Jaadu, the number
leadership trusts.
A per-developer analytics layer for AI coding tools, Claude Code, Copilot, Codex and more. It measures requests, code produced, money spent, bad habits and skill opportunities, captured automatically, without anyone changing how they work. Browser-only, GitLab-hosted, no server, no database, nothing leaves your repo.
Four numbers, per
developer, every day.
Refreshed nightly per employee. The same figures the ROI reports → roll up to the board.
Total AI messages sent, one conversation turn per request.
Distinct conversation windows opened across all AI tools.
Lines generated by AI and actually accepted by the developer.
Monthly API cost vs budget, with daily burn-rate tracking.
Captured locally.
Stays in your GitLab.
No agent watches anyone. A nightly job summarises the day on each machine and pushes it to a repo you own.
Raw AI usage, turned
into board-ready signal.
Per-developer intelligence and the honest framing the reports are built on, scored on a simple band, 75–100 strong, 50–74 watch, 0–49 act.
Seven pages, each
answers one question.
From "how is this developer doing?" to "is the right context loading?", every screen leads with the question it answers.
How is this developer doing overall?
Persona, an overall 0–100 score, headline stats and a preview of every other view.
When do they use AI, and when are they in deep focus?
Peak hour and day, active-day streaks, plus activity and focus-density heatmaps.
What did they actually produce with AI?
AI vs user lines, split by language, tool and workspace.
Are they on track with their token budget?
Cumulative spend vs budget ceiling, daily spend, and tokens per session.
What bad AI habits do they have, and how serious?
Detected habits ranked by severity, each with a concrete suggested fix.
Which repeated tasks should become a saved skill?
Repeated prompt clusters, cancel rate and the time saved by promoting them.
Are their projects set up to give AI good context?
Per-workspace audit of instruction files and skills, worst gaps first.
Habits, skills
and context.
The three views that turn raw usage into something a team lead can act on this week.
Anti-Patterns
bad-habit detectionFive practice groups, Prompt Quality, Session Hygiene, Code Review, Tool Mastery and Context Management, each scored and trended. Every finding carries a severity, an occurrence count and a concrete fix, built for coaching, not for performance reviews.
Skill Finder
repeated prompts worth savingSurfaces prompts a developer types over and over, with cancel rate and average correction turns. A high cancel rate flags a task that should become a well-framed saved skill, with the estimated time it would save.
Context Health
workspace readiness auditScores each repository on whether AI has what it needs, a CLAUDE.md and configured skills, and lists the exact gaps to close. "Add CLAUDE.md to api-server", sorted worst-first.
