Install
Get rho and the rho-egui desktop UI running on your machine.
What you need
- A Rust toolchain — edition 2024, so a current stable.
- PowerShell 7+ (
pwsh) — rho runs shell commands through it. This is deliberate: the model generates PowerShell, not bash, so what you see is what runs. - A model — either a local server (LM Studio, Ollama) or an API key for a hosted provider. You choose; rho has no opinion.
Get rho
git clone https://github.com/crustyrustacean/rho-coding-agent
cd rho-coding-agent
cargo build --release --package rho
Put the binary somewhere on your PATH. target/release/rho.exe is where it
lands on Windows.
Get the UI
rho-egui is the official desktop
UI — a native app that spawns rho in the background and renders the agent as a
chat.
git clone https://github.com/crustyrustacean/rho-egui && cd rho-egui
cargo run
It looks for rho on your PATH. If you'd rather point it at a specific build,
set RHO_PATH:
RHO_PATH=/path/to/rho cargo run
Point it at a model
Local model — start LM Studio or Ollama on localhost:1234, then:
rho --model <model-id>
Hosted provider — put your key in an environment variable and add a
provider to .rho/config.toml. Keys are never stored in config:
[agent]
model = "gpt-4o"
token_budget = 131072
[[providers]]
preset = "openai"
api_key_env = "OPENAI_API_KEY"
export OPENAI_API_KEY="sk-..."
rho --accept-external-provider
That flag skips the one-time consent prompt telling you your code is about to leave your machine. You should read it the first time.
If something goes wrong
rho keeps two records worth knowing about. Session data — the full entry
history for every project — lives under ~/.rho/sessions, filed by project.
And logs/ inside the project holds the runtime logs for the work you did
there. When something misbehaves, the session file tells you what the agent
saw; the log tells you what the agent did.
The documentation covers providers, tools, and configuration in full.