Usage & Analytics

When running large language models and complex autonomous Agent workflows, real-time consumption metrics help you protect budgets, optimize prompts, and maximize efficiency.

The Aitok Hub Console includes a live Usage & Analytics Dashboard that illuminates every single request and cent spent.

Analytics Dashboard Overview

After logging into the console, navigate to Usage & Analytics (/hub/usage) in the sidebar:

  1. Balance & Spending Metrics:
    • Available Balance: Your live remaining balance.
    • Today's Spend: Cumulative consumption for the current calendar day.
    • Monthly Spend: Total expenditures incurred this calendar month.
  2. Usage Trend Charts:
    • Filter by past 7 days, past 30 days, or a custom date range.
    • Visual charts display daily request volume and total token throughput over time, highlighting peak usage cycles.
  3. Model Distribution Breakdown:
    • Shows percentage breakdown across models (e.g., Claude 3.5 Sonnet vs. GPT-4o vs. Llama 3.3).
    • Identify at a glance which models are driving the bulk of your expenditure.

Detailed Request Logs

Below the summary charts, you will find granular logs for every completed request:

  • Timestamp: Second-level timestamp of request execution.
  • Model Used: The exact model ID invoked.
  • Token Breakdown:
    • Input (Prompt Tokens): Size of the prompt context delivered to the model.
    • Output (Completion Tokens): Number of tokens generated in response.
  • Deducted Cost: Exact currency deducted from your balance for this invocation.
  • Source Credential: Pinpoints which specific AI Key or Authorized Agent made the request.

Tips for Cost Optimization

  1. Investigate Runaway Agents: If spending spikes unexpectedly, filter logs by date and key name to diagnose whether an Agent is stuck in an infinite retry loop or excessive context window inflation.
  2. Tier Model Selection: Reserve premium reasoning models (such as Claude 3.5 Sonnet or o1) for mission-critical tasks like intricate code generation and multi-step reasoning. Route routine summarization, formatting, or translation to budget-friendly models like GPT-4o-mini or Llama-3.3-70b.
  3. Reset Long Conversation Threads: Periodically start fresh chat sessions to avoid carrying excessive conversation history, which inflates the input token count on every consecutive turn.