Go to Dashboard

Claude Code analytics for engineering teams

See how your team uses AI. Then get measurably better.

Local Claude Code sessions become prompt-quality scores, timelines, and coaching. Private, local, and yours alone. Your code never leaves the machine.

Get TrayToken

Take AI seriously. Measure it like you mean it.

Private, local tracking. Yours alone.

A tray agent reads local session logs — only for the projects each engineer opts in. Prompts and metrics sync. Nothing else.

  • Works on personal Pro/Max plans
  • File references reduced to bare names before syncing

Data flowLocal first

Your machine~/.claude session logs, selected projects only
prompts + metrics only
Team dashboardscores, timelines, trends per role
Excluded from sync
Never leaves homeyour code, file contents, Claude's responses, tool outputs

Every session scored across four dimensions.

Prompt Quality 30%, Context Management 25%, Iteration 25%, Tool Usage 20%. One overall score, rolled up daily, banded for instant reading.

  • Rewards file references, /compact, a 4-8 prompt sweet spot
  • Token spend paired with outcomes, so effort is measurable
  • One engine on every platform, locked by shared test fixtures

Scores7-day avg

Prompt Quality24.9
Context Mgmt20.5
Iteration46.0
Tool Usage71.3

8.4M→6 commitsTokens → outcome, straight from the day summary

Team activity, day by day.

Prompts and sessions for the whole org, on one chart. Adoption dips show up before anyone asks for a report.

Activity Trends

Company, team, and personal views.

Admins see the company, leads see their team, engineers see themselves. Competition on top, private progress underneath.

  • KPI cards, Score Trends, Team Comparison, Leaderboard
  • 30d / 60d / 90d ranges with per-dimension toggles
  • Compare view lines up two engineers side by side
Engineers
6
Active
Avg Score
42.2 ▲8.9
vs last week
Sessions
128
This week
Prompts
764
This week

Score Trends

Leaderboard

1M. Cartersenior58
2J. Brooksmiddle52
3D. Fostermiddle46
4A. Walshjunior38
5P. Harmonmiddle31

Coaching that arrives before the 1:1 does.

When a dimension dips below 40, concrete fixes appear. The Attention panel flags anyone dropping fast or going quiet.

  • Flags: declining score, weak dimension, silent week
  • Graded critical / warning / info, with an all-on-track state

Attention Needed1 critical · 1 warning

P. Harmon

Score declining fast

Down 14 pts in 7 days. Tool Usage fell most: 41 → 19.

R. Mercer

No sessions this week

No Claude Code sessions in the last 7 days.

2 items need attention

Recommendations

Improve your prompts

Reference files with @-mentions, add constraints and acceptance criteria.

Manage context

Run /compact between tasks. Keep standing instructions in CLAUDE.md.

Analytics for the follow-up questions.

Active days, sessions per day, tokens, accepted lines and accept rate — with per-day Activity and Lines of Code charts underneath. Personal for engineers, org-wide for admins and leads.

  • Activity, Models & Tokens, and Tools & Adoption tabs
  • 7 / 30 / 90-day ranges, always in UTC
My AnalyticsYour Claude Code usage · Data in UTC
Active days
14
of 30 days
Sessions/day
2.4
— vs previous 30d
Total tokens
1.4B
— vs previous 30d
Lines accepted
7.3K
— vs previous 30d
Accept rate
97.0%
— vs previous 30d

Activity

Lines of Code

Lives in your menu bar. Stays out of your way.

Parses new sessions in the background, syncs every 5 to 60 minutes, starts at login. Fails loudly, succeeds quietly.

  • Only selected projects are ever read
  • Tokens encrypted at rest with OS key storage where available

Several orgs, three roles, one clean sign‑in.

Switch organizations without signing in again. Roles enforced everywhere. Browser sign-in returns through a state-verified traytoken:// link.

  • Admin, team lead and engineer scopes on every view and API call
  • One-click CSV export of sessions, scores included

Choose your workspace

→ traytoken://auth/callback?state=verified ✓

Three steps, then it runs itself.

No API keys, no instrumentation. If Claude Code runs on the machine, TrayToken can score it.

Install the tray app

Pick your build, sign in through the browser, choose which projects to share.

It scores your sessions

New local sessions are parsed and scored across four dimensions, on your schedule.

Watch the team improve

Dashboards, timelines and coaching update as scores roll in.

Day timeline

One engineer. One day. The whole story.

The dashboard's day view, rebuilt here with sample data. Hover any segment.

‹ Wed, Jun 24 ›
  • Research / audit
  • Implementation
  • Debugging
  • Review / verify
  • Docs / tests
  • Ops
  • ◆ human turn
  • ✓ commit
  • ✕ abandoned
  • ⋯ idle
  • --- waiting
lane 1
✓ ✓
lane 2
✓ ✓
lane 3
✕ ✓ ✓

Segment opacity = token intensity--- = waiting on humandashed border = uncertain classification⋯ = idle gap

6
Sessions
6.5h
Active hours
3×
Peak concurrency
8.4M → 6 commits
Tokens → outcome
1
Abandoned
0.4
Babysitting ratio beta
  • Implementation32%
  • Research / audit21%
  • Debugging19%
  • Review / verify6%
  • Ops11%
  • Docs / tests11%

Project split · of 6h 30m active

projects/checkout62%
projects/platform38%

Privacy, by construction

Built on consent, not surveillance.

The parser extracts prompts and metrics. Nothing else is ever stored.

Synced to the dashboard

What we collect

  • ✓Your prompts to Claude and their quality metrics
  • ✓Computed session metrics: timing, token counts, tool use, accepts and rejects
  • ✓File names, reduced to bare basenames before syncing
  • ✓Timestamps of sessions and turns
  • ✓Project names, from the projects you selected

Only from projects you opt in.

Never stored, never sent

What we never store

  • ✕Your code. Not a single line, ever
  • ✕File contents of any kind
  • ✕Claude's responses, including reasoning and generated text
  • ✕Tool outputs: command results are read in memory only to count accepts and rejects
  • ✕Unselected projects, memory files, or settings

You choose the projects; everything else stays on your machine. The onboarding says so. The code enforces it.

Pick your platform.

One tray app, three operating systems. The dashboard runs in any browser.

macOS

Universal DMG · Apple Silicon + Intel in one Checking for latest version… Download

Windows

Installer .exe · NSIS, x64, choose your folder Checking for latest version… Download

Linux

AppImage · x64, run anywhere Checking for latest version… Download

Linux

.deb package · x64, Debian & Ubuntu Checking for latest version… Download
Web dashboardScores, timelines and leaderboards in any browser.
Go to Dashboard

Common questions

No. Your code, file contents, Claude's responses, and tool outputs are never stored and never leave your machine. File references are reduced to bare file names before syncing.

Five things: prompt text, computed session metrics, file names, timestamps, and project names. Only from the projects you selected in the tray app.

macOS (universal DMG), Windows x64, and Linux x64 (AppImage and .deb). The dashboard runs in any modern browser. There are no mobile apps.

No. Download the new build and reinstall over the old one. Settings and project selections are kept, so nothing is lost.

No. TrayToken exists precisely because engineers run personal Pro and Max plans. Everything comes from the local session files Claude Code already writes.

Yes, it is the first thing the app asks. Projects are opt-in one by one, and Cloud Sync can be paused from the tray at any time.