Struct AgentConfig 

pub struct AgentConfig {

    pub default_model: String,
    pub context_window: usize,
    pub reserve_tokens: usize,
    pub max_tokens: usize,
    pub max_spawn_depth: Option<u8>,
    pub subagent_model: Option<String>,
    pub fallback_models: Vec<String>,
    pub max_tool_repeats: usize,
    pub max_tool_errors: u32,
    pub tool_retry_on_malformed: bool,
    pub session_max_age: u64,
    pub session_max_count: usize,
    pub post_compaction_sections: Vec<String>,
    pub checkpoints_enabled: bool,
    pub max_checkpoints: usize,
    pub active_memory: ActiveMemoryConfig,
    pub permission_level: PermissionLevel,
    pub auto_approve_loopback: bool,
}

Fields

default_model: String

context_window: usize

reserve_tokens: usize

max_tokens: usize

Maximum tokens for LLM response

max_spawn_depth: Option<u8>

Maximum depth for spawn_agent tool (default: 1, no nested spawning)

• 0: spawn_agent tool disabled
• 1: single level only (subagents cannot spawn more agents)
• 2+: limited nesting allowed (not recommended)

subagent_model: Option<String>

Model to use for spawned subagents (default: same as default_model or claude-cli/sonnet)

fallback_models: Vec<String>

Fallback models to try if primary provider fails with retryable errors (rate limits, server errors, timeouts). Providers are tried in order. Example: [“openai/gpt-4o”, “ollama/llama3”]

max_tool_repeats: usize

Maximum times the same tool can be called with identical arguments before loop detection triggers. Default: 3. Set to 0 to disable loop detection.

max_tool_errors: u32

Maximum consecutive failures of the same tool before blocking it. Default: 3. Set to 0 to disable tool error tracking.

tool_retry_on_malformed: bool

Allow one retry when a tool call has malformed arguments. Default: true. Set to false to fail immediately.

session_max_age: u64

Maximum age for session files before pruning (in seconds). 0 = keep forever. Default: 30 days.

session_max_count: usize

Maximum number of sessions to keep per agent. 0 = unlimited. Default: 500.

post_compaction_sections: Vec<String>

Sections from AGENTS.md (or SOUL.md) to re-inject after session compaction. Default: [“Session Startup”, “Red Lines”]. Empty array disables injection.

checkpoints_enabled: bool

Save checkpoint of session transcript before compaction (enables restore). Default: true.

max_checkpoints: usize

Maximum compaction checkpoints to keep per session. Default: 5.

active_memory: ActiveMemoryConfig

Active memory recall configuration — automatically search memory before replies

permission_level: PermissionLevel

Session permission level. Tools requiring a higher level need approval. Default: “elevated” (all tools run without prompting, matching current behavior). Set to “safe” to require approval for dangerous tools (bash, file write, etc.).

auto_approve_loopback: bool

Auto-approve elevated tool calls from loopback connections (localhost). Default: true (backward compatible — CLI and local HTTP skip approval).

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 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.