rho

A coding agent you pair with.

Reads your code, runs your tests, and works the problem with you — on your machine, with your approval at every step.

Forged in Rust

No runtime panics mid-edit, no GC pauses in the tool loop, and 1600+ tests gating every commit.

The agent you trust with your working tree shouldn't crash taking your files with it.

Why the language matters

rho is a single Rust workspace with no runtime between it and the metal. That buys three things you can feel in daily use:

  • No surprise termination. No runtime exception that kills the process with your edits unsaved. Errors are values, handled where they occur.
  • No pauses. There's no garbage collector stopping the world mid-tool-loop, so a long session behaves the same at hour three as at minute one.
  • Predictable resource use. Streaming is bounded, memory is released when a session closes, and an idle rho costs effectively nothing.

Structured diagnostics

Most agents shell out to cargo check and hand the model the raw output. rho parses the NDJSON instead: each diagnostic carries its error code, message, source spans, and machine-applicable suggestions. Dependency noise is filtered out, so the model sees your code's problems rather than a thousand lines about someone else's crate.

The practical difference: the model points at a specific span and makes a targeted edit, instead of guessing from prose. A fix that doesn't compile produces a real error it can read, not a shrug.

Tested like it matters

Every commit to rho runs the same gate: format, lint, audit, build, and the full test suite — 1600+ tests across the workspace. Branch protection enforces it on trunk. Nothing lands because someone was confident; it lands because the suite passed.