More sessions per GPU
56×
Read the case study
YBacked by YC W25
A continuous
A continuous
inference engine
Bring your streaming model. We integrate it and run the inference, managing execution, session state, and scheduling.
Streaming performance.
More live sessions.
Kept on time.
Faster shared steps create room for more live sessions within their timing budgets.
Swipe to compare
| What matters | Conventional LLM serving | .Wave engine | Benefit |
|---|---|---|---|
| Unit of work | Batch steps from bounded requests | Batch compatible steps from live sessions | Share GPU cost across live streams |
| Timing | Response and token latency targets | Per-session timing requirements | Keep streams on time |
| Scaling | Requests or tokens per second at target latency | Concurrent sessions meeting timing targets | Grow capacity while protecting active streams |
Built for streaming models.
Models that process incoming data incrementally, without waiting for the full input sequence.
Articles
Continuous inference needs a different GPU engine
Engine
Why request-serving stacks fail continuous inference, and how WPK keeps recurring work on the GPU to serve 56× more real-time sessions on the same H100.
Read the article
Duplex inference runs on a clock
Systems
AI inference is moving beyond requests. New input arrives incrementally, model state persists, and each output is useful only while it is still current.
Read the articleA streaming model?
A streaming use case?
Tell us about your model, expected traffic, and timing requirements.
Talk to an engineer