See where every Claude Code token goes.

Why Claude Code usage is hard to follow

Claude Code reports its token usage and cost at the end of each run, in the terminal where it ran. Once you work across several client projects, that information is scattered across sessions you have already closed. You can see a monthly total from your provider, but not which project, client or task consumed it.

Pandore keeps that record for you. Every run started from Pandore is attached to a project, a client and an agent, and its measures are stored locally, in the same file as the rest of your workspace.

What Pandore measures

• Tokens, in detail: input, output, cache reads and cache writes for each run, as the CLI reports them.
• Reported cost: the cost Claude Code declares for the run. Pandore never applies a price list of its own.
• Context: the project, client, agent and model behind each run, and whether it came from an agent session, a Mission or the chat.
• Honest gaps: a run that reported nothing stays marked as unmeasured, instead of counting as zero.

Read it at a glance

A single switch counts everything in tokens, in runs or in reported cost. The table shows each model's exact identifier, such as claude-sonnet-5-5, next to its readable name, and splits its tokens into input, output and cache. A model that mostly re-reads its cache costs far less than one that receives new context on every call, and the bar shows the difference at a glance.

Budgets per workspace, client and project

Set a token or cost budget for the whole workspace, for one client or for one project, with a warning threshold. Pandore shows when a budget is getting close to its limit and when it is exceeded. Missions also have their own spending cap, and pause when they reach it until you decide to continue.

Codex, Gemini CLI and local models in the same view

The dashboard is not limited to Claude Code. Runs from Codex and Gemini CLI appear in the same charts, with the detail each CLI actually reports: Codex reports its tokens but not a cost, and Gemini CLI runs are counted as unmeasured. Pandore's 43 downloadable local models report their tokens and cost nothing per call.