Model freedom · BYOK voice AI
Most platforms hide the model bill inside your subscription, and quietly pick the model too. Bring your own key and it stays yours.
Add an OpenAI, Anthropic, or other provider key in Settings, and every agent you point at it runs on your rate, your data terms, and your own infrastructure if you already have one, not a blended cost Zumu decided on your behalf.
- AES-256-GCM envelope encryption, per organization
- Validated with a live ping before it serves a call
- Your own base_url, for your own gateway
- Provider and model selection, per agent
The key lifecycle
Four stages stand between pasting a key and a call actually using it
None of it happens by hand, and none of it is optional. A key that fails its ping never gets a call routed through it, and every call that follows keeps going through the same four steps in order.
Your provider key
Add an OpenAI, Anthropic, or other provider key straight into Settings, the same place you already manage everything else about an agent.
Sealed the moment it lands
The key is encrypted per organization before it is ever written to a database, so nobody, Zumu staff included, reads it back in plain text again.
Checked against the provider
Before the key is trusted with a real call, Zumu sends one live request to the provider and confirms the key actually works.
Serving calls right now
Every agent configured to use it routes its model, speech, or voice requests through your key from that point on, call after call.
The same prompt-cache warm ping that keeps first-call latency low runs the same way on your own key, because the cache lives in Zumu’s routing layer, before the request ever reaches your provider.
Why it matters
A key you bring is still the relationship you already have
None of this is about Zumu working harder. It is about four things that stay exactly where you already put them.
Your rate, not a markup on it
The provider account is still yours, so whatever you already negotiated with OpenAI, Anthropic, or another provider is the rate that applies to every call your key serves. Zumu does not sit between you and that invoice.
Your data agreements, not a shared one
Every request goes out under your own provider account, so it is covered by whatever data-processing agreement you already signed with that provider, not one blanket agreement Zumu negotiated for every customer at once.
Swap a provider without a migration project
Model and provider selection happens per agent, not per platform. Point one agent at a different model or provider, and the rest of your account keeps running on whatever it was already using.
Your own gateway, through base_url
A base_url override lets a key point at any OpenAI-compatible endpoint you already run, instead of the provider’s default address, for teams that already sit a gateway or their own deployment in front of a model.
What doesn't change
Bringing your own key doesn't make the platform free
Your provider bills you directly for the model. The phone call itself, the recording, the transfer, and the wallboard your supervisors watch are a separate thing, and Zumu still meters that part.
Changes with your own key
Your provider bills you directly for every model, speech, or voice request that key serves.
The rate and the data-processing terms are whatever you already agreed with that provider.
A key can point at your own base_url instead of the provider’s default endpoint.
Stays the same
Zumu still meters every AI-handled minute on the call, on the same usage-based basis as any other agent.
Recording, transcription, and the operations surface, the wallboard, transfers, and QA, keep running exactly the way they already do.
A transferred call still only bills for the minutes the AI agent actually handled, key or no key.
Bringing your own key changes who you pay for the model. It does not change what Zumu meters for running the call.
Where this fits
A provider key is one part of the platform, not the whole of it
- Security
How the rest of a call is secured
Envelope encryption on a provider key is one control among several. See what protects a call end to end, told as controls rather than a certification badge.
See security - The platform
Where a provider key sits in the architecture
A key is one part of the agent runtime layer, which sits beside the call layer, the knowledge graph, and the operations surface your supervisors work from.
See the architecture - Pricing
What the platform itself meters
Your provider bill is separate from what Zumu meters for running the call. See how the usage-based rate card actually works.
See pricing
Questions
What teams ask before bringing their own key
Still have questions?
Bring your own provider contract. We will walk through exactly how it plugs in.
Book a demo
Bring the account you already have.
Point Zumu at the OpenAI, Anthropic, or other provider key you already pay for, and we will show you exactly which agent is using it and what it cost, call by call.
- No credit card for the demo
- Bring the key you already pay for