Track engineering output
See code output, pull request cost, and tool reliability for your engineering teams on the Engineering page.
The Engineering page connects AI coding spend to the code it produced: lines changed, commits, pull requests, and which tools fail most often.
Who can see this: Admins.
Pick a range in the top right: Last 7 days, Last 30 days, or Last 90 days.
Stats
| Stat | What it shows |
|---|---|
| Lines added and Lines removed | Lines of code changed with Claude Code. |
| Commits | Commits made, with the number of pull requests under it. |
| Active hours | Time spent actively working in Claude Code. |
| Edit acceptance | The share of suggested edits people accepted, with the total number of edits. |
| PR attributed cost | AI spend attributed to pull requests in the range. |
These numbers come from Claude Code. A stat shows "—" when Tokenize didn't receive that measurement.
Tabs
Pull Requests
Lists pull requests linked to Claude Code sessions, with Attributed cost, Usage, and Last Claude activity. This tab covers Claude Code sessions only.
How pull requests get here. When Claude Code opens or links a pull request during a session, Tokenize links that pull request to the session. No other setup is needed, but the person must be using Claude Code with the desktop app installed. See Install Tokenize on your computer.
When a session touches more than one pull request, Tokenize splits the session's cost between them.
If some requests couldn't be priced, a warning says the dollar totals count them as $0 and are understated. See How cost is calculated.
If nothing is linked yet, the tab shows "No linked pull requests".
Languages
Edits by programming language, with Accepted, Rejected, and Accept rate.
Branches, Skills, and Effort
Spend broken down by git branch, by skill, or by reasoning effort level, with Spend, Requests, and Sessions for each. Spend that isn't tied to a branch, skill, or effort level is shown on its own row.
Tools
- Tool reliability shows each tool's calls and failures.
- Failures by user shows who hits the same failures, with Calls, Failures, Fail rate, and Last failure. Repeated failures for one person often mean a broken credential or query, especially on background agents.
- API errors by user lists errors returned by the AI provider, by Error type.
Codex
Codex usage from OpenAI's ChatGPT Enterprise analytics. This covers every Codex client, including Codex web and the ChatGPT apps, not only sessions on enrolled computers.
- Accepted lines by user shows Turns, Threads, Lines added, and Lines removed for each person. When OpenAI reports commit data, the table adds Committed lines and Codex commits.
- Client mix shows Users, Threads, Turns, and Tokens for each Codex client, such as Codex CLI, Codex Desktop, or the JetBrains extension.
This tab needs an OpenAI Enterprise key in Settings → Providers. Without one, it shows "No Codex Enterprise usage in this period." See Connect OpenAI.