Half your work is done by AI now.
Who keeps the books?
Not hours, work state.
A local ledger of your day: your hours and your agents’ work, in an invoice where every line holds up.
How it works- 12+AI agents trackedClaude Code, Codex, Cursor, Copilot, Gemini CLI, Cline/Roo, Kiro, OpenCode, Warp…
- 0 bytesof your activity on our servers by defaultMobile sync is opt-in and end-to-end encrypted
- 30 daysfree, no cardNo account and no auto-renewal
- PDF · CSVinvoices with provenanceEach line traces back to sessions and commits
We track CPU load more precisely than our own day. After eight hours at the computer it's hard to answer a simple question: what did I actually do?
LogAgent is the local ledger of your working day — your hours and your agents’: sessions, tokens, cost, commits. It reconstructs the day automatically, explains the why behind every number, and turns it into an invoice where each line can be defended. Offline. No cloud.
From agent session to invoice line
Claude Code, 2h 14m, $3.80 in tokens → three commits in payment-service → task PAY-142 → line 7 of the March invoice, with its “why this is billable” note attached. Nobody else closes this loop — not Toggl, not RescueTime, not WakaTime.
AI tools made the day fragmented
Cursor, Claude Code, Codex and console agents changed how we work. Days got more fragmented — and more and more of the work is done by agents, so there's nothing to remember it by.
Read in the manifestoConstant switching
IDE, browser, terminal, the agent dialog — focus fractures dozens of times an hour.
High cognitive load
Prompt, review the result, fix, prompt again. From the outside it just looks like "browser and VS Code".
Agent spend is real money
Cursor, Claude, tokens, subscriptions — a budget line nobody attributes to tasks or clients. You can’t bill what you can’t see.
Your agents work. LogAgent keeps their books
A new accounting layer: not where you coded, but what your agents did — Claude Code, Codex, Cursor, Copilot, Gemini CLI, Cline/Roo, Kiro, OpenCode, Warp and more — what it cost, and what survived to production.
- Every session, meteredTime, tokens and dollars — per tool, per project, per task.
- Outcomes, not vibesWhich sessions landed in commits and which got reverted — measured, not remembered.
- Efficiency, not just spendOne-shot rate, cache hits and hints on where a session burned tokens without a result.
Here "AI" is the object of analysis, not a neural net inside.
The invoice comes from the real day
Auto-grouping segments into projects, clients, and rates — with an invoice where every hour has a reason.
- Segments → projectsActivity maps itself to clients and rates.
- Invoice with rationale"Why this time is billable" next to each line.
- Explainable basisNot just "logged work" — show how it composed.
- Agent hours includedAI session time and cost enter the invoice as explainable lines, not a mystery markup.
One signal lies.
A series gives an honest picture.
The decision is not one sensor but their agreement. Disputed segments go to review automatically.
Ten sources — one coherent state assessment. And when there’s no data, the day says unknown — never a fake 0% productivity.
Work state instead of hours
Not "how long in an app", but what state the day was in — and what follows from that.
Read in the manifesto"5 h at the computer, 61% productivity, 58% fatigue. Draw your own conclusions."
- →Numbers without context are not an explanation.
- →Unclear what to do with them.
- →One score hides the shape of the day.
"3.5 effective hours. Drop after 4 PM. Two 25-min focus sessions. Break skipped three times. Posture slipped after 40 min."
- ✓Every conclusion rests on clear signals.
- ✓Segments have confidence; disputed ones go to review.
- ✓You see not only how long, but why.
An agent is about proactivity,
not passive tracking
All analytics run on rules and heuristics — local and explainable. No black-box single numbers.
Collects signals itself
Windows, domains, git, IDE, AFK, focus, breaks, camera — without your input.
Classifies with rules
Transparent heuristics instead of a magic score. Every segment can be explained.
Intervenes when it matters
Spots anomalies and nudges only when signals say it's needed.
Not pomodoro on a timer
A nudge comes when signals say it's needed — not on a blind countdown.
- 20-20-20 for eyesContextual, by load and fatigue — not every N minutes blindly.
- Posture and fatigueA gentle reminder when signals have accumulated.
- Pomodoro as a modeClassic timing is available as one option.
Eye-load guidance aligns with NIOSH / CDC / OSHA. LogAgent is not medical advice: nudges are soft and signal-based.
