AgentTelemetry

Your AI work, in focus

Every agent.
One clear
picture.

See the tokens, time, and estimated cost behind your AI coding. One private dashboard, right on your machine.

Install the appChoose an option
Install with pip

Free & open source macOS & Windows apps · Linux via Python MCP setup

Today SAMPLE
2.4M
tokens · $3.21 est. API cost

Illustrative preview

Less guesswork. More perspective.SCROLL TO EXPLORE

One view across your tools

01The dashboard

A little perspective.
A lot more clarity.

Bring usage from your coding tools into one calm view. Follow tokens and estimated cost by model, project, session, day, and more — using the logs already on your computer.

AgentTelemetry / Today
Interactive preview · Sample data

YOUR USAGE · ILLUSTRATIVE

Tokens ⓘ2.4Macross your tools
Est. API cost$3.21API list-price estimate
Sessions18today
Cache hit rate82%cache reads
ACTIVITY

A day in context

InputOutput
Illustrative token activity across the dayAn area chart showing sample input and output tokens rising toward evening.
8 AM10 AM12 PM2 PM4 PM6 PM8 PM

Sample data for demonstration. Your dashboard starts with your own local logs.

COST · SAMPLE DATA

Estimated spend, by tool.

Today · API list prices
ESTIMATED TODAY$3.21Subscriptions may differ
HIGHEST SHAREClaude Code58% of the sample total
WEEK TO DATE$18.74Illustrative estimate
Claude Code$1.86
Codex$0.82
Copilot$0.36
Cursor$0.17

Sample estimates use API list prices; subscription plans and actual bills can differ.

MODELS · SAMPLE DATA

See what each model carries.

4 models in this sample
C

Claude · Sonnet

Claude Code · 9 sessions

984Ktokens
O

OpenAI · GPT

Codex · 6 sessions

716Ktokens
G

Google · Gemini

Copilot · 2 sessions

421Ktokens
C

Cursor · Composer

Cursor · 1 session

279Ktokens

Illustrative totals are grouped by model; tool attribution stays with the coding surface.

TOOLS · SAMPLE DATA

Usage across your coding tools.

4 tools

Claude Code

9 sessions · 1.1M tokens

46%

Codex

6 sessions · 720K tokens

30%

Copilot

2 sessions · 390K tokens

16%

Cursor

1 session · 190K tokens

8%

An illustrative mix. Each tool may record different amounts of usage detail.

PROJECTS · SAMPLE DATA

Work, grouped by project.

Across 4 projects
⌘

api-service

Claude Code · 8 sessions

812K tokens
◇

storefront

Codex · 5 sessions

604K tokens
▤

documentation

Copilot · 3 sessions

486K tokens
⌑

playground

Cursor · 2 sessions

498K tokens

Project names and usage above are invented examples from a local dashboard.

OPTIMIZE · SAMPLE DATA

Patterns worth a closer look.

Observations, not instructions
↗

Cache reads are doing useful work

In this sample, 82% of repeated context came from cache reads.

82%
◷

Most activity arrived later in the day

Usage rose after the afternoon coding sessions began.

4–8 PM
◇

Compare models on routine edits

Review task needs alongside the model and cost breakdowns.

Explore

These are descriptive sample observations; the page does not change settings or estimate savings.

STORAGE · SAMPLE DATA

Your records stay close to home.

Stored locally
Local ledger

Parsed usage is summarized on this computer.

READY
▤Source logsRead from their existing locationsUNCHANGED
▦Usage ledgerLocal summary used by the dashboardON DEVICE
◷RefreshUpdates as new local usage appearsINCREMENTAL

Illustrative status only. This marketing preview never reads or displays your real ledger.

SESSIONS · SAMPLE DATA

Work, with its context.

4 example sessions

Build a search API

Claude Code · api-service

486Ktokens
Today, 10:42

Refine checkout flow

Codex · storefront

392Ktokens
Today, 09:18

Ship the docs

Copilot · documentation

218Ktokens
Today, 08:14

Explore a new idea

Cursor · playground

104Ktokens
Today, 07:36

Names and counts above are invented examples. Some tools record partial usage; missing data is marked in the dashboard.

YOUR DEVICES · SAMPLE DATA

A wider view, when you want it.

Sharing is off by default
⌘
MacBookThis device
1.5Mtokens
▣
Windows PCConnected device
0.9Mtokens
Combined sample total2.4M tokens

Illustrative only. Optional device sharing works over your local network after you turn it on.

SAMPLE DATA · NO PERSONAL USAGE SHOWNAgentTelemetry v2.0

Eight views, one consistent picture. Compare tools, models, projects, sessions, and time without leaving your machine.

TOKENS · TIME · COST
02Simple by design

Already on your machine.
Finally in view.

No account or API key is needed. Use the dashboard, check recorded sessions in your editor, or connect an AI client through optional MCP access.

01

Install or run

Choose the native app, the editor Agent desk, or start the local dashboard with one command.

YOUR COMPUTER
02

Reads existing logs

Usage is parsed and summarized on your device, then kept in a local ledger you control.

NO NEW TRACKER
03

Find your patterns

See where time and tokens go, compare your tools, and notice what changes.

YOUR OWN PICTURE
A QUIETER WAY TO CHECK IN

At a glance.
When you need it.

The optional macOS menu bar and Windows system tray app brings today’s usage close by. Open the full dashboard whenever you want more detail.

