Struct OrchestratorConfig 

pub struct OrchestratorConfig {

    pub max_turns: usize,
    pub max_tokens: u32,
    pub context_strategy: Option<ContextStrategyConfig>,
    pub summarize_threshold: Option<u32>,
    pub tool_timeout_seconds: Option<u64>,
    pub max_tool_output_bytes: Option<usize>,
    pub run_timeout_seconds: Option<u64>,
    pub enable_squads: Option<bool>,
    pub reasoning_effort: Option<String>,
    pub enable_reflection: Option<bool>,
    pub tool_output_compression_threshold: Option<usize>,
    pub max_tools_per_turn: Option<usize>,
    pub tool_profile: Option<String>,
    pub max_identical_tool_calls: Option<u32>,
    pub max_fuzzy_identical_tool_calls: Option<u32>,
    pub max_tool_calls_per_turn: Option<u32>,
    pub dispatch_mode: Option<DispatchMode>,
    pub routing: RoutingMode,
    pub escalation: bool,
    pub multi_agent_prompt: Option<bool>,
    pub spawn: Option<SpawnConfig>,
    pub max_tokens_in_flight_per_tenant: Option<usize>,
}

Orchestrator-level settings with sensible defaults.

Fields

max_turns: usize

max_tokens: u32

context_strategy: Option<ContextStrategyConfig>

Context window management strategy for the orchestrator’s own conversation.

summarize_threshold: Option<u32>

Token threshold for summarization of the orchestrator’s own context.

tool_timeout_seconds: Option<u64>

Timeout in seconds for the orchestrator’s own tool calls.

max_tool_output_bytes: Option<usize>

Maximum byte size for tool output on the orchestrator’s own tools.

run_timeout_seconds: Option<u64>

Wall-clock deadline in seconds for the entire orchestrator run.

enable_squads: Option<bool>

Enable the form_squad tool for dynamic agent squad formation. When None (default), auto-enabled when there are >= 2 agents. Set to false to disable for a simpler prompt with fewer tokens.

reasoning_effort: Option<String>

Reasoning/thinking effort level. Enables extended thinking on models that support it (e.g., Qwen3 via OpenRouter, Claude with extended thinking). Valid values: “high”, “medium”, “low”, “none”.

enable_reflection: Option<bool>

Enable reflection prompts after tool results. When true, the agent pauses to assess tool outputs before deciding the next action (Reflexion/CRITIC pattern).

tool_output_compression_threshold: Option<usize>

Tool output compression threshold in bytes. Outputs exceeding this size are compressed via an LLM call that preserves factual content.

max_tools_per_turn: Option<usize>

Maximum number of tool definitions sent per LLM turn. When agents have many tools, filtering to the most relevant reduces context usage and cost.

tool_profile: Option<String>

Tool profile for pre-filtering tool definitions. Valid values: “conversational”, “standard”, “full”. Defaults to no filtering.

max_identical_tool_calls: Option<u32>

Maximum consecutive identical tool-call turns before doom loop detection triggers. When reached, tool calls get error results instead of executing.

max_fuzzy_identical_tool_calls: Option<u32>

Maximum consecutive fuzzy-identical tool-call turns before doom loop detection. Fuzzy matching compares sorted tool names (ignoring inputs).

max_tool_calls_per_turn: Option<u32>

Maximum number of tool calls allowed in a single LLM turn. When a turn contains more tool calls than this limit, the excess calls are rejected with an error result (per-turn cap, not cumulative).

dispatch_mode: Option<DispatchMode>

Dispatch mode for orchestrator delegation. When Sequential, the delegate_task schema constrains maxItems: 1 so the LLM dispatches one agent at a time. Defaults to Parallel when absent.

routing: RoutingMode

Task routing strategy: auto (default), always_orchestrate, single_agent. auto uses heuristic scoring + capability matching to route simple tasks to a single agent and complex tasks to the orchestrator.

escalation: bool

Escalate from single-agent to orchestrator on failure. Default: true. When a single-agent run fails with MaxTurnsExceeded, doom loop, or excessive compaction, the task is re-run through the orchestrator.

multi_agent_prompt: Option<bool>

Append the multi-agent collaboration prompt to sub-agent system prompts. Teaches sub-agents blackboard protocol, dedup, cross-verification, and structured execution. Default: true.

spawn: Option<SpawnConfig>

Dynamic agent spawning configuration. When present, enables the spawn_agent tool on the orchestrator, allowing the LLM to create specialist agents at runtime.

max_tokens_in_flight_per_tenant: Option<usize>

Per-tenant in-flight token cap for the TenantTokenTracker. When None, in-flight token tracking is disabled (effectively unbounded). Must be > 0 when set.

Trait Implementations

impl Debug for OrchestratorConfig

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

Formats the value using the given formatter.

impl Default for OrchestratorConfig

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for OrchestratorConfig

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

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.