Configuration

Rho stores persistent config at ~/.rho/config.toml by default.

provider = "openai"
model = "gpt-5.5"
max_output_bytes = 12000
auth = "api-key" # or "codex"
reasoning = "medium" # off, minimal, low, medium, high, or xhigh
auto_compact = false
compact_threshold_percent = 85
compact_recent_messages = 8

CLI overrides

Passing --provider, --model, --auth, or --reasoning updates the config file and makes the choice the future default.

rho --provider openai --auth api-key --model gpt-5.5
rho --provider openai-codex --auth codex --model gpt-5.5
rho --reasoning high

These values select authentication and models.

You can load and save a specific config file with:

rho --config ~/.rho/config.toml

--no-system-prompt and --no-tools are only available on the command line and apply only to the current run.

TUI updates

In the interactive TUI, /config opens a picker for configuration values. The reasoning row cycles through off, minimal, low, medium, high, and xhigh, saves immediately, and applies to the current session. The max_output_bytes row opens a numeric input and saves for the next session.

/login stores credentials in the OS credential store, not in this config file. /logout deletes stored credentials. /model saves the selected provider and model. The picker shows entries from Rho's static model catalog for providers with available auth, and /model provider/model can switch explicitly.

Reasoning options

reasoning is the user-facing thinking level. Supported values are off, minimal, low, medium, high, and xhigh. For supported OpenAI Responses providers, off omits the reasoning object. Other levels send reasoning.summary = "auto"; minimal maps to effort low, while low, medium, high, and xhigh map to matching effort values.

Tool output limit

max_output_bytes controls how much output Rho keeps from tool calls such as command output and file reads.

Auto compaction

auto_compact enables summarizing older conversation history when the estimated current context approaches the model window. It is disabled by default. compact_threshold_percent controls the trigger point, and compact_recent_messages controls how many recent messages are kept verbatim after older history is summarized.