Tools know "how long".
Not "in what state".
An hour can be flow, procrastination disguised as research, a stall, or a call. App time alone explains nothing.
Read in the manifestoKnow: hours per app.
Miss: whether it was real work.
Know: what was planned.
Miss: what actually happened.
Know: steps and sleep.
Miss: five hours in the IDE.
Knows: meetings.
Misses: focus between them.
The problem lives between tools.
At the intersection of privacy and automation
Most trackers force a trade-off: automatic capture in the cloud, or privacy at the cost of manual timers. LogAgent covers both corners at once.
Compared on typical capabilities per tool category; individual products vary. LogAgent sits where automation doesn't cost you privacy.
Your ledger never leaves your machine
Privacy-first is an engineering choice: capture, analytics and invoices live in a local database. Mobile sync is optional and end-to-end encrypted; the optional camera runs frame-by-frame in memory and stores nothing.
Value compounds over time
A day is thin context. The longer LogAgent stays, the more reliable the picture.
Read in the manifestoOne slice is not yet a pattern.
18% data confidenceYou see when focus peaks and dips.
42% data confidenceAnd you see what each client and each agent actually costs.
70% data confidenceEnough data to trust conclusions.
100% data confidenceFair and local
30 days free, no card. Data stays on your device on every plan.
- Explainable day and analytics
- AI sessions: tokens, cost, commits
- Invoices with provenance, PDF/CSV
- Context-aware breaks
- Local, offline
- Same features as monthly
- Cheaper than 12 monthly licenses
- One payment a year, no auto-renewal
Payments on the English site are coming soon. The 30-day trial is fully free.
Early-stage product, built in public.
Questions before you install
What is LogAgent?
A local desktop app that records your working day and your AI agents' work — sessions, tokens, cost, commits — and turns it into explainable reports and invoices.
Which AI agents does it see?
Claude Code, Codex, Cursor, Copilot, Gemini CLI, Cline/Roo, Kiro, OpenCode, Warp and more. It reads their local session logs; nothing is sent anywhere.
Where does my data go?
Nowhere: activity data stays on your device. Our server only knows your purchase — email, device ID and license key.
Can I bill AI agent time to clients?
Yes. AI sessions carry time and cost, attach to projects and tasks, and enter the invoice as lines traceable to the sessions and commits behind them.
What happens after the free trial?
The trial lasts 30 days, no card. After it, features including export are limited until you buy a license; your data stays on the device.
Is there auto-renewal or a refund?
No auto-renewal: each license is paid once for 30 or 365 days. Full refund within 14 days of purchase.
Can my own AI assistant read my day?
Yes, locally: the optional MCP server lets your agents query the ledger — read-only by default, actions only after approval.
Built for people who live at the computer
LogAgent is made for people whose working time — and whose agents’ working time — is the product.
- Freelancers
Friday: the invoice is ready
LogAgent splits your time across clients and projects on its own and rolls it into an invoice — every line can survive a client’s “what’s this hour?”.
Why it matters: A freelancer sells time but can’t recall where it went by the end of the day. Automatic, explainable tracking protects both income and client trust.
- Consultants
Precise per-client attribution
Time attaches to the right project and client automatically, and reports and invoices come together in a couple of clicks.
Why it matters: With many clients and projects, you need to show transparently what you billed for. Precise attribution removes disputes and saves hours of reporting.
- Working with agents
Know what your agents did
Sessions, tokens and cost per tool and per project; which AI work landed in commits, what got reverted, and where tokens were wasted.
Why it matters: You prompt, review, fix, re-prompt. From the outside it’s “browser and VS Code” — LogAgent shows what actually got produced and what it cost.
- Confidential work
Nothing to send to the cloud
Client work you can’t hand to someone else’s servers: capture, analytics and invoices live in a local database on your machine, offline by default.
Why it matters: When client work can't go to someone else's servers, local-first isn't a feature — it's the requirement.
Your AI can read your ledger — locally
A local MCP server turns your day into context for your own agents: ask your assistant what you actually did on Tuesday and it answers from the ledger — read-only by default, actions only after your approval. LogAgent is the first practical module of DuoHuman, a personal self-digitization layer.
Start the free trial
Download LogAgent and see your first explainable day. Free, no card, all local.