The mobile companion for tokscale

Every token. Every model. Every dollar.

TokStats is the phone companion to tokscale — the open-source CLI that tracks token usage and cost across 25+ AI coding agents. Pair once with a QR code, and your entire AI footprint lives on your home screen.

  • Open-source core · MIT
  • Your machine stays the source of truth
  • Built on tokscale by Junho Yeo
TokStats Usage screen showing token breakdown, cache-hit rate and cost per message
TokStats Overview screen with total tokens, total cost and a GitHub-style activity heatmap
TOTAL COST $1,422.63
CACHE HIT 90%

One view for your entire stack — tokscale scans them all locally

Claude Code Cursor Codex CLI Gemini CLI Copilot CLI OpenCode Kilo Zed Agent Warp Amp Kimi CLI Qwen CLI Droid Goose Antigravity Trae Cline Roo Code + more
25+ AI coding agents unified in one app
4 Token types broken out — input · output · cache · reasoning
live Real pricing via LiteLLM & OpenRouter
MIT Open-source core you can audit

// What you get

Built for people who measure things.

Not a dashboard you check once and forget — a precise, daily read on where your tokens and dollars actually go. Light and dark, system-aware.

Overview: total tokens 1.9B, total cost, 127 active days, current streak and per-provider glance

01 — Overview

The contribution graph that still tells the truth.

GitHub’s green squares stopped reflecting your real work the day coding agents started committing for you. What you actually do now is drive tokens across models — TokStats makes that your daily heatmap, streaks and all.

  • Tracks the activity that counts now — your work with the models, not commits an agent authored
  • Year-over-year calendar with current & max streaks, best day, daily drill-down
  • Every connected provider, side by side
Providers: total spent, spend mix across Anthropic, Cursor and Gemini, plus per-model breakdown

02 — Providers

Know exactly where the money goes.

Spend mix across every provider, your leading model, average cost per provider, and a per-model breakdown — right down to claude-opus-4-8.

  • Live spend share with a clear leading-provider call-out
  • Messages, models, and average spend per provider
  • Drill into spend by individual model
Usage: token breakdown with input, cache write and reasoning, plus cache-hit rate, cost per message and peak day

03 — Usage

Token-level accounting, not estimates.

Input, output, cache reads/writes, and reasoning tokens — separated. Track your cache-hit rate, cost per message, and peak days. The numbers a cost-conscious engineer optimizes.

  • Cache-hit rate front and center — the cheapest token is the one you reuse
  • Cost per message and peak-spend day
  • Input / output / cache / reasoning split, exactly

// AI News, filtered

Never miss a limit change, model drop, or price cut.

A curated feed for people who live at the frontier. New products, rate-limit changes, and pricing moves — each with an AI-written “Why it matters” and a link to the source. No timeline to doomscroll.

New product Jun 4, 13:39

NVIDIA Nemotron 3 Ultra launches in Kilo Code with free limited-time access

Kilo

A 550B-parameter open-weights model, now available free for a limited time across Kilo Code’s terminal, VS Code, and JetBrains integrations.

✦ Why it matters

Frontier-class 550B open weights, tuned for agentic coding, with a 1M-token context window — at zero cost while the window is open.

Read source →
New product Jun 3, 09:02

Google launches Gemini 3.5 Flash with new Interactions API and thinking controls

Google

65K max output tokens and a new Beta Interactions API with a default medium thinking-effort level.

✦ Why it matters

Configurable thinking effort lets you trade latency for depth per call — a direct lever on both quality and cost.

Read source →
Limits Jun 2, 18:20

Weekly rate limits arrive for Pro coding plans

Provider update

A rolling weekly cap now sits alongside the existing 5-hour window for high-volume agent users.

✦ Why it matters

If you run agents all day, you’ll want to pace heavy jobs — TokStats’ streak and daily views show you exactly when you’re trending toward the cap.

Read source →
Pricing Jun 1, 11:10

Prompt-caching discounts deepen for long-context workloads

Pricing change

Cache-read tokens drop further, rewarding workflows that reuse large system prompts and context.

✦ Why it matters

Your cache-hit rate is now worth even more. The Usage screen surfaces it so you can see the savings land.

Read source →

// Pairing

Log in by pointing your camera at your terminal.

No web form, no copy-pasting API keys onto your phone. tokscale already lives on your machine — TokStats just reads from it after a one-time, device-to-device handshake.

01

Install the CLI

tokscale scans your machine for every AI coding agent you use — locally.

zsh
$ npx tokscale@latest
 scanned 14 agents · 1.9B tokens · $1,422.63
02

Show the pairing QR

One command renders a QR that encodes your token and username — nothing else.

zsh
$ npx tokscale@latest qr

scan with TokStats → {"token":"tt_••••","username":"…"}

03

Scan with TokStats

Point your phone at the terminal. Your full history syncs in seconds — and stays in sync.

Synced 1m ago

Device-to-device. Your token is read straight from the screen — it never travels through a browser, form, or our servers.

// Standing on tokscale

Powered by an open-source core.

TokStats is the mobile companion to tokscale — a high-performance CLI and visualization dashboard for tracking token usage and cost across AI coding agents. The name nods to the Kardashev scale: tokens are the new energy. tokscale collects everything locally on your machine; TokStats gives those numbers a home screen.

AuthorJunho Yeo
LicenseMIT
Agents25+ supported

tokscale is © Junho Yeo, MIT-licensed. TokStats is an independent companion app.

tokscale — TUI
┌─ OVERVIEW ───────────────────────────┐
 total tokens   1.9B
 total cost     $1,422.63
 active days    123 / 127
 cur. streak    55 days
├─ SPEND MIX ──────────────────────────┤
 anthropic   ████████        37%
 cursor      █████████       40%
 gemini                      2.9%
├─ TOKENS ─────────────────────────────┤
 input · output · cache 90% · reasoning
└──────────────────────────────────────┘
$ npx tokscale@latest qr  # pair your phone

// Early access

Get TokStats before everyone else.

We’re running a closed beta on Android now, with an iOS TestFlight round next. Leave your email and we’ll send a single invite when your platform opens.

  • Android — closed testing
  • iOS — TestFlight waitlist
  • Free during beta

// Questions

The details, for the detail-oriented.

Is my usage data private?

Yes. tokscale collects your usage locally on your own machine, and your machine stays the source of truth. You decide what to sync to TokStats — your prompts and code are never read.

Do I need the tokscale CLI to use TokStats?

TokStats is a companion app, so the CLI is what gathers your numbers. Install it with npx tokscale@latest, run npx tokscale@latest qr, and scan. That’s the whole setup.

Which AI tools are supported?

Everything tokscale scans — 25+ agents including Claude Code, Cursor, Codex CLI, Gemini CLI, Copilot, OpenCode, Kilo, Zed, Warp, Amp, Kimi, Qwen, and more. New ones land in the CLI and show up in the app.

How does QR login work, exactly?

npx tokscale@latest qr renders your saved API token and username as a QR code on your terminal. TokStats reads it directly from the screen — the credential never passes through a browser or our servers.

When will it be available?

Android closed testing is live now via Google Play; iOS follows on TestFlight. Join the beta above and we’ll send one invite when your platform is ready.

Is it really open source?

The core — tokscale — is MIT-licensed and built by Junho Yeo. You can read every line at github.com/junhoyeo/tokscale.