tokenwise

Find out what burned tokens in a Claude Code session and how to cut it. Runs a local, zero-dependency analyzer over the session transcript (main + subagents) and reports the biggest token drivers — large tool results that re-bill on every later turn, redundant file reads, subagent spend, cache efficiency — then turns that into concrete, prioritized fixes. Use when the user asks why a session/agent was expensive, wants to reduce token or context usage, make an agent cheaper, or "what used all my tokens?". Read-only.