Why A Coding Agent Is Better Without A Runtime.

Anvaya CLI is a single static Rust binary: 12 ms to first TUI frame, 17 ms for --version, 4.2 MB kernel footprint at idle (12.4 MB ps RSS), 2.8 MB in headless setup, and roughly 11 MB held across a 17.6-hour session. macOS footprint is the honest instrument: RSS includes shared mapped pages and hides compressed ones, so we publish both and the protocol behind them.

In Phase 1 of our benchmark program (six harnesses, three gateway models, ten task classes, 238 runs, external verification) the harness measured 25.6 MB median own-process RSS with a 21 MB idle floor. The Node-based harnesses measured 432 MB and 582 MB; the Python one 249 MB. Their runtimes set floors no context-management improvement removes, which is exactly why the number matters: it is what a user pays before any work happens.

Compile times and a smaller extension ecosystem. Rust also raises the bar for contributors, unsafe code, lifetimes, and a type system that makes the wrong abstraction feel expensive. For a process that must stay resident for hours and never surprise the user, that trade is worth it; for a weekend script, it emphatically is not.

And a static binary is not automatically light: the persistent-memory daemon is a separate process that warms to roughly 50 MB with its embedding model mapped. We treat that as its own subsystem with its own budget, and it can be left out entirely.