Struct AgentConfig 

pub struct AgentConfig {

    pub provider: String,
    pub api_key_env: String,
    pub default_model: String,
    pub theme: String,
    pub system_prompt_mode: SystemPromptMode,
    pub tool_documentation_mode: ToolDocumentationMode,
    pub enable_split_tool_results: bool,
    pub todo_planning_mode: bool,
    pub ui_surface: UiSurfacePreference,
    pub max_conversation_turns: usize,
    pub reasoning_effort: ReasoningEffortLevel,
    pub verbosity: VerbosityLevel,
    pub temperature: f32,
    pub refine_temperature: f32,
    pub enable_self_review: bool,
    pub max_review_passes: usize,
    pub refine_prompts_enabled: bool,
    pub refine_prompts_max_passes: usize,
    pub refine_prompts_model: String,
    pub small_model: AgentSmallModelConfig,
    pub onboarding: AgentOnboardingConfig,
    pub project_doc_max_bytes: usize,
    pub instruction_max_bytes: usize,
    pub instruction_files: Vec<String>,
    pub custom_api_keys: BTreeMap<String, String>,
    pub credential_storage_mode: AuthCredentialsStoreMode,
    pub checkpointing: AgentCheckpointingConfig,
    pub vibe_coding: AgentVibeCodingConfig,
    pub max_task_retries: u32,
    pub harness: AgentHarnessConfig,
    pub include_temporal_context: bool,
    pub temporal_context_use_utc: bool,
    pub include_working_directory: bool,
    pub user_instructions: Option<String>,
    pub default_editing_mode: EditingMode,
    pub require_plan_confirmation: bool,
    pub autonomous_mode: bool,
    pub circuit_breaker: CircuitBreakerConfig,
    pub open_responses: OpenResponsesConfig,
}

Agent-wide configuration

Fields

provider: String

AI provider for single agent mode (gemini, openai, anthropic, openrouter, xai, zai)

api_key_env: String

Environment variable that stores the API key for the active provider

default_model: String

Default model to use

theme: String

UI theme identifier controlling ANSI styling

system_prompt_mode: SystemPromptMode

System prompt mode controlling verbosity and token overhead Options: minimal (~500-800 tokens), lightweight (~1-2k), default (~6-7k), specialized (~7-8k) Inspired by pi-coding-agent: modern models often perform well with minimal prompts

tool_documentation_mode: ToolDocumentationMode

Tool documentation mode controlling token overhead for tool definitions Options: minimal (~800 tokens), progressive (~1.2k), full (~3k current) Progressive: signatures upfront, detailed docs on-demand (recommended) Minimal: signatures only, pi-coding-agent style (power users) Full: all documentation upfront (current behavior, default)

enable_split_tool_results: bool

Enable split tool results for massive token savings (Phase 4) When enabled, tools return dual-channel output:

• llm_content: Concise summary sent to LLM (token-optimized, 53-95% reduction)
• ui_content: Rich output displayed to user (full details preserved) Applies to: grep_file, list_files, read_file, run_pty_cmd, write_file, edit_file Default: true (opt-out for compatibility), recommended for production use

todo_planning_mode: bool

Enable TODO planning helper mode for structured task management

ui_surface: UiSurfacePreference

Preferred rendering surface for the interactive chat UI (auto, alternate, inline)

max_conversation_turns: usize

Maximum number of conversation turns before auto-termination

reasoning_effort: ReasoningEffortLevel

Reasoning effort level for models that support it (none, low, medium, high) Applies to: Claude, GPT-5, GPT-5.1, Gemini, Qwen3, DeepSeek with reasoning capability

verbosity: VerbosityLevel

Verbosity level for output text (low, medium, high) Applies to: GPT-5.1 and other models that support verbosity control

temperature: f32

Temperature for main LLM responses (0.0-1.0) Lower values = more deterministic, higher values = more creative Recommended: 0.7 for balanced creativity and consistency Range: 0.0 (deterministic) to 1.0 (maximum randomness)

