Module count 

Pure token estimation for LLM requests

Estimation layer built on top of TokenEstimator: counts messages, tools, and multimodal content. Transport-level counting (provider count-tokens API first, estimation as fallback) lives in client::LlmClientImpl::count_tokens_inner; this module never touches HTTP clients or codecs.

Functions

default_estimator

Estimator used by counting helpers.

estimate_image_tokens

Estimate tokens for an image reference.

estimate_message_tokens

Estimate tokens for a single message: content (text / rich blocks), thinking, tool calls, plus a fixed metadata overhead per message.

estimate_messages

Estimate the token count of the current conversation messages.

estimate_request_tokens

Estimate tokens for a full LLM request: messages + tool declarations.

estimate_tool_declarations

Estimate tokens for tool declarations alone (the per-request dynamic overhead that never accumulates into a message-array ledger).