Struct EnrichmentConfig 

pub struct EnrichmentConfig {
    pub enabled: bool,
    pub max_parallel_prefetches: u32,
    pub prefetch_budget_tokens: u32,
    pub prefetch_timeout_ms: u32,
    pub respect_rate_limits: bool,
}

Speculative-execution settings for the Paper 3 enrichment planner.

Off by default. Operators (or tune analyze --auto-enrichment) flip enabled to true once the corpus statistics show that speculation would have paid off. Once enabled, the host enforces the budget and concurrency limits below per turn.

Schema-v4 — added in CURRENT_SCHEMA_VERSION = 4.

Fields

enabled: bool

Master switch. false (default) means the planner runs only in telemetry-only mode: recent_tools is tracked, should_skip can be consulted, but no out-of-band tools/call is dispatched. Flip to true to enable real speculative pre-fetch.

max_parallel_prefetches: u32

Maximum number of speculative pre-fetches the host issues in parallel from a single turn’s EnrichmentPlan. Caps fan-out so a Glob → 12 Read does not melt the API rate-limit. Default: 3 (matches the corpus finding that top-3 prefetch covers > 80% of cited follow-ups).

prefetch_budget_tokens: u32

Token ceiling for the speculative part of one turn — distinct from the per-response budget. EnrichmentPlanner::build_plan reads this when constructing TurnContext. Default: 8000 tokens (~32 kB at the 4-byte-per-token heuristic).

prefetch_timeout_ms: u32

Wall-clock budget the host waits for prefetches before returning the main response. Past this, the prefetch keeps running in the background (its result lands in the dedup cache when it returns) but the LLM gets the main response immediately

• a hint that a prefetch is in flight.

Default: 1000 ms — wide margin so typical Glob/Read can land synchronously, but small enough that a slow API never holds the agent.

respect_rate_limits: bool

Honour [tools.<name>].rate_limit_host when scheduling prefetches. When true, the host counts how many prefetches per class are inflight this turn and skips new ones once the cap is hit. Default: true — the only reason to disable is for a benchmark harness with a known sandbox API.

Trait Implementations

impl Clone for EnrichmentConfig

fn clone(&self) -> EnrichmentConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for EnrichmentConfig

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

Formats the value using the given formatter.

impl Default for EnrichmentConfig

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for EnrichmentConfig

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

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for EnrichmentConfig

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

where __S: Serializer,

Serialize this value into the given Serde serializer.