Skip to main content

default_compact_threshold_chars

recursive::llm::pricing

Function default_compact_threshold_chars 

Source 
pub fn default_compact_threshold_chars(model: &str) -> usize
Expand description

Compute the default compaction character-count threshold for a model.

Strategy (mirrors fake-cc getAutoCompactThreshold):

  1. Start from the model’s context window in tokens.
  2. Reserve 20 000 tokens for the compaction summary output.
  3. Take 80 % of the remainder as the trigger point (leaves a comfortable 20 % buffer before the hard limit is hit).
  4. Convert tokens → characters using a conservative 4 chars/token ratio.

Limitation: the 4 chars/token ratio works well for English but significantly underestimates token density in CJK languages (~1 char ≈ 2-3 tokens). Prefer default_compact_threshold_tokens when actual prompt-token counts are available from the API response.