Lake Meter

Optimize AI agent spending with per-step model selection, run caps, and detailed cost reporting for Claude Code.

Lake Meter helps developers manage costs and token consumption in Claude Code through per-step model recommendations, execution safeguards, and detailed financial receipts. Operating entirely locally with zero telemetry, it integrates directly via Claude Code hooks, MCP, and statusline extensions. During plan mode, Lake Meter estimates step complexity to recommend the most cost-effective model tier, while an interactive spending cap pauses execution before tool calls if budget thresholds are exceeded. It provides precise post-run receipts comparing measured usage to original estimates, offering developers fine-grained control and visibility over local agentic workflows.

主な機能

01Configurable run and session caps that prompt for permission before overspending
02End-of-run receipts detailing token and dollar usage across primary models and subagents
03Per-step model recommendations that match task complexity to Haiku, Sonnet, or Opus tiers
04MCP server interface enabling plan estimation within tools like Cursor
05Real-time statusline metrics showing current run spend, session spend, and 5-hour rate limits
060 GitHub stars

ユースケース

01Preventing accidental API spend overruns caused by runaway subagent loops in Claude Code
02Forecasting token budgets and model costs directly within Cursor or markdown plans
03Optimizing agent workflow expenses by delegating simpler plan steps to lighter models