token-watch
por vedantsingh60
Coding Agents & IDEsResumen
**Track, analyze, and optimize token usage and costs across AI providers.
Descripción
Qué hace esta skill
TokenWatch is a local Python library for tracking token usage and costs across AI providers. It records API calls, calculates costs against a pricing table of 41 models across 10 providers, and fires budget alerts when configured thresholds are crossed. All usage data stays on your machine with no external connections.
Unlike provider billing dashboards, it captures per-call cost data in real time and surfaces ranked optimization suggestions before the monthly bill arrives.
Cuándo usarla
- Checking daily AI spend before a budget limit is reached
- Comparing model costs for a specific input/output token count before choosing a model
- Getting automatic alerts when monthly spending hits a set percentage of budget
- Identifying which tasks or models are driving the highest per-call costs
- Estimating cost before running a large batch of API calls
Flujo de trabajo de ejemplo
User asks: How much did I spend on Claude this week and how can I reduce it?
- 1
Calls get_spend_by_model(period="week") to retrieve spending broken down by model
- 2
Calls get_spend_by_provider(period="week") to see Anthropic's share of total spend
- 3
Calls get_optimization_suggestions() to generate ranked cost-reduction tips with estimated monthly savings
- 4
Calls format_dashboard() to render a visual summary with budget progress bars
Weekly spend by model alongside optimization suggestions, such as swapping Claude Opus for Sonnet on non-reasoning tasks to save an estimated $8.20/month
Instalación
clawhub install vedantsingh60/token-watchEjecuta el comando de la CLI o pega el prompt en Claude Code, Cursor, Codex o el agente de programación que uses.