refine_temperature: f32

Temperature for prompt refinement (0.0-1.0, default: 0.3) Lower values ensure prompt refinement is more deterministic/consistent Keep lower than main temperature for stable prompt improvement

enable_self_review: bool

Enable an extra self-review pass to refine final responses

max_review_passes: usize

Maximum number of self-review passes

refine_prompts_enabled: bool

Enable prompt refinement pass before sending to LLM

refine_prompts_max_passes: usize

Max refinement passes for prompt writing

refine_prompts_model: String

Optional model override for the refiner (empty = auto pick efficient sibling)

small_model: AgentSmallModelConfig

Small/lightweight model configuration for efficient operations Used for tasks like large file reads, parsing, git history, conversation summarization Typically 70-80% cheaper than main model; ~50% of VT Code’s calls use this tier

onboarding: AgentOnboardingConfig

Session onboarding and welcome message configuration

project_doc_max_bytes: usize

Maximum bytes of AGENTS.md content to load from project hierarchy

instruction_max_bytes: usize

Maximum bytes of instruction content to load from AGENTS.md hierarchy

instruction_files: Vec<String>

Additional instruction files or globs to merge into the hierarchy

custom_api_keys: BTreeMap<String, String>

Provider-specific API keys captured from interactive configuration flows

credential_storage_mode: AuthCredentialsStoreMode

Preferred storage backend for credentials (OAuth tokens, API keys, etc.)

• keyring: Use OS-specific secure storage (macOS Keychain, Windows Credential Manager, Linux Secret Service). This is the default as it’s the most secure.
• file: Use AES-256-GCM encrypted file with machine-derived key
• auto: Try keyring first, fall back to file if unavailable

checkpointing: AgentCheckpointingConfig

Checkpointing configuration for automatic turn snapshots

vibe_coding: AgentVibeCodingConfig

Vibe coding configuration for lazy or vague request support

max_task_retries: u32

Maximum number of retries for agent task execution (default: 2) When an agent task fails due to retryable errors (timeout, network, 503, etc.), it will be retried up to this many times with exponential backoff

harness: AgentHarnessConfig

Harness configuration for turn-level budgets, telemetry, and execution limits

include_temporal_context: bool

Include current date/time in system prompt for temporal awareness Helps LLM understand context for time-sensitive tasks (default: true)

temporal_context_use_utc: bool

Use UTC instead of local time for temporal context in system prompts

include_working_directory: bool

Include current working directory in system prompt (default: true)

user_instructions: Option<String>

Custom instructions provided by the user via configuration to guide agent behavior

default_editing_mode: EditingMode

Default editing mode on startup: “edit” (default) or “plan” Codex-inspired: Encourages structured planning before execution.

require_plan_confirmation: bool

Require user confirmation before executing a plan generated in plan mode When true, exiting plan mode shows the implementation blueprint and requires explicit user approval before enabling edit tools.

autonomous_mode: bool

Enable autonomous mode - auto-approve safe tools with reduced HITL prompts When true, the agent operates with fewer confirmation prompts for safe tools.

circuit_breaker: CircuitBreakerConfig

Circuit breaker configuration for resilient tool execution Controls when the agent should pause and ask for user guidance due to repeated failures

open_responses: OpenResponsesConfig

Open Responses specification compliance configuration Enables vendor-neutral LLM API format for interoperable workflows

Implementations

impl AgentConfig

pub fn validate_llm_params(&self) -> Result<(), String>

Validate LLM generation parameters

Trait Implementations

impl Clone for AgentConfig

fn clone(&self) -> AgentConfig

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for AgentConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for AgentConfig

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for AgentConfig

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl JsonSchema for AgentConfig

fn schema_name() -> Cow<'static, str>

The name of the generated JSON Schema.

fn schema_id() -> Cow<'static, str>

Returns a string that uniquely identifies the schema produced by this type.

fn json_schema(generator: &mut SchemaGenerator) -> Schema

Generates a JSON Schema for this type.

fn inline_schema() -> bool

Whether JSON Schemas generated for this type should be included directly in parent schemas, rather than being re-used where possible using the $ref keyword.

impl Serialize for AgentConfig

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.