Prism reasons over your caller's actual audio, checks against your guardrails, and fine-tunes on the call data you already have. Run it on-prem, in your VPC, or in the cloud.
Join the private beta
Ownership has never sounded better
Speech-to-speech you can scale without compromising control.
Fine tuning
It learns from your real business data
Generic models guess at your policies, your products, your menu. Prism trains on your own calls, so accuracy, timing, and voice improve together. The result: more calls finished without a human.
guardrails
You see every word before it's spoken
Prism writes text before it speaks. Your guardrails, policies, and required disclosures sit between the model and the caller. If your industry requires you to review what an AI says, this is the model you can actually deploy.
deployment
It runs where your data lives
On-prem, in your VPC, or in the cloud, with HIPAA compliance and SOC 2. The frontier realtime APIs are cloud-only. For many enterprise teams, that ends the conversation before it starts.
Natural speech above. Coordinated work underneath.
Made for real builders and real outcomes
Prism is for enterprises with high call volumes, in-house agent teams, and calls too sensitive to hand a model you can't control.
61%
Of callers choose Rime voices
4x
Fewer hangups
42%
Increase in call containment
Frequently asked questions about Rime’s speech-to-speech model, Prism
Doesn't a text step make it a cascade?
No. A cascade reasons over a transcript and loses the caller's tone. PRSM reasons over the audio and emits text so your rules can review every word before it becomes speech. The text is the output checkpoint, not the input to the reasoning.
What does a fine tuning of PRSM require?
Data you already have: call transcripts and tool-call logs. No paired audio training data. During the private beta, our team runs fine-tuning with you.
Where does PRSM run?
On-prem, in your VPC, or in the cloud. HIPAA and SOC 2. Telephony-ready.
How does turn taking work?
PRSM is turn-based at launch, like every developer voice API on the market today. Full duplex is on our roadmap, and the text checkpoint stays as it arrives.
Is there an API?
Not yet. v0 is in private beta. Public API is coming soon.
Try Prism in private beta
We're adding new partners to the private beta now. Partners get direct access to the Rime research team, a model fine-tuned on their own call data, and a hand in shaping the roadmap.
If you're building your own voice agent and your calls demand control, we want to talk.
Request access to the private beta
You're on the list. We review every request and reply to strong fits within a week.
Oops! Something went wrong while submitting the form.