Software Engineer, Runtime
Confirmed on the employer's own hiring board on Oct 16, 2026. First seen by Alion on Jul 9, 2026. Ollama scores B on the Alion truth index.
Ollama is the most popular way for developers to access open models. What started as an open-source, local-first runtime is now the largest developer network in the open-model ecosystem: 8.9 million monthly active developers and over 67,000+ community-built integrations. We're backed by Y Combinator, Benchmark, 8VC, and Theory Ventures.
Our team is small and talent dense. We're flat, low-ego, and fast-moving. We like people who are truth-seeking, passionate, design-driven, and who enjoy shipping code.
About the role
You'll work on the heart of Ollama - the local runtime that runs open models on developers' own machines. It loads models, manages memory, drives GPU acceleration across NVIDIA, AMD, Intel, Qualcomm, and Apple Silicon (including our MLX integration), and makes all of it feel instant. You'll work in Go and C/C++ and touch the model formats and inference engines underneath, shipping to macOS, Linux, and Windows across an enormous range of hardware.
What you'll do
Make open models run fast and reliably on consumer and enterprise hardware - from a MacBook Pro to server-grade NVIDIA GPUs.
Own pieces of the runtime: model loading & scheduling memory management, quantization, GPU hardware backends.
Integrate new model architectures and quantization formats so the latest open models work on day one.
Improve cold-start, time-to-first-token, and throughput
Partner with model labs and hardware vendors on early access and deep integrations.
Ship in the open: Ollama is open source, and you'll work with the community
Example projects
Add support for a new model family end-to-end - format parsing, weights loading, and the defaults that make it useful out of the box.
Cut cold-start for a popular model in half by streaming weights and lazy-loading layers.
Land a new quantization format so a 70B model runs on a single consumer GPU.
Wire up a new GPU backend and find a 2x throughput win with kernel selection and memory tuning.
Improve the "Auto" experience - picking the right model and settings for a machine's hardware without the user thinking about it.
You may be a fit if
You have strong systems fundamentals and are comfortable in Go, C, or C++
You've worked close to the metal - GPU compute, inference, game engines, databases, OS, or networking.
You care about performance and have profiled and optimized real workloads.
You're comfortable shipping to millions of users and handling the long tail of hardware and OS combinations.
Bonus: experience with model quantization, GPU programming (CUDA/Metal/SYCL), Apple MLX