AI-native engineering analytics

Your team codes with AI now.
Measure like it.

Tempo reads the work itself: real Claude Code sessions plus git history. It shows you who ships what, what it costs and whether AI is actually paying off, and it never needs a screenshot to do any of it.

Works with Claude Code today, including the Pro and Max subscription seats that no admin API can see.

Tempo · reconstructed proof Source evidence → shipped outcome
  1. Claude CodeSession S1 · 42m
  2. Git3 commits · guardrails/*
  3. GitHubPR #184 merged
Feature landed

Release guardrails shipped

Scope
Deploy checks
Sessions
4 linked
Attended
2h 18m
Follow the proof inward Release guardrails shipped
One worker · one attended window

J. Park’s day, reconstructed.

Attended window
09:12 → 11:54
Total attended
2h 18m
10:04–10:26 · three sessions active together

4 sessions · 3 in parallel · 2h 18m attended · 1 feature shipped

Release guardrails shipped

J. Park attended from 09:12 to 11:54 for 2 hours 18 minutes. Session S1, Build loop, ran from 09:12 to 10:26. Session S2, Test sweep, ran from 09:38 to 11:17. Session S3, Docs pass, ran from 10:04 to 11:02. Session S4, Release check, ran from 11:18 to 11:54. The maximum concurrency was three sessions from 10:04 to 10:26. All four sessions link to the shipped feature Release guardrails.

Analytics that survive the AI era

Commit counts stopped meaning anything.
Substance didn't.

When AI does the typing, volume is free. Anyone can push fifty easy pull requests a week. Tempo scores what mattered: the features that landed, on the priorities you declared, and whether they stayed fixed.

  • Feature-level scoring. Work is grouped into features and scored on quality times importance. Mediocre work nets zero, and a broken launch on a critical surface scores negative.
  • Importance you declare, weighting everyone can see. Leadership tags what counts as P0 this quarter, and the leaderboard weights work by those declarations.
  • Churn and rework detection. Reverted work, repeated fixes to the same subsystem, pull requests that stall or get superseded. The "tries but never lands it" pattern is surfaced as flags with evidence attached, never as a silent score penalty.
  • Who steers the AI, and who ships its output unread. On a team where everyone uses AI, usage share tells you nothing. What separates people is whether they validate what the model produced, and the session record shows exactly that.
  • Two windows, always. Full-period and trailing weeks, side by side, so an old mistake never buries a strong recent quarter, and a hot streak cannot hide an old mess either.
Contribution score · trailing 8 weeks
Sum of impact times neutralised quality per feature landed. Top 4 of 12 shown. Illustrative data.
Re-fix loop flagged. The same subsystem was patched four times in three weeks without a durable fix.Evidence: 4 pull requests, 6 sessions attached
Shipped-unread pattern. AI output committed with no validation step anywhere in the session.Evidence: 3 sessions, 2 commits attached
Cost governance

Know what your AI spend bought, not just what it cost.

Most teams can say they spent seventy grand on models last month. Very few can say what it bought. Tempo attributes spend to people, projects and models, then follows it all the way down to the features that landed.

Spend by model · month to date
Illustrative data.
Monthly burn vs budget
Cumulative spend, July. Illustrative data.
  • Cost per feature landed. Spend is attributed through sessions to the work that shipped, which turns a vague monthly total into a price per landed feature: $800, this week.
  • Every seat, including Pro and Max. Subscription seats are invisible to every official admin API. Tempo measures from the session record instead, which puts your whole team on one ledger, API keys and subscriptions alike.
  • Spend by person, project, model and skill. Find the workflow burning Opus tokens on work Haiku handles perfectly well, and the engineer whose $2,500 of agents replaced a week of work.
  • ⚑ Burn-rate flags. They fire while there is still time to act, whenever a project tracks over its declared budget, and they arrive with the sessions that drove them attached.
  • Your analysis costs stay yours. Deep-dive analysis runs through the coding agent and model you already use. Tempo does not meter your questions.
The part no dashboard can do

Ask it where you already work.

Tempo ships as an MCP server for any compatible coding agent. Dashboards show you charts. Tempo answers questions, with citations, wherever you already work.

