Base Compute

Base Compute is an AI inference lab that builds runtimes and infrastructure to run powerful AI models directly on your own hardware — laptops, phones, and tablets. Their flagship product, BaseRT, is a runtime optimized for Apple Silicon that claims to prefill up to 6.4x faster than llama.cpp and 3.9x faster than MLX, with decode speeds up to 1.33x faster than MLX. The company's mission is to make AGI run on-device, meaning lower latency, real privacy, and zero marginal cost once the model is running on hardware you already own.

BaseRT is the core of what they offer. Benchmarks on the page show decent performance gains over competing runtimes on Apple M5 Pro hardware. The company is also building enterprise-grade infrastructure for on-premise, hybrid, or fully air-gapped deployments, letting organizations run AI inside their own environment with full control over data and systems. Their research focuses on automated research pipelines and GPU kernels tuned per hardware and per model to squeeze the most throughput out of local silicon.

The main audience for Base Compute is developers and researchers who need to run LLMs locally without sending data to third-party servers. It's also a good fit for enterprises with strict data privacy requirements, like healthcare, finance, or defense. If you're building an app that needs fast, private inference on Apple hardware, BaseRT is worth a look. The company is small, senior team based in Melbourne and Berlin, and they're hiring.