Settings

Settings File

Rusty stores its persistent configuration at ~/.rusty/settings.json. This file is created by the setup wizard on first run.

{
  "api_key": "sk-...",
  "api_base": "https://api.xiaomimimo.com/v1",
  "default_model": "mimo-v2.5-pro",
  "models": [],
  "active_model": "mimo-v2.5-pro",
  "api_keys": {},
  "allowed_tools": [
    "bash:git status",
    "bash:cargo check"
  ],
  "credential_store": "keyring",
  "thinking_level": "normal",
  "permission_mode": "default",
  "permissions": {}
}

Fields

Field	Type	Description
api_key	string	API key for the LLM provider (stored if not using keyring)
api_base	string	Base URL for the API endpoint
default_model	string	Model identifier to use by default
models	array	Model registry entries (see Model Registry below)
active_model	string	Which model entry is currently active
api_keys	object	Per-model API keys (keyed by model ID)
allowed_tools	string[]	Permanently allowed tool invocations
credential_store	enum	"keyring" or "settings_file"
thinking_level	enum	"minimal", "normal", or "deep"
permission_mode	enum	"default", "accept-edits", "bypass", or "plan"
permissions	object	Per-tool permission decisions

Model Registry

The models array defines available LLM models. Each entry has:

Field	Type	Description
group	string	Display grouping (e.g. “Xiaomi”, “Anthropic”)
name	string	Human-readable name
provider	enum	OpenAI or Anthropic (determines wire format)
api_base	string	API endpoint URL
model	string	Model ID sent to the API
available_models	string[]	Models to offer in the picker
context_window	number	Context window size in tokens
thinking_budget	number	Token budget for reasoning/thinking
max_tokens	number	Max output tokens
temperature	number	Sampling temperature
extra_headers	object	Additional HTTP headers

The active_model field selects which model entry is used. The TUI /settings command provides a visual editor for the model registry.

Thinking Levels

The thinking_level field controls the token budget allocated to reasoning:

Level	Token Budget	Use Case
minimal	1024	Simple queries, quick responses
normal	4096	Standard multi-step tasks
deep	16384	Complex reasoning, architecture decisions

Dynamic adjustment: The agent automatically adjusts the thinking level based on context:

• Multi-step tasks (2+ tool turns) are boosted from Minimal to Normal.
• Context usage above 70% triggers a one-level step-down.
• Context usage above 85% forces Minimal regardless of setting.

CLI Flags

All settings can be overridden at runtime via CLI flags:

Flag	Description
--model	Override the model
--api-key	Override the API key
--api-base	Override the API base URL
--preset	Use a named preset (overrides api_base and model)
--permissions	Set permission mode: default, accept-edits, bypass, plan
--max-turns	Maximum agent loop iterations
--max-tokens	Maximum tokens in the response
--temperature	Sampling temperature
--thinking-budget	Token budget for reasoning/thinking content
--plan-with-tasks	Enable structured task tracking in responses
--cwd	Set the working directory
--prompt	Run in headless mode with a single prompt
--headless	Run in stdin REPL mode
--resume	Resume a saved session by ID
--list-sessions	List all saved sessions
--verbose	Enable verbose logging
--setup	Force the setup wizard to run
--no-claude-md	Disable discovery of AGENTS.md/CLAUDE.md/RUSTY.md context files
--append-system-prompt	Append additional text to the system prompt

Environment Variables

Variable	Purpose
OPENAI_API_KEY	API key (lower priority than RUSTY_API_KEY)
RUSTY_API_KEY	API key (higher priority)
OPENAI_BASE_URL	API base URL
RUST_LOG	Logging level (debug, info, warn)

Resolution Order

Settings are resolved in this order (later wins):

1. Preset defaults
2. ~/.rusty/settings.json
3. Environment variables
4. CLI flags