Install the app
2.4M•••
AgentTelemetryToday Monitoring
TOKENS2.4M
EST. API COST$3.21
View dashboard
SAMPLE DATA
MACOS MENU BAR · WINDOWS SYSTEM TRAY
03Private by default

A clear boundary, by design

Your work stays
your business.

AgentTelemetry reads and summarizes logs on your computer. Optional LAN sharing lets your devices share usage with each other. Provider account lookups contact that provider only when you explicitly request one. Local MCP access is also opt-in.

01

Local by default
Your usage ledger stays on your device. No AgentTelemetry account or API key required.

02

Sharing is a choice
Optional device sharing is enabled by you and works across your local network.

03

Account checks need consent
One-time provider lookups run only after you explicitly ask for them.

04

Choose what MCP shares
Requested analytics go to your configured client; a cloud-backed client may forward results to its provider.

Honest numbers. Costs are estimates at published API list prices; subscription plans can work differently. Recorded usage, estimates, and unavailable data are distinguished.

How the data works
04Get started

AGENTTELEMETRY · V2.1.0

Meet your new
point of view.

Free, open source, and ready to run on your machine.

macOS

Native menu bar app · macOS 13+

Download for Apple siliconmacOS · DMG · Python included Download for IntelmacOS · DMG · Python included

Unsigned and unnotarized. See release notes for install guidance.

Windows

Native system tray app · Windows 10/11

Download for Windowsx64 · EXE · Python included

Ready-to-run installer
No Python setup needed

Unsigned installer. Review release details and SHA-256 checksums.

Agent desk for VS Code

Editor extension · 2.1.0

See recorded usage, estimated cost, and linked sessions beside your code. Supports desktop VS Code 1.85+; compatible desktop editors can install the VSIX.

Install from Visual Studio MarketplaceVS Code · AgentTelemetry Team · 2.1.0 Install from Open VSXCursor, Antigravity & compatible editors · 2.1.0

If your editor does not offer either listing, download the universal VSIX and follow the install guide. Requires Python 3.9+ or an installed AgentTelemetry runtime. Choose Start local monitoring in the extension to begin.

Connect your agent with MCP

MCP server · 2.1.0 · off by default

Ask your agent for daily tokens and estimated cost by project or tool, comparisons, and insights. Start the dashboard, then enable Settings → Local MCP access.

Set up MCPEnable access in the dashboard settings

You choose whether to share project/device names or usage from other devices. A cloud-backed client may send these results to its provider. Prompts and credentials are excluded.

PREFER THE TERMINAL?

Install with pip

Linux runs the browser dashboard; native tray apps are available for macOS and Windows.

Python 3.9+ · macOS, Windows & Linux

01 INSTALL

pip install agenttelemetry-app

02 LAUNCHagenttelemetry

Available as agenttelemetry-app 2.1.0 on PyPI. Then open 127.0.0.1:7878 in your browser.

Or run with uvx

Already using uv? Launch in an isolated environment:

uvx agenttelemetry-app

RELEASE V2.1.0BUILT FOR YOUR DEVICEAll releases & checksums
05Good to know

A few clear
answers.

Know what to expect before you bring your usage into focus.

01Is AgentTelemetry free?

Yes. AgentTelemetry is free and open source under the MIT license. You can inspect the code, run it from source, or use the native installer on macOS and Windows.

02Does it upload my usage?

AgentTelemetry reads and stores app usage locally by default. Optional LAN sharing and provider account lookups require your action. When you enable Local MCP access, a compatible client receives analytics you request; a cloud-backed client may forward those results to its provider. Project/device display names and connected-device usage have separate opt-ins. The marketing website uses Google Analytics to count visits; it sends standard page-view details, including the page URL, to Google. That website measurement does not send your AgentTelemetry logs or dashboard data. No AgentTelemetry account is needed.

03Is estimated cost my actual bill?

No. Cost uses published API list prices as an estimate. Subscription plans, negotiated rates, and provider billing details may differ. Some per-request pricing rules cannot be represented exactly in daily summaries.

04Which tools and usage are covered?

It reads logs from Claude Code and Claude Desktop agent mode, Codex, GitHub Copilot, Cursor, OpenCode, Hermes Agent, and OpenClaw. Coverage depends on what each tool records: some sources provide partial usage or estimates, and recorded, estimated, and unavailable data are distinguished.

05How do I get started?
Install the app

Or install with pip (Python 3.9+):

01 INSTALL

pip install agenttelemetry-app

02 LAUNCHagenttelemetry

Or run with uvx

Requires uv. Installs and launches in an isolated environment.

uvx agenttelemetry-app

After launch, open 127.0.0.1:7878.

06Can I use AgentTelemetry in my code editor?

The Agent desk extension supports desktop VS Code 1.85+ and compatible desktop editors such as Cursor and Antigravity. It shows recorded usage and linked sessions. If no dashboard is running, choose Start local monitoring; installation and opening the extension do not start it. See editor install options.

07How does Local MCP access work?

It is off until enabled in dashboard Settings, and the dashboard must already be running. Eight read-only tools return analytics to your configured client; a cloud-backed client may forward results to its provider. Project/device display names and connected-device usage require separate opt-ins. Results exclude prompts, session titles, native IDs, paths, credentials, and account identities. Read the MCP setup and privacy guide.

06The people

PEOPLE BEHIND AGENTTELEMETRY

A small team.
A shared point of view.

Design, data, and product come together to make AI coding work easier to understand.

  1. Uttam Deb

    Data analyst and AgentTelemetry creator, making AI coding usage easier to understand.

  2. Shad Iqbal

    Product and UX systems designer behind the macOS companion.