Products
Make memory work harder for large AI models.
Experimental: compression is still being tested.GecoStream investigates memory-efficient execution of Mixture of Experts models. Compression aims to reduce stored model data. Expert streaming loads relevant model parts from NVMe storage, with caching and prefetching intended to reduce waiting. These are complementary techniques, not a guarantee that every model fits every computer.
Imagine a library whose complete collection cannot fit on one desk. Keep frequently needed books nearby and bring others from the shelf when needed. This illustrates streaming only: storage speed, available memory and model behavior still constrain performance.
Experimental measurements apply to their tested setup, and compression quality is still being tested on full models. OpenAI-compatible interfaces and multi-computer execution are architectural directions, not a guarantee of compatibility with every client.
Begin in the GecoChat app. Check that the feature is available for your account and device, review permissions and execution location, then try a small task with non-sensitive information. Use the related guides below before expanding the task.
Check the current release and compatible devices in the GecoChat app. A model’s operating system, processor, available memory and storage requirements must be checked separately. A platform listed in the roadmap may not be supported yet.
How does it fit into GECO? Use the related products and guides below to follow the connection between conversations, assistants, computer tasks and model execution. Does every described capability work today? Not all of them: the status at the top of the page shows where each product stands, and the GecoChat app shows what is available on your account.