Agent Report Card

Turn local coding-agent transcripts into a daily cost and cache-hit scorecard

Problem

Coding-agent sessions vary wildly in cost. DeepSeek Harness surfaces live cache-hit rates so users can track spend, but Claude Code and Codex offer no such view — so devs only discover waste when the bill arrives, with no idea which prompts burned context.

Solution

Parse local agent session logs (cache hits, input/output tokens, retries) into per-day and per-project cost and hit-rate reports, plus actionable hints: which sessions to split, which repeated context is burning money. Everything runs locally; transcripts never leave the machine.

Users

Heavy devs running multiple coding agents, indie hackers paying for their own AI coding subscriptions, and tech leads at small teams

MVP

Log parsers for 2-3 major agents; per-session token cost and cache-hit view; daily scorecard via notification or email; local SQLite with one-click data wipe

Revenue

Free desktop client; Pro one-time license for multi-machine aggregation and team dashboards; later an enterprise compliance reporting tier

Why now

Coding-agent plans are shifting from flat monthly fees to metered quotas, making users far more cost-sensitive while clients still lack session-level spend views