alexgladkov/quickai

quickai is a profiling tool for Claude Code, the command line AI coding assistant. It reads the transcript log files that Claude Code already writes to disk on a person's own machine and turns them into reports showing where tokens, time, and effective cost went, broken down by task, project, subagent, or model. Nothing is sent anywhere else and no extra tracking is added to Claude Code itself, the tool simply reads files that already exist locally. Because most Claude Code users pay a flat subscription rather than per token, the tool leads with token counts and time rather than a bill. Any dollar figures it shows are described as an api equivalent value, meaning what those tokens would have cost on pay as you go pricing, used as a rough measure of volume and value rather than an actual charge. After installing the tool through Homebrew, as a Claude Code plugin, or by building it from source with Rust, a user first runs an indexing command that scans their transcript files into a local database, which can be rebuilt or refreshed at any time. From there, a set of command line subcommands can show an overall summary, list the biggest tasks by tokens or cost, find unusually slow tasks and how much of that time was spent idle versus actively working, check tool error rates, spot repeated wasted turns, and generate a full HTML report that opens in a browser with sortable tables and drill down views into each task's subagents. The project can also run as an MCP server, which lets a person ask questions like which project used the most tokens this week directly inside a chat conversation, rather than using the command line. The README notes that the tool currently only supports Claude Code, though its core data model is described as general enough that support for other coding assistants could be added later. It is released under the MIT license.