pi · Claude Code · CLI
Know how long your agent worked on each client, project and task. And what it cost you.
Quick start
npm install -g kankaku
kankaku setupThen run kankaku to open the dashboard.
$ kankaku today site tasks 3 wall 32m work 24m wait 8m $0.82 total tasks 3 wall 32m work 24m wait 8m $0.82
Illustrative kankaku CLI example
1. Install
Two commands: install, then set up.
2. Work
kankaku measures on its own. In pi it asks which client a new project is for; in Claude Code it takes it from your hub catalog.
3. Report with data
Time and cost by client, project and task. In the terminal or the hub.
For anyone who works by project
Freelancers, studios and teams that deliver by client and want to know what part of the work the AI did, how long it really took, and what it cost.
By client and project
Every minute and every token, assigned to where it belongs.
Work vs. waiting
Counts what the agent worked, not what it waited on you.
Sessions and tasks
From a loose prompt to a delivered task, with its history.
Subagents, no double-count
Never counts the same minute twice.
One dashboard for the business
The hub: clients, projects, tasks and trends. Self-hosted.
Export and decide
CSV or JSON for your invoicing or your report.
- Local by default — no hub, no network calls.
- Prompt text never uploads unless you turn it on.
- Self-hosted hub — no third party.
Better with Engram
If you work with Engram, the hub also tells you what got done.
- Each session comes out with its goal, not just an id.
- The summary of what got done, next to time and cost.
- Without Engram, nothing changes.
Available kankaku on npm.
Live demo
Try the hub with fictional data, read-only. Nothing you do there is saved.
demo@kankaku.io- Password
demokankaku
