> SIE is an open-source inference engine that runs the models behind every agent task through one API: search and retrieval, document-to-markdown conversion, structured output, content safety, and the agent loop itself. It replaces the patchwork of a separate model server per task with one system that serves 100+ models, loading each on demand.
3rd time reading this and it's still confusing.
> an open-source inference engine that runs the models behind every agent task through one API
"the models"? - it means "various models that are called by an agent, per task"
(then lists all the tasks that it covers)
> "It replaces the patchwork of a separate model server per task with one system that serves 100+ models, loading each on demand."
SIE exposes 4 task oriented primitives: encode, score, extract, generate. The application calls the relevant primitive with a supported model, while SIE automatically handles the model specific runtime, loading, batching and GPU sharing. Models load on demand, and the least recently used ones are evicted when GPU memory is needed. So the "100+ models" refers to the supported catalog, not 100 models loaded simultaneously.
3rd time reading this and it's still confusing.
> an open-source inference engine that runs the models behind every agent task through one API
"the models"? - it means "various models that are called by an agent, per task" (then lists all the tasks that it covers)
> "It replaces the patchwork of a separate model server per task with one system that serves 100+ models, loading each on demand."
Which says nothing about how it works.
The About block is poorly written.