Struct CompactionConfig 

pub struct CompactionConfig {
    pub compact_at_pct: f64,
    pub compact_budget_threshold_pct: f64,
    pub compaction_scope: CompactionScope,
    pub keep_first_turns: usize,
    pub keep_recent_turns: usize,
    pub max_summary_tokens: usize,
    pub tool_output_max_lines: usize,
    pub focus_message: Option<String>,
    pub in_memory_strategy: Option<Arc<dyn CompactionStrategy>>,
    pub block_strategy: Option<Arc<dyn BlockCompactionStrategy>>,
}

Full compaction policy — controls both WHEN and HOW to compact.

Fields

compact_at_pct: f64

Fraction of max_context_tokens below which headroom is measured. Compaction fires when headroom drops below compact_budget_threshold_pct. Default: 0.90 (90%).

compact_budget_threshold_pct: f64

Minimum remaining headroom fraction before compaction fires. Default: 0.05 (5%). With defaults at 100k/4k: fires at ~81k tokens.

compaction_scope: CompactionScope

Scope controlling how many earlier loops to compact and load. Default: FixedCount(3).

keep_first_turns: usize

Turns to keep verbatim from the start (most recent loop only). Default: 2.

keep_recent_turns: usize

Minimum turns to keep from the end (most recent loop only). Extended to turn boundary so ToolCall/ToolResult pairs are never split. Default: 10.

max_summary_tokens: usize

Token budget for the summarised middle section. Default: 2_000.

This is a budget, not a per-turn limit. Implementations of BlockCompactionStrategy::keep_compacted() should aim to summarise ALL turns in the range within this budget — e.g. by producing shorter per-turn summaries or an LLM-generated holistic digest. DefaultBlockCompaction is a basic implementation that generates per-turn one-liners and drops remaining turns when the budget runs out.

tool_output_max_lines: usize

Max lines per tool output in the keep_recent section. Default: 50.

focus_message: Option<String>

Optional focus message to guide compaction summarization. When set, prepended to the compacted section to tell the model what to prioritize. Example: “Focus on specification details, API contracts, and architectural decisions.”

in_memory_strategy: Option<Arc<dyn CompactionStrategy>>

Custom in-memory compaction strategy. When set, replaces DefaultCompaction. Used when AgentContext.session is None (sub-agents, tests, sessionless runs).

block_strategy: Option<Arc<dyn BlockCompactionStrategy>>

Block-based compaction strategy for Session-aware compaction. When set, replaces DefaultBlockCompaction. Used when AgentContext.session is Some.

Trait Implementations

impl Clone for CompactionConfig

fn clone(&self) -> CompactionConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for CompactionConfig

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

Formats the value using the given formatter.

impl Default for CompactionConfig

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for CompactionConfig

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

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for CompactionConfig

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

where __S: Serializer,

Serialize this value into the given Serde serializer.