HiRoute acts as a local coordination layer across macOS and Linux, directing agent requests to tailored model pools rather than committing an entire session to a single expensive provider. It selects models anew on fresh conversational turns while freezing choices during tool executions. Although it runs locally, requests pass onward to external model services, and reported benchmark savings remain specific to tested configurations rather than universally guaranteed. Rebuilt conversation contexts or unrecognizable tool continuations trigger new routing evaluations.
How you can use it
Install the desktop app on macOS or the headless service on Linux, then connect a decision provider and the models that will perform the work. Create a routing plan and point your agent at HiRoute. Inspect its session records to check which model actually ran; a locally installed router can still send your task to outside services.
