Tokaware · Your limits
36%room
Go
use work
workClaude Max 20x36% left
resets in 1h 48m
personalClaude Pro8% left
resets in 34m
side projectCodex Pro79% left
resets in 4h 11m
example accounts · demo clock
// README.md
# Overview
Tokaware shows how a team uses Claude Code, Codex and the other AI coding tools, and what it costs: every session, limit, computer and dollar, by person, team and project. For one developer it is a free limits dashboard that puts every Claude and ChatGPT/Codex limit, across all their accounts, on one screen. It says which account to use right now and when each limit comes back.
- Every Claude and ChatGPT/Codex limit across all your accounts, free
- A team console with live sessions, people, computers and projects
- Seats and API spend by person, group and project, with budgets
- Alerts in Slack, Microsoft Teams or email before a limit or budget runs out
- Works with Claude Code, Codex, Gemini CLI, OpenCode, Kilo Code and Cline, plus Cursor and GitHub Copilot
- Desktop app that switches CLI logins and starts the 5-hour window on a schedule
- Computers share usage numbers by default, and conversations only if allowed
- Encrypted at rest, with single sign-on, two-factor and an audit log
spec.json
{
"name": "Tokaware",
"stack": "Next.js, React, libSQL, Electron",
"platforms": ["Web", "Windows", "macOS", "Linux"],
"license": "Free for up to 3 people, then from $5 a seat",
"support": "Email + online guides"
}// screenshots/
# Gallery
// ARCHITECTURE.md
# Under the hood
- 1Next.js 16 and React 19 on libSQL, with no other runtime dependency but analytics
- 2The browser polls each account's limits through a server proxy: Claude every 3 minutes, Codex every minute
- 3The server refreshes the tokens it owns, caches per token, and serves the last good reading while it backs off from a rate limit
- 4An Electron desktop app ships its own copy of the pages and checks in with the hosted API
- 5Connected computers read the coding tools' local session logs and report numbers, not content, by default
- 6Provider tokens and computer reports are encrypted at rest with AES-256-GCM
./capabilities
- Every rate-limit window: the 5-hour session, the weekly limit and per-model weekly limits
- Pace, burn rate and a daily budget per window, with a projection to the reset
- Tokens, API value, prompts, lines changed and active time per person and day
- Claude Code and Codex sessions started on a teammate's computer, within what that computer allows
- Provider connectors that bring in what Anthropic, OpenAI, Cursor and GitHub bill
- Budgets, forecasts, cost-center chargeback and CSV exports
// DEVLOG.md
# How we built it
./challenges
- Reading limits from usage endpoints the providers do not document and can change without notice
- Anthropic rate-limits its usage endpoint per token, with no Retry-After, for hours at a time
- Refreshing tokens without logging the developer's own CLI out, since both providers rotate refresh tokens
- Counting a conversation once when it was resumed, forked or copied into several logs
./insights
- Answering one question first, can I keep working and on which account, beats showing every number
- Numbers by default and content only by consent made computers easy for teams to connect
- API responses only ever grow, never repurpose a field, so every installed desktop version keeps working
- Flushing state files to disk before renaming them means a crash leaves the old file or the new one, never zeros
Next.js 16 + React 19Tailwind CSS 4libSQL / TursoElectron desktop appVercel
// docs/
# Documentation
./guides
- Claude Code usage limits, explained
- When does the Claude Code limit reset?
- Codex usage limits, explained
- How much does Claude Code cost for a team?
- How to track Claude Code usage and cost across your team
- Install the Tokaware desktop app
