Struct ContextConfig 

pub struct ContextConfig {

    pub snip_threshold: f64,
    pub micro_threshold: f64,
    pub collapse_threshold: f64,
    pub auto_threshold: f64,
    pub renewal_threshold: f64,
    pub target_after_compress: f64,
    pub snip_per_msg_ratio: f64,
    pub carryover_ratio: f64,
    pub recovery_content_ratio: f64,
    pub preserve_recent_msgs: usize,
    pub preserve_recent_turns: usize,
    pub render_dashboard: bool,
    pub verbose_control_notes: bool,
    pub collapse_assistant_narration: bool,
    pub micro_compact_idle_minutes: u32,
    pub preserved_tool_results: usize,
    pub autocompact_buffer: u32,
    pub spool_threshold_bytes: u32,
    pub spool_preview_bytes: u32,
}

All compression and context management parameters expressed as fractions of max_tokens. This is the single control surface for the compression pipeline: changing max_tokens (e.g. switching model) rescales every derived limit automatically with no other configuration change required.

Invariant: snip < micro < collapse < auto < renewal (strictly increasing).

Fields

snip_threshold: f64

micro_threshold: f64

collapse_threshold: f64

auto_threshold: f64

renewal_threshold: f64

target_after_compress: f64

Target rho after any compression pass. Must be < snip_threshold.

snip_per_msg_ratio: f64

Fraction of max_tokens any single message may occupy after SnipCompact. Messages smaller than this are never touched.

carryover_ratio: f64

Fraction of max_tokens worth of history tokens to carry across renewal. Renewal stops carrying messages once this token budget is exhausted.

recovery_content_ratio: f64

Maximum fraction of max_tokens a recovery/replay payload may occupy.

preserve_recent_msgs: usize

Recent history messages always kept during render.

preserve_recent_turns: usize

Number of most-recent turns (user+assistant pairs) preserved by CollapseCompactor and AutoCompactor. Each turn = 2 messages, so the actual message count kept is preserve_recent_turns * 2. Must be ≥ 1. Default: 2 (= 4 messages).

render_dashboard: bool

Include the dashboard block in the rendered system context. Defaults to false; enable only in explicit agent-os mode.

verbose_control_notes: bool

Use verbose internal control notes (e.g. “[SYSTEM] Transaction rollback: …”). Defaults to false; uses concise natural-language notes instead.

collapse_assistant_narration: bool

Collapse the narration text of OLD assistant turns (those past the preserve_recent_msgs window that also carry tool calls) to a short stub at render time — non-destructively (the full text stays in partitions.history). The model’s user-facing preamble (“好的，我来…先X”) has no value once it has aged out of the recent window, but re-feeding it verbatim every turn primes the model to keep emitting the same preamble (an in-context repetition trap). Tool calls and pairing are untouched; current progress lives in the TASK STATE turn. Defaults to true.

micro_compact_idle_minutes: u32

Minutes of inactivity before triggering Micro-Compact (Layer 3). Defaults to 60 minutes — assumes Prompt Cache has expired by then.

preserved_tool_results: usize

Number of recent tool results to preserve during Micro-Compact.

autocompact_buffer: u32

Buffer size for Auto-Compact trigger (Layer 5). Trigger threshold = max_tokens - autocompact_buffer. Defaults to 13K tokens (p99.99 of summarizer output length + safety margin).

spool_threshold_bytes: u32

Byte size above which a single tool result is spooled (Layer 1): the kernel keeps only a preview in context and emits LargeResultSpooled for the SDK to persist the full content to disk. Default: 50 KiB. 0 disables spooling.

spool_preview_bytes: u32

Preview byte budget kept in context when a tool result is spooled. Default: 2 KiB.

Implementations

impl ContextConfig

pub fn target_tokens(&self, max_tokens: u32) -> u32

Token budget to target after a compression pass.

pub fn snip_per_msg_tokens(&self, max_tokens: u32) -> u32

Per-message token cap used by SnipCompact. Floor of 50 ensures very small context windows still get useful output.

pub fn carryover_tokens(&self, max_tokens: u32) -> u32

Token budget for history carryover across renewal.

pub fn recovery_content_tokens(&self, max_tokens: u32) -> u32

Token cap for a single recovery/replay payload.

pub fn autocompact_threshold(&self, max_tokens: u32) -> u32

Auto-Compact trigger threshold (Layer 5). Returns max_tokens - autocompact_buffer (absolute value).

Trait Implementations

impl Clone for ContextConfig

fn clone(&self) -> ContextConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ContextConfig

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

Formats the value using the given formatter.

impl Default for ContextConfig

fn default() -> Self

Returns the “default value” for a type.