yakuikai/tokentab

Local-first CLI and web dashboard that reads Claude Code, Codex and Gemini CLI session logs and shows token usage and cost by model, project, day and activity. No account, no API key, nothing leaves your machine

What it solves

Tokentab provides a way to track and analyze token usage and costs for various AI coding tools. It solves the problem of fragmented logs across different tools (like Claude Code, Codex, and Gemini CLI) by aggregating them into a single, local dashboard that breaks down spending by model, project, and date.

How it works

The tool reads existing session logs stored on your local disk. It does not require an API key or an account. It calculates costs using a locally maintained pricing table for different models, employing fuzzy matching to handle model versioning. To categorize activity (e.g., coding, debugging), it uses deterministic guesses based on tool usage and the initial session message.

Who it’s for

Developers who use multiple AI coding assistants and want to monitor their spending and token consumption without sending their data to a third-party service.

Highlights

  • Local-first approach: Runs entirely on your machine with no network requests, no API keys, and no external CDN fonts.
  • Crosstool aggregation: Supports Claude Code, Codex, and Gemini CLI out of the box.
  • Flexible reporting: Offers both a CLI and a web dashboard for viewing usage statistics.
  • Extensible architecture: New tools can be added by creating a single provider module that returns a standardized usage record.
  • Zero-dependency server: The web dashboard is built using Python's standard library HTTP server.

Related

  • Project
    tomyud1/godot-mcp

    A Model Context Protocol (MCP) server that gives AI assistants direct access to the Godot editor to read, write, and manipulate scenes, scripts, and project settings.

  • Project
    cristicretu/diri

    A native desktop application that allows developers to run and coordinate dozens of AI coding agents in parallel, providing isolated Git worktrees and a unified review interface.

  • Project
    Ancienttwo/repo-harness

    A file-backed workflow and runtime for AI agents that replaces chat memory with repository-based state, enabling resumable sessions and token-efficient development.

  • Project
    FujitsuResearch/OneCompression

    OneComp is a Python library that automatically compresses large language models by selecting per‑layer bit‑widths, applying advanced post‑training quantization (QEP, LPCD, AutoBit, JointQ, MDBF, etc.), and optionally fine‑tuning with LoRA. It supports Llama, Qwen, Gemma, GPT‑OSS, and more, and can serve the quantized models with vLLM and a local web UI. Install via pip (or the `uv` tool for developers) and run with a single CLI command or the `Runner.auto_run` API.