  • Ask a question, get an answer. "Who is carrying the P0 work?" "What did the auth rewrite cost?" "Is anyone stuck in a rework loop?" Plain language in, evidence-backed answer out.
  • Citations on every claim. Session IDs, commits, PR numbers. Negative findings run a refute-first pass before they are shown, and no single number ever stands without its rubric and caveats.
  • Bring your own frontier model. Analysis runs through the agent and model you already use. You stay in control of the model and subscription, and Tempo never meters your questions.
  • Your reports are your files. Tempo records that an analysis ran, never its contents. What you generate stays on your machine.
Transparent by design

Not bossware. Verifiably.

The monitoring industry trained everyone to expect stealth modes and screenshots. We built Tempo so that its privacy claims are architectural facts, the kind your security team can verify in the product itself.

No screenshots. No screen recording.

The code to capture a screen was never written, so there is nothing to switch on. The same goes for desktop activity: Tempo does not read window titles, app usage or the accessibility tree. It sees agent sessions and git, and nothing else.

No stealth mode.

The collector is a visible app the employee installs and can see running. A hidden mode does not exist to be turned on.

Company repositories only.

Collection is scoped to projects in your company's GitHub organisation, enforced on the device and re-checked on the server. Personal projects never leave the laptop.

Employees see their own data.

Everyone can query everything Tempo holds about them, including an access log of who looked and when.

Delete means delete.

An employee can remove a session after the fact. It leaves storage and every statistic, numerator and denominator, permanently.

Every read is logged.

Manager queries are audit-logged, and the log cannot be switched off. Symmetry is the point: this is measurement you would accept being on the other end of.

How it works

Running before lunch.

Install the collector

A small desktop app on each dev machine syncs Claude Code session data, from company repositories only, on a gentle schedule. It never asks for screen access because it never uses any.

Connect the MCP

A small MCP configuration adds Tempo to any MCP-compatible coding agent, with manager scope for leads and self scope for everyone else. The dashboard covers installs and team timelines.

Ask

Run /tempo:assess, /tempo:churn, or just ask in plain language. Declare this quarter's priorities and the scoring follows them.

The people being measured get the tool too.

Tempo works both ways. The engineer being measured gets useful answers from the same data, on the same day, in the same tool.

/tempo:how-am-i-doing

Your own trend, your strongest shipped work and concrete tips, computed from your data, visible to you, never comparative.

/tempo:team-vision

What leadership has declared as this quarter's priorities. The actual weighting, not a guess at what matters.

/tempo:score-task

Score a task against the declared vision before you spend a week on it. Knowing the weighting up front is what makes the system fair.

Questions

Fair questions, straight answers.

What exactly does Tempo collect?

Coding-agent session data (Claude Code today, other agent CLIs as raw capture) and git metadata, from company-organisation repositories only. That scope is enforced on the device and verified again on the server. It never touches keystrokes, browsing history, window titles, the accessibility tree, email or chat, and there is no screenshot capture of any kind.

Is this employee monitoring?

It measures the work, and you can check the difference yourself instead of taking our word for it. No screen capture exists in the codebase. There is no stealth mode. Employees see and can delete their own data, and every manager read is audit-logged.

Can it really see Pro and Max subscription usage?

Yes. That is a structural advantage of measuring from the session record rather than from billing APIs. Official admin consoles exclude individual Pro and Max seats entirely, whereas Tempo puts every seat on the same ledger, with costs computed from public per-model pricing.

Do scores decide anything automatically?

No. Tempo informs human judgement and never renders verdicts. Every claim carries citations, negative findings run an adversarial refute-first pass before they are shown, and no score is ever displayed without its rubric, its caveats and an explicit list of what the data cannot see.

Where does our data live? Who can read it?

In your Tempo organisation's isolated store. Manager-scope tokens can query team data, always audit-logged. Self-scope tokens see only your own. Nothing is sold or used to train models, and processing stays with the sub-processors that run the service, all of them listed in your agreement.

What does it cost?

We are onboarding early teams directly and pricing with them. Book a demo and we will talk concrete numbers for your team size.

Early access

See your team's real tempo.

A 30-minute walkthrough with the founders, built around your questions rather than a canned deck.

Book a demo

Book a demo

Tell us a little about your team and we will be in touch within a day.

We use this to reply to you and for nothing else. See our privacy policy.