Status reports,writtenby the AIReported the moment a session ends. You just open the dashboard.See the demo dashboardNo sign-up. Demo data is ready.Claudethat didthe work.
Claude helped L Lin Xiao fix an idempotency bug in the refund callback.
Everyone on the team has an AI partner.
Nobody knows better what got done today.
When a session ends, the partner writes it up in one line and sends it to your dashboard with the time and changes.
Gemini helped W Wang Ke backfill 30 days of daily-active data.
Codex helped C Chen Mo refactor the order state machine to be table-driven.
Five questions every boss asks. One dashboard answers.
Good morning, boss.
Today 5 people and their AI partners finished 11 sessions, mostly in payments-apiinfra-k8s.
7
↑ 0% vs previous 7 days
73
↑ 14% vs previous 7 days
54.7h
23.9M tokens
+12klines
−4.3k lines · 70 commits
Worth a look
1
is solidly held only by . Export a SKILL.md so the team can reuse it.
AI sessions per day
Claude CodeCodexCursorGemini CLI
People
Person
Tools
Trend
Sessions
Time
Last active
L
Lin XiaoBackend
18
13.6h
1 h ago
Z
Zhou YiFrontend
13
8.2h
2 h ago
C
Chen MoBackend
12
9.7h
1 h ago
S
Sun HeInfra
12
10.3h
1 d ago
W
Wang KeData
9
5.0h
4 h ago
Latest activity
Codex helped in infra-k8s: Turned the staging deploy scripts into a Helm chart
1 h ago50 min4 turns5 files+226−691 commit
Codex helped in infra-k8s: Added image caching to CI, cutting build time from 11 to 4 minutes
1 h ago57 min14 turns3 files+73−501 commit
Claude helped in payments-api: Refactored the order state machine to be table-driven, removing 3 duplicate branches
2 h ago48 min13 turns10 files+128−871 commit
Claude helped in mobile-app: Implemented deep links from push notifications
2 h ago31 min10 turns7 files+97−631 commit
Cursor helped in mobile-app: Fixed the keyboard covering the input field on iOS 17
2 h ago56 min15 turns10 files+214−131 commit
Codex helped in infra-k8s: Debugged frequent pod OOM kills and tuned memory limits and JVM flags
4 h ago1.0 h13 turns6 files+152−18
Gemini helped in data-pipeline: Added data quality checks and alerts to the ETL
4 h ago25 min3 turns1 file+120−681 commit
Codex helped in payments-api: Debugged the reconciliation job timeout, traced it to an N+1 query and added batch loading
4 h ago50 min4 turns4 files+243−331 commit
What your best people know, written down.
Skills, inferred
Rules over the work summaries infer what each person is good at: debugging, refactoring, performance. Nobody fills in a form.
Bus factor, visible
When only one person solidly holds a skill, the dashboard tells you.
Export as SKILL.md
Export one person's way of working as a file, drop it into .claude/skills/, and everyone's Claude Code can follow it.
First, what is collected and what is not.
Surveillance tools do not last. HeyBoss takes the minimum a roll-up needs, and the team can see all of it.
Reported
Tool, project folder name, session time and length
Number of turns, token usage
Files changed, lines added and removed, commits
A one-line summary (the session title, or the first 200 characters of the first prompt)
Never reported
Code or file contents
The full conversation
Keystrokes, screens, browsing history
Anything outside an AI session
Visible to the team
The collectors are a few dozen readable lines. Set HEYBOSS_DRY_RUN=1 to see the exact payload before anything is sent.
Your data stays yours
Self-hosted. The data is a JSON file on your own server and never touches a third party.
Never in the way
If reporting fails it is skipped silently. It never blocks or interrupts an AI session.
Three steps in, automatic after.
1
Deploy HeyBoss
One Next.js app. Set HEYBOSS_TOKEN, start it, done.
HEYBOSS_TOKEN=secret npm run start
2
Install the collectors
A SessionEnd hook for Claude Code, notify for Codex. One line of config each.
That is why the boundary comes first: summaries and counts only, open scripts, and a preview of every payload. Tell the team what is collected before you roll it out. That matters more than any technical measure.
Do more sessions or more lines mean better work?
No. The numbers show where AI collaboration happens. They are not a performance score. HeyBoss replaces “what are you working on?”, not a ranking.
Which AI tools are supported?
Claude Code and Codex CLI have ready-made collectors. Cursor, Gemini CLI or anything else can POST to /api/ingest.
How are skills worked out?
Keyword rules over the work summaries, with the sessions behind each skill listed so you can check. Treat it as a lead, not a verdict.
What do I see before real data arrives?
Demo data. The dashboard switches over when the first real report comes in.