Why do some runs show as unmeasured?

They did not report their tokens, usually because they ran before Pandore tracked usage or because the CLI does not report it. Pandore labels them instead of counting them as zero.

Does it work with Codex and Gemini CLI?

Yes, they appear in the same dashboard with what each CLI reports. Codex reports tokens but not a cost; Gemini CLI runs are counted as unmeasured.

It is the cost Claude Code reports for each run. Pandore does not apply its own price list.

How to track Claude Code usage and cost by project

Claude Code tells you what a run cost when it ends. This guide shows how to keep that number, attach it to the right project and client, and read it in charts, in about five minutes.

October 19, 2026 · 3 min read · by Dorian Richard

Pandore AI usage dashboard with donut charts by model, engine and type of work, shown with fictional data — The AI usage dashboard after a few weeks of work. Shown with fictional data.

What you will set up

By the end of this guide, every Claude Code run you start from Pandore will be recorded with its tokens, its reported cost, and the project, client, agent and model behind it. You will also have a budget that warns you before a project overspends.

You need Pandore (free, for Mac and Windows) and a working Claude Code account. If Claude Code is not installed yet, Pandore can install it for you.

Step 1: Connect Claude Code

Open Pandore, then Settings › AI.
Choose Claude Code as the remote agent. If it is missing, install it from the same screen.
Check that the status shows it as ready.

Pandore uses your existing Claude Code login. There is nothing to paste and no API key to create.

Step 2: Run an agent inside a project

Usage is only useful when it is attached to something. Start from a project:

Open a project, or create one and link it to its folder on disk.
Go to the project's Agents tab and pick an agent, for example Reviewer.
Add instructions if you want, then click Run.

The agent works directly in the project folder. When the run ends, Pandore stores what Claude Code reported: input, output, cache read and cache write tokens, and the cost.

Step 3: Open the AI usage dashboard

Click AI usage in the sidebar. The page has four parts:

Overview: measured tokens, runs, cache reads and reported cost for the period.
By model, Local or remote, By usage: three donut charts. Hover a slice to see its value in the center.
Over time: a day-by-day chart, split between local and remote work.
Model details: each model with its exact identifier, its runs, its token mix and its reported cost.

Use the period switch at the top (7 days, 30 days, 90 days, All) and the Measure by switch to count in tokens, runs or cost.

Pandore AI usage chart over time and model details table, shown with fictional data — Over time and model details. Shown with fictional data.

Step 4: Find what costs the most

Three readings answer most questions:

Switch Measure by to Cost. The By model chart now shows which model spends your money. Larger models cost more per token, so a few runs on a large model can outweigh many runs on a smaller one.
Look at the token mix bar in Model details. A bar dominated by cache reads is cheap; a bar dominated by input means the agent receives a lot of new context on every call.
Scroll to the project and client rankings. The FinOps signal at the top of the page also names the project that concentrates the most tokens.

Step 5: Set a budget

Click Budgets at the top of the page.
Set a token limit, a cost limit, or both, for the whole workspace, a client or a project.
Choose the warning threshold, for example 80%.

Pandore flags a budget when usage nears the threshold and when it is exceeded. If you are not sure how to split a global budget, the suggested allocation follows each project's observed usage while keeping a share for quieter projects.