Where is the budget actually going?
Engineering is your largest line item, and you're inferring its shape from ticket states. You can name every cloud cost to the dollar — but not how your engineering weeks are actually spent.
AI engineering telemetry
You bought AI coding tools.
Now prove they work.
Your engineers now work alongside Claude Code, Cursor agents, and Codex. More code ships than ever, tokens burn around the clock — and when the board asks what changed, you have a Jira board and a feeling.
GitHub shows what changed. Jira shows what people claimed. 3an.ai shows how the work actually happened.
Sign in with Google · self-serve · 2-week free trial · no card required
AI changed engineering. Your dashboards didn't.
Software is now built by humans and AI working together — in editors, agent sessions, and terminals your management stack has never seen. Commit counts show leftovers. Standups show what people remember. The how is invisible — and in the AI era, the how is where the money is.
Is AI making you faster — or just louder?
More code is shipping. So is more churn, more review burden, more token burn. Without human and AI work side by side, "AI productivity" is an opinion, not a number.
Which teams actually use the tools you pay for?
Some teams run agents in parallel all day. Some have licenses. From the org chart they look identical — until you can see AI hours, sessions, and tokens per team. "The developers seem to like it" is not a slide.
Where is work getting stuck?
Blocked, distracted, grinding through rework — the activity record knows. Your ticket tracker doesn't, because nobody files a ticket called "I rewrote the same file four times."
Right now, the most AI-leveraged era in the history of software is being managed with Jira theater, GitHub leftovers, and vibes. You'd never run sales or infrastructure this way. You are flying blind — not because leadership is careless, but because the instruments didn't exist.
One telemetry layer across every surface where engineering happens.
3an.ai captures the actual texture of engineering work — in the editor, in AI sessions, in Git, and across the desktop — and turns it into an executive dashboard.
-
Install the capture surfaces
Each engineer gets a personal ingest token — create users and email setup instructions in one click. No engineer sign-in required.
-
Work happens as normal
The extension and desktop agent stream activity in the background. Nothing to start, tag, or remember — and autofetch noise is filtered out by design, so only real activity counts.
-
Leaders see the system
Team overview, per-engineer drill-downs, AI session transcripts, Git flow, GitHub PR throughput, and system-wide focus — filterable by team and date range, minutes after the work happens.
VS Code extension
Editor activity — typing, selections, focus, file switches — straight from the editor where the work happens. One setting to configure: paste the token.
Cursor extension
The same extension runs natively in Cursor (and VSCodium, Windsurf) via Open VSX — and it reads Cursor's own Composer and Agent sessions directly, so Cursor AI work shows up as first-class sessions, not a blind spot.
Claude Code, Codex & agents
Full session capture for Claude Code / CLI and Codex: prompts, responses, tool calls, message counts, tokens in and out — every session badged by source. Claude Desktop registers as a presence signal: timestamps only, never chat content.
Git, every local repo
Commits, branch switches, merges, and rebases with per-commit lines added and removed — read from local Git itself, so it works with any host and needs no webhooks.
GitHub pull requests — optional
Connect a read-only GitHub App — pull requests and repo metadata, nothing else — and the leaderboard gains PRs opened, PRs merged, average time to merge, and reviews per engineer. You choose the repositories; GitHub usernames map to engineers once, in Manage users.
macOS desktop app
A notarized, self-updating menu-bar agent for work outside the editor: keyboard and mouse intensity (counts only — never which keys), app focus, and AI sessions even when no editor is open. Deploys by hand or by MDM with a ready-made permissions profile.
Browser visibility — no extension needed
The desktop agent reads the active tab straight from Chrome, Safari, Arc, Edge, Brave, Vivaldi, and Opera. Domains only by default. Nothing extra to install or manage in the browser.
Editor-only tools see a fraction of the day. Git-only tools see the aftermath. 3an.ai sees the work.
The numbers your current stack can't produce.
Measured activity, defined precisely — because a metric you can't explain to the board is a metric you can't defend. Hours on 3an.ai are activity-based: counts of five-minute slots that contain real work, not timers or screen time.
Human hours vs. AI hours
Every five-minute slot is classified: human work (editor activity, Git, real keyboard and mouse input) and AI work (active agent sessions). AI hours are agent-time — four parallel sessions in one slot count four, because that's the capacity you are actually deploying.
AI ratio, per engineer and team
AI hours over human hours — the single number most leaders have never seen for their own org. It separates the teams that adopted AI from the teams that have AI licenses.
Trends with momentum
Daily, weekly, and monthly trends per engineer or whole team — hours, lines, tokens — with 7-day momentum deltas, so you see direction, not just position.
