm.
Choose a model from the suggestions or type any model your provider offers. Suggestions include Claude models for Anthropic, and GPT-6.1 Sol, the GPT-6 family and GPT-4.1 for OpenAI. Which models you can use depends on your own provider account. An optional fast model handles conversation titles and tabular extraction. Effort runs from model default through low, medium, high, extra high and maximum, and levels the selected model does not accept are disabled.
The OpenAI card adds the API mode, a Pro reasoning mode where the model supports it, and the service tier. The OpenAI-compatible card takes a base URL over HTTPS, so a model gateway or a self-hosted model server can serve Forseti.
Test connection checks the card's saved settings with one short request that contains no customer data. Your provider may charge for it. The card shows the result, the time and the provider's short explanation, and it tells you when it has unsaved changes the test would not cover.
One provider is active at a time, and only a provider with a saved key can be made active. Every Forseti feature then routes to it: chat, workflows, tabular reviews, document redlines and titles.



The API key is write-only. After it is saved, the page shows only its last four characters, and no screen returns it.

Changes are recorded. A key set, replaced or removed, a provider switched and a connection test each leave an audit event naming the administrator, and never the key itself.

The record keeps up with the choice. Every Forseti request is written to the usage ledger with the provider, model and effort that handled it, and the Usage page reports it by period.

ComplianceOne is customer-managed and self-hosted. The configured provider or compatible endpoint determines where model inference runs and how its inputs are handled. Delegated reading uses routing and budgets set by the organisation.
Ready to see Forseti switch to the provider your organisation already uses, pass a connection test, and show the routing on the Usage page? Request a personalised demo.


Run a pilot on your own provider account. Connect the provider, pass the connection test, and compare a month of Forseti usage with the provider's invoice.

Talk to our team about connecting Forseti to the AI provider your organisation has already approved.
Anthropic, OpenAI, and any endpoint that implements the OpenAI API, such as a model gateway or a self-hosted model server reached over HTTPS. Each provider needs its own API key, entered by your organisation's administrator. One provider is active at a time.
No. Prepaid token bundles and top-ups are gone. Your organisation pays its AI provider directly for what Forseti uses, and the Usage page reports the consumption so you can compare it with the provider's invoice.
The prompts and documents Forseti sends go to the provider your organisation configured, under your organisation's own agreement with that provider. How long the provider keeps them is set by that agreement and the provider's terms. Record the provider as a processor in your vendor and processing records.
Yes. You choose the model, an optional faster model for titles and tabular extraction, and an effort level from model default to maximum. The page offers only the effort levels the chosen model accepts. Which models you can call depends on your provider account.
No. The key is stored encrypted and is write-only. After saving, the page shows only its last four characters, and the audit trail records who changed the key without recording the key.