Function EstimateMessageTokens

pkg/compaction/compaction.go:41–64

EstimateMessageTokens returns a rough token-count estimate for a single chat message based on its text length. This is intentionally conservative (overestimates) so that proactive compaction fires before we hit the limit. The estimate accounts for message content, multi-content text parts, reasonin

(msg *chat.Message)

// reasoning content, tool call arguments, and a small per-message overhead
// for role/metadata tokens.
func EstimateMessageTokens(msg *chat.Message) int64 {
	// charsPerToken: average characters per token. 4 is a widely-used
	// heuristic for English; slightly overestimates for code/JSON (~3.5).
	const charsPerToken = 4

	// perMessageOverhead: role, ToolCallID, delimiters, etc.
	const perMessageOverhead = 5

	var chars int
	chars += len(msg.Content)
	for _, part := range msg.MultiContent {
		chars += len(part.Text)
	}
	chars += len(msg.ReasoningContent)
	for _, tc := range msg.ToolCalls {
		chars += len(tc.Function.Arguments)
		chars += len(tc.Function.Name)
	}

	if chars == 0 {
		return perMessageOverhead
	}
	return int64(chars/charsPerToken) + perMessageOverhead
}

// SplitIndexForKeep walks messages from the end and returns the earliest
// index whose suffix fits in maxTokens, snapping to user/assistant