Getting started

Rho is a lightweight agent harness for interactive coding sessions and one-shot terminal prompts.

Quick path

1. Install Rho.
2. Set up authentication and models.
3. Run rho to open the interactive TUI.
4. Read tools and workspace to understand how Rho reads, edits, and runs commands in your project, including its security and workspace boundaries.

First successful run

For an interactive session, start Rho and authenticate from the command palette:

rho
/login openai
/model openai/gpt-5.6-sol

For a one-shot API-key workflow, keep the key out of configuration and provide it through the provider's environment override:

OPENAI_API_KEY=... rho run "summarize this repository"

The active model and provider are saved when you choose them through /model or pass the corresponding CLI flags. See authentication and models for other providers and auth modes.

Choose a workflow

• Use the interactive TUI when you want an ongoing session with streaming output, tool calls, slash commands, and session resume.
• Use automation and CLI when you want a single answer for a script, hook, alias, pipeline, or CI job.

After the first run

Rho stores persistent configuration in ~/.rho/config.toml. The model catalog and /model command can update the active provider and model from inside the interactive TUI.

For local project work, see development. If authentication, model selection, or environment setup fails, run /doctor in the TUI; use /info to inspect the active selection and /limits to inspect supported OAuth usage windows (including last-observed Claude Code limits after a claude-cli run).

To delegate through Claude Code on a subscription, install the Claude Code binary, run /login claude-code, and define an agent with runtime: claude-cli. See when this is useful and how to use it.

Built-in host and shell integrations need no plugin store: see integrations. Herdr turns on under a Herdr pane, and RTK rewrites agent shell commands when the binary is on PATH.