Marathon CLI

Marathon CLI now has a much more capable interface with an Ink-based TUI, context compaction, tool input streaming, session search, pluggable workflows, and stronger safety and upgrade controls. The result is a better long-running agent loop for terminal work, not just a thin wrapper around API calls.

For developers, that means longer sessions stay usable, tool behavior is easier to inspect, and the CLI can adapt to different workflows like TDD or code review. It's especially helpful to analyze why long task runs are not succeeding, as you can export the entire session for review. That is used internally at Runtype to deepen our understanding of what models need to perform optimally.