.Wave
.wave provides a purpose-built inference engine for continuous, real-time AI workloads like full-duplex voice and streaming speech recognition. The platform replaces request-based serving with persistent, session-scoped steps governed by timing contracts, ensuring outputs meet recurring deadlines. It offers a real-time API with no audio or transcript retention, and supports concurrent sessions sharing GPU capacity efficiently.
- Artificial Intelligence
- Developer Tools
- Software Only
Funding
Raised to date
$500KRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
Founders
Founder details are not available yet.
Product
Problem
Traditional AI inference engines are optimized for bounded requests, where a small delay only affects response latency. However, a growing class of workloads—such as full-duplex voice and streaming ASR—requires persistent, session-scoped state and must produce output on a recurring wall-clock deadline. Missing a deadline causes a gap in audio or a model state that no longer matches reality, making standard request-based serving inadequate for these live applications.
Solution
.wave provides an inference engine purpose-built for continuous, live AI workloads. It replaces request-based serving with recurring steps from persistent live sessions, each governed by a timing contract that ensures outputs meet their deadlines. The engine shares GPU capacity across concurrent sessions, allowing more live sessions to run on time without falling behind schedule. This approach supports full-duplex voice and streaming ASR workloads by keeping pace with recurring frame deadlines. The service is accessed via a real-time API, with no audio or transcript data stored anywhere in the service.
Target Audience
Primary users are developers and organizations building real-time voice agents, streaming speech recognition systems, and other continuous-inference applications that require low-latency, deadline-bound model execution.
Features
- Inference engine that schedules recurring steps from live sessions rather than bounded requests
- Timing contracts per session to enforce recurring deadlines and manage playback buffer margin
- Concurrency-aware scheduler that shares GPU capacity safely across multiple live sessions
- Optimized for small-batch, heterogeneous model paths (audio encoder, backbone, decoders, codec) in duplex systems
- Real-time API with TLS 1.3, scoped and revocable API keys, and no audio or transcript retention
- Session records retained for 90 days containing only counts, timestamps, and metadata; no payload content
- Dedicated Lanes option for reserving inference capacity with guaranteed latency and an SLA, matching reservation to workload and hours needed