Roadmap

From engine
to release

Loculus Core is built. It is now being measured, model by model and machine by machine, before its first release on Linux.

1. 01

   Complete

   The engine

   A new inference engine written from first principles: its own CPU and GPU kernels, memory governor, expert streaming from SSD, prefix caching and lossless speculative decoding, behind the standard chat APIs.

2. 02

   In progress

   Benchmarking & compatibility

   Measuring speed and memory across model families, quantization formats and GPUs, and validating every supported model against an exact reference.

3. 03

   Next

   Linux release

   The first release. Sign in with Google or Apple, buy once, and your purchase is linked to your account, with regular engine updates.

4. 04

   Then

   Windows

   The same engine and the same account: download on any machine you sign in on.

5. 05

   Later

   Android, macOS and iOS

   In that order, so one purchase follows you from desktop to phone, across different hardware.