pandev-metriks/pandev-cli
A CLI tool that breaks down what your AI coding agent actually costs, by task instead of by day
pandev reads local log files left behind by AI coding agents like Claude Code and Codex CLI and calculates how much each task, branch, model, or file actually cost. Everything runs on your own machine with no account and no data sent out. It launches with a single command, npx pandev, with stable releases for macOS and Linux and an experimental installer for Windows.
What it does
- The problem it targets: AI coding tools bill by token, but nothing connects that spend to the actual task or ticket someone was working on
- It found that Claude Code always writes 'HEAD' instead of a real branch name into its logs, which can inflate a single task's cost by up to 70x if taken at face value — pandev fixes this by reconstructing the real branch at each point in time from git reflog
- Commands like pandev task, pandev files, and pandev models show cost broken down by task, file, and model along with cache usage patterns, and pandev web launches a local-only dashboard at 127.0.0.1
- It reads raw prompt text to display it, but never stores or transmits it, and the binary itself makes no outbound network calls
- An early-stage project with 8 GitHub stars; the source is proprietary-licensed while the executable is distributed via npm
Why it matters
For developers or team leads using AI coding agents day to day, this turns a single opaque bill into a breakdown of which specific task drove the cost and why, making it actually usable for budgeting and prompt improvement. It also documents how to independently verify that no data leaves the machine, which matters for a tool handling sensitive code and prompt data.
Terms in this repo
- token · the basic unit AI models process text in; most AI service pricing is based on token count
- git reflog · a git log that records, in time order, changes to references such as branch switches
- loopback · an address like 127.0.0.1 that is only reachable from the same machine, never from an external network
- cache read/write · cost categories that occur when an AI model reuses previously processed content, priced differently from fresh input or output
Repository description (English)
by task, branch, model and file. Runs locally, sends nothing.
Open on GitHubTrending repos
- vorssaint/vorssaint-utilsOne free menu bar app replaces a dozen paid Mac utilities
- Alishahryar1/free-claude-codeA local proxy that lets coding AI agents run on 49 free or cheap model providers instead of one paid service
- freestylefly/awesome-gpt-image-2A library of 532 reverse-engineered prompts that turn GPT-Image2 into a predictable image-making tool
- block/buzzAn open-source workspace where humans and AI agents chat, code, and review in the same rooms
- NousResearch/hermes-agentNous Research's Hermes is an AI agent that gets smarter the more you use it
- virgiliojr94/book-to-skillA tool that turns technical book PDFs into on-demand reference skills for AI coding agents
- VoltAgent/awesome-agent-skillsA single hub collecting over 1000 'how-to' manuals that make AI coding assistants act like experts
- anthropics/claude-plugins-communityA shared shelf where anyone's Claude add-ons get listed for install