tokflow_skill

Overview

This skill monitors and analyzes token consumption and cost for LLM usage, providing real-time channel balances, model-level statistics, and actionable optimization suggestions. It also analyzes user prompt patterns and estimates savings from prompt and invocation changes. Use it to track expenses and reduce token spend across paid model channels.

How this skill works

It queries a local TokFlow API (http://localhost:8001/api) to fetch dashboard, model, provider, balance, and prompt-statistics data. The skill interprets JSON outputs into human-friendly insights: totals, trends, top cost drivers, and suggested optimizations. It can regenerate suggestions and run deeper analysis on demand to produce up-to-date recommendations.

When to use it

• Check daily or monthly token and cost overviews
• Identify which models or calls drive the highest cost
• Monitor real-time balances of paid channels
• Get concrete optimization suggestions for models and prompts
• Analyze prompt patterns to estimate token savings

Best practices

• Run dashboard and balance checks before planning heavy usage
• Use model-detail and analysis for root-cause of cost spikes
• Generate fresh suggestions after configuration changes or traffic shifts
• Prioritize prompt-stats findings when conversational costs are high
• Keep TokFlow service running locally on port 8001 for accurate data

Example use cases

• Ask "How much did I spend this month?" to get month_cost and breakdown by model
• Request "Which model costs the most?" to receive a ranked list by total_cost
• Query "What are my channel balances?" to retrieve real-time provider balances
• Run "Any optimizations available?" to get model replacements, caching, and prompt suggestions with estimated savings
• Analyze conversational cost by asking for prompt-stats to see average lengths, buckets, and potential token reductions