Struct AgentConfig 

pub struct AgentConfig {

    pub name: String,
    pub max_tool_iterations: usize,
    pub auto_update_check: bool,
    pub instruction_files: Vec<PathBuf>,
    pub instruction_auto_detect: bool,
    pub max_tool_retries: usize,
    pub tool_repeat_threshold: usize,
    pub max_retry_duration_secs: u64,
    pub focus: FocusConfig,
    pub tool_filter: ToolFilterConfig,
    pub budget_hint_enabled: bool,
    pub supervisor: TaskSupervisorConfig,
    pub time_reminder_enabled: bool,
    pub time_reminder_interval_requests: u32,
}

Core agent behavior configuration, nested under [agent] in TOML.

Controls the agent’s name, tool-loop limits, instruction loading, and retry behavior. All fields have sensible defaults; only name is typically changed by end users.

Example (TOML)

[agent]
name = "Zeph"
max_tool_iterations = 15
max_tool_retries = 3

Fields

name: String

Human-readable agent name surfaced in the TUI and Telegram header. Default: "Zeph".

max_tool_iterations: usize

Maximum number of tool-call iterations per agent turn before the loop is aborted. Must be <= 100. Default: 10.

auto_update_check: bool

Check for new Zeph releases at startup. Default: true.

instruction_files: Vec<PathBuf>

Additional instruction files to always load, regardless of provider.

instruction_auto_detect: bool

When true, automatically detect provider-specific instruction files (e.g. CLAUDE.md for Claude, AGENTS.md for OpenAI).

max_tool_retries: usize

Maximum retry attempts for transient tool errors (0 to disable).

tool_repeat_threshold: usize

Number of identical tool+args calls within the recent window to trigger repeat-detection abort (0 to disable).

max_retry_duration_secs: u64

Maximum total wall-clock time (seconds) to spend on retries for a single tool call.

focus: FocusConfig

Focus-based active context compression configuration (#1850).

tool_filter: ToolFilterConfig

Dynamic tool schema filtering configuration (#2020).

budget_hint_enabled: bool

Inject a <budget> XML block into the volatile system prompt section so the LLM can self-regulate tool calls and cost. Self-suppresses when no budget data is available (#2267).

supervisor: TaskSupervisorConfig

Background task supervisor tuning. Controls concurrency limits and turn-boundary abort.

time_reminder_enabled: bool

Inject a <current_time> reminder into the volatile system prompt block every N agent turns (#6361, spec 070 FR-003). Opt-in — defaults to false so existing prompt content and token budget are unaffected unless explicitly enabled (NFR-005). Complementary to the always-available get_current_time tool, which covers time-awareness within a single long-running turn where this per-turn injection cannot re-fire.

time_reminder_interval_requests: u32

Number of agent turns between <current_time> reminder injections when time_reminder_enabled = true (#6361, spec 070 FR-004). Named after Codex’s reminder_interval_model_requests, but counts agent turn-cycles (sidequest.turn_counter) rather than individual model requests — the mandated injection hook (rebuild_system_prompt) runs once per turn, before the tool loop, so a literal per-model-request cadence is unreachable there.

Trait Implementations

impl Debug for AgentConfig

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

Formats the value using the given formatter.

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

fn deserialize<__D>( __deserializer: __D, ) -> Result<AgentConfig, <__D as Deserializer<'de>>::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for AgentConfig

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

where __S: Serializer,

Serialize this value into the given Serde serializer.