L2: Incremental session memory compaction — session_memory_compact • codeagent

Summarises early turns while retaining recent context. Keeps between min_tokens and max_tokens in the summary.

Usage

session_memory_compact(
  chat,
  model = .HAIKU_MODEL,
  min_messages = 5L,
  min_tokens = .COMPACT_L2_MIN_TOKENS,
  max_tokens = .COMPACT_L2_MAX_TOKENS,
  pending_turn = NULL
)

Arguments

chat: An ellmer::Chat object.
model: Character. Haiku model for summarisation.
min_messages: Integer. Minimum number of text messages to keep.
min_tokens: Integer. Minimum tokens to retain.
max_tokens: Integer. Maximum tokens for the summary section.
pending_turn: Optional outgoing pending turn used for pairing validation.

Value

Invisibly TRUE after a validated history write, otherwise NULL.