New Claude Code & OpenAI Codex usage dashboard
The Claude Code and Codex usage dashboard: cost by developer, team and project.
Your AI coding bill counts seats. Ganivra prices every request and shows who spent it, on which team, project and client, from one command on each machine.
- Claude Code in the terminal, IDE and desktop app
- Codex CLI and the ChatGPT desktop app
- Prompts and code are never sent
curl -fsSL https://ganivra.com/install.sh | shNo credit card · Works on API billing and on Claude and ChatGPT plans · From $29/month after the trial
- acme-checkoutclient acmeCCCX$412.80
- mobile-appinternalCC$268.15
- data-pipelineclient northwindCX$97.40
- No projectassign in SessionsCX$31.20
At a glance
Claude Code and Codex cost tracking, in short.
Updated
- What it is
- Ganivra tracks what Claude Code and OpenAI Codex cost, priced per request, by developer, team, project, client and model.
- Tools
- Claude Code (terminal, IDE, desktop app), OpenAI Codex CLI and Codex in the ChatGPT desktop app.
- Setup
- One command per machine on macOS and Linux, and one committed .claude/settings.json per repository. Manual settings on Windows.
- Data sent
- Token counts, model, time, session, developer and your labels. Never prompts, responses or source code.
- Plans
- Works on API billing and on Claude and ChatGPT subscriptions, where it shows usage at API list rates.
- Price
- 14-day free trial with no credit card, then from $29 a month.
The gap
Your AI coding bill can't answer the questions you get asked.
Invoices count seats
They say how many seats you pay for, not who used them, which seats sit idle or which team ran up the overage.
Each tool reports on itself
Claude and ChatGPT each have their own console and export. None shows one figure across tools, by team.
Nobody owns the project number
Finance asks what the Acme work cost in AI. Engineering can't say, because no tool knows which repository a request came from.
What you can answer
Every question, one dashboard.
Switch between developers, teams, projects, models, tools and sessions. Every view downloads as CSV.
Who is spending, and on which team?
Every request carries the developer who made it and a team label. See cost, tokens and cost per million tokens by person and team, across both tools.
What does each project and client cost?
Label each repository once. Every request from it carries its project and client, ready for chargeback or a client invoice.
Is the seat worth it?
On a Claude or ChatGPT plan, see what the same usage would cost at API rates, and set it next to the seat price.
Would a cheaper model do?
Re-price the tokens your team used on another model, hosted or on your own GPUs.
- Claude Sonnet 5.5
- Claude Haiku 5.5
- GPT-6.1 Terra
- Gemini 4 Argon
- Kimi K3.1 Pro · hosted
- Your own GPUs
Usage nobody labelled?
Assign desktop-app sessions, or anything from before you set labels, to a project and client in Ganivra.
Setup
One command per machine. One file per repository.
The command finds Claude Code and Codex, asks for your workspace key once, checks it, and backs up their settings before changing them.
- 1
Connect each machine
Every developer runs it once. It configures whichever tools it finds.
curl -fsSL https://ganivra.com/install.sh | sh - 2
Label each repository
Commit the file it writes, and everyone who clones the repository sends the same labels.
ganivra label --project acme-checkout --client acme git add .claude/settings.json - 3
Work as usual
Check the setup, then open Dashboard → Coding agents after a request or two.
ganivra status
Rolling out to a whole team? Claude Code's managed settings can push the export to every machine, and ganivra uninstall removes everything the command added. Read the script first at ganivra.com/ganivra.sh, or follow teams, projects and clients in the guide.
Supported tools
Where it works today.
Claude Code
SupportedDeveloper, team, project and client. Projects can also come from the git repository with no labels.
OpenAI Codex CLI
SupportedDeveloper, team, project and client, read from the same project file Claude Code uses.
Codex in ChatGPT
SupportedDeveloper and team. Assign projects in the Sessions view: the app doesn't report a folder.
GitHub Copilot
PartlyArrives as priced model calls, not yet in the coding-agents view.
Cursor
Not yetCompare its seats in the AI coding tools cost calculator.
Privacy
Costs, not conversations.
Ganivra needs to know who used which model and how many tokens. It never needs what was asked or what was written. When a folder name becomes a project, only the name is kept, never the path.
Sent
- Token counts and model
- Time and session
- Developer email or account ID
- Your team, project and client labels
Never sent
- Prompts, responses, source code
- File contents and diffs
- Full folder paths
Related
Plan the budget before you track it.
Questions
Claude Code and Codex cost tracking FAQ
Something else? The integration guide has every setting.
Is there a Claude Code or Codex usage dashboard for teams?
Claude Code's /cost command shows the current session, and Anthropic's and OpenAI's consoles show organization usage by API key, project or user. Neither breaks usage down by your teams, projects or clients, or puts Claude Code and Codex side by side. Ganivra is that dashboard: every request from both tools, priced and grouped by developer, team, project, client and model.
How do I track Claude Code costs per developer?
Claude Code can export OpenTelemetry from each developer's machine. Run the Ganivra setup command, or add the export settings to ~/.claude/settings.json by hand, and every Claude Code request arrives priced, with the developer's email, model, tokens and session. Ganivra groups it by developer, team, project and model.
Can I track OpenAI Codex usage by team and project?
Yes. Codex exports each completed response with its token counts. The Ganivra setup command adds the export to ~/.codex/config.toml and a small shell function that gives each codex run the project's labels. The ChatGPT desktop app reports the developer and team; you assign its sessions to projects in Ganivra.
Does Ganivra see our prompts or code?
No. Claude Code redacts prompts in its telemetry by default, the Codex settings Ganivra writes turn prompt logging off, and Ganivra would not store prompt or response text anyway. It receives token counts, model, timing, the developer and the labels you set.
Does it work on Claude Pro, Max or Team subscriptions?
Yes. The telemetry is sent whatever plan the developer is on. On a subscription, the dollar figures show what the same usage would cost at API list rates, not your invoice, which is how you judge whether each seat is worth its price.
How are projects detected?
Best: a .claude/settings.json committed to each repository with its team, project and client, which both Claude Code and the codex shell function read, so everyone on the repository sends the same labels. Without it, Claude Code can use the git repository name, and anything left over can be assigned in Ganivra.
How do we charge AI coding costs back to teams or clients?
Group by team, project or client and download the table as CSV, ready for a chargeback sheet. Allocation rules in Reports can also map developers to cost centres. The chargeback guide explains how to reconcile the totals to your invoices.
Does the setup work on Windows?
The one-command setup is for macOS and Linux. On Windows, add the same settings by hand from the integration guide: Claude Code's settings.json and Codex's config.toml.
Know what AI coding costs, by team and project, this week.
Create a workspace, run one command on each machine and label your repositories. 14-day free trial, no credit card, from $29 a month after.
curl -fsSL https://ganivra.com/install.sh | sh