The leaderboard
Team and engineer totals side by side: hours (total, human, AI), AI ratio, lines edited, tokens, commits — and with GitHub connected, PRs opened, merged, time to merge, and reviews. Sortable, filterable by team, one click from any row to the full individual picture.
AI sessions, fully legible
Every Claude, Cursor, and Codex session: title, project, message count, tokens in and out — and the full transcript, tool calls included. Admin-visible by design and disclosed as such; see the trust section for where the lines are.
Git flow, commit by commit
Per-repo activity with commit subjects, branch switches, and per-commit line stats. Lines edited is honest raw churn — adds plus removes — so heavy rework is visible instead of flattering.
PR throughput, defined honestly
With GitHub connected: PRs opened, PRs merged, and average open-to-merge time in the selected window, plus distinct PRs reviewed. Self-reviews excluded, bots ignored, and closed-without-merge counted separately — not dressed up as shipped work.
Company rhythm
Company-wide and per-engineer Mon–Sun heatmaps at 15-minute resolution, with peak-hour callouts. The weekly shape of your org's work, at a glance.
Beyond the editor
Top apps and top sites per engineer, plus keys, clicks, and active minutes from the desktop agent — the part of the workday no editor plugin has ever seen.
What we won't pretend to measure: we don't turn token counts into invented dollar figures, we don't score code quality from line counts, and hours are activity slots — not surveillance timers or screen recordings. Fewer numbers you can trust beats a wall of theater.
Built for the people who answer for engineering.
CEOs
You sign the bill for engineering — and for AI tooling on top of it. Get a defensible answer to "is it working": adoption, hours, and output in one view, instead of a translated summary of a summary.
CTOs & VPs of Engineering
You're accountable for the engineering system, not individual heroics. See whether AI adoption is real or nominal, which teams have changed how they work, and where the activity says the system is jammed.
Engineering managers
Skip the archaeology before every 1:1. Each engineer's week — human work, AI sessions, Git activity, focus — in one drill-down. Context for coaching, not a stick for compliance.
Platform & developer-productivity leaders
You rolled out the AI tooling. Now show the receipts: sessions, tokens, AI hours, and adoption curves per team — the evidence layer for every tooling decision you'll make next year.
Real visibility, honestly labeled.
3an.ai is not spyware, and we refuse to market it like spyware. It is engineering telemetry your company licenses — disclosed to the team, configured deliberately, and pointed at the system, not used to ambush individuals.
Here is the honest version: it captures editor activity including edit content, full AI session transcripts (visible to your org's admins), local Git history, and — through the desktop agent — app focus, browser activity, and input intensity. That is real visibility, and it should be deployed as such: tell your team it's running. We build on the assumption that you will.
No keylogging outside the editor
System-wide keyboard and mouse capture is counts-only by construction — the agent records that keys were pressed, never which ones. There is no key content to leak because none is recorded.
Domains, not URLs, by default
Browser visibility records the bare domain unless your org explicitly opts into full URLs. Window-title capture is a separate org-level toggle. Both are enforced server-side at ingest — not in a client setting someone can bypass.
No screenshots. No screen recording. No timers.
Hours are derived from five-minute activity slots, not from watching screens or clocking seconds. Claude Desktop usage is a timestamp, never chat content.
Per-org isolation, org-controlled
Each organization's data is its own. Privacy flags are set per org by your admins, per-user tokens are revocable at any time, and deleting a user deletes their events.
Surveillance asks "who's slacking." Telemetry asks "is the system working." We built the second one — and wrote this page so you can defend the difference to your own team.
Pricing
Per engineer. Self-serve. Every new workspace starts on a 2-week free trial.
Monthly
$9.99
/ user / month
Billed monthly
- All capture surfaces, full dashboard
- Unlimited events & AI session history
- Per-user tokens, revocable any time
- Email support
Yearly
$5.99
/ user / month
Billed annually
- Everything in Monthly
- Priority email support
- Locked-in rate for 12 months
On-prem
$20k
/ year, starting at
Self-hosted — your cloud or your metal
- The full stack inside your perimeter
- Data never leaves your infrastructure
- MDM rollout support
- Direct line to the founder
Stop managing the AI era on last era's instruments.
Humans aren't going anywhere — but the way they work has already changed, and your dashboards haven't noticed. Every quarter without real telemetry is another quarter of AI budget justified by anecdote. Sign in with Google and watch the first real picture of your org assemble itself today.
Self-serve. Your workspace is provisioned the moment you sign in.