Struct Ai 

pub struct Ai {
    pub backend: Option<String>,
    pub model: Option<String>,
    pub coding_backend: Option<String>,
    pub coding_model: Option<String>,
    pub max_tool_iterations: u32,
    pub max_turn_seconds: u64,
    pub rounds: u32,
    pub skip_knobs: bool,
    pub cross_scheduler_refs: bool,
    pub build_fix_attempts: u32,
    pub runtime_fix_attempts: u32,
}

Fields

backend: Option<String>

Backend for the planner/reasoner role: an OpenAI-compatible API base URL (built-in openai backend), or one of the keywords claude, codex, opencode, cursor-agent (shell out to that agent CLI). Falls back to $SCX_FORGE_BACKEND.

model: Option<String>

Planner/reasoner model for the built-in openai backend.

coding_backend: Option<String>

Backend for the patch/apply (coding) role: an OpenAI-compatible API base URL or one of claude, codex, opencode, cursor-agent. Falls back to $SCX_FORGE_CODING_BACKEND, then to the planner backend. Set this (with coding_model) to run the coding role on a separate backend.

coding_model: Option<String>

Patch/apply model for the built-in openai backend. Empty means use model.

max_tool_iterations: u32

Max tool-calling iterations per LLM turn for the built-in openai backend.

max_turn_seconds: u64

Wall-clock cap per planner or coding model turn.

rounds: u32

Maximum optimization rounds, shared across both phases. The AI-driven knob-tuning phase ends when the model decides it is done; the remaining rounds (up to this cap) go to code changes.

skip_knobs: bool

Skip the AI-driven knob-tuning phase and go straight to code changes.

cross_scheduler_refs: bool

Allow read-only cross-scheduler reference tools (list_schedulers, grep_schedulers, read_scheduler_file) that inspect other scheduler crates under scheds/rust. Disabled by default: the model focuses only on the target crate, which keeps the prompt and tool surface (and thus token usage) smaller. Enable to let the planner port mechanisms from other schedulers.

build_fix_attempts: u32

LLM edit attempts to fix a broken build before reverting the round.

runtime_fix_attempts: u32

LLM edit attempts to fix a runtime scheduler failure before reverting.

Implementations

impl Ai

pub fn turn_timeout(&self) -> Duration

Trait Implementations

impl Clone for Ai

fn clone(&self) -> Ai

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for Ai

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

Formats the value using the given formatter.

impl Default for Ai

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for Ai

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

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.