Auto now balances quality, speed and cost as the work changes. Quick interactive actions favor UltraFast, careful refactoring and difficult reasoning favor UltraCode, and understood, bounded work favors UltraSave.
- Use
/mode autoto adapt during a conversation, or select a fixed mode. The footer shows the choice and the mode currently in use. - Subagents inherit your mode choice, including later changes. Each Auto subagent can choose the mode appropriate to its own action.
- Explicit model and effort selections stay authoritative. Fast and economical execution keep native MCP tools available.
- Usage accounting now includes initial Auto decisions and subagent setup.
In a paired 20-task sample, Auto and Normal both passed every task. Auto used 57% less aggregate completion time and 77% fewer estimated LLM subscription credits; Jev USD costs were measured separately and were also lower. These are sample results, not a universal performance guarantee. Specialist comparisons and validation details are in PR #45.
Downloads include native binaries for macOS (Apple Silicon and Intel), Linux (ARM64 and x86-64), and Windows (x86-64), with SHA256SUMS.