About this skill

Problem

Long-running LLM workloads create a practical accounting gap: developers need to know how much context remains in the current session, how many tokens were used in this conversation, what today or this week cost, whether a budget threshold was crossed, and how much balance a provider still has. These signals are often scattered across session state, billing APIs, and local logs, making cross-day or cross-model comparison awkward.

How it works

token tracker consolidates the query path into local scripts and JSON state files:
- Live session: reads session_status for token usage, cost, cache hit rate, and context occupancy;
- Historical usage: generates today's summary, detailed bill, weekly report, budget status, and budget updates;
- Data source: prefers configured provider APIs, merges usage into usage.json, and falls back to current-session recording when no provider is available;
- Automation: a session hook checks daily reports, weekly reports, and budget alerts at session start, with state files preventing duplicates; automation/ templates can be used for external scheduling.
It supports openai, anthropic, openrouter, deepseek, kimi, and minimax, including provider listing, usage fetch, and balance queries where available.

Boundaries

session_status is only for the live session and should not be used for today's total questions;
some providers require Admin or Management API keys, and deepseek, kimi, and minimax are primarily balance-oriented;
zero usage is a valid result, but a corrupted usage.json should be surfaced as an error;
the response language follows the user, while script-backed outputs remain the primary source for engineers who want usage checks embedded in agent workflows.

Screenshots

Use Cases

Inspect live session tokens, cache hit rate, and remaining context before compacting a long agent conversation.
Generate today's summary, detailed bill, weekly report, and budget status during a recurring LLM cost review.
Fetch provider usage and balance after connecting OpenRouter or DeepSeek, then verify local usage records.
Check missed daily reports, weekly reports, and budget alerts when starting a new agent session.

Best For

Engineers maintaining LLM workflows who need live session context, tokens, and cost checks.
Technical owners responsible for API budgets who need daily or weekly usage, bills, and alerts.
Platform engineers integrating model providers who need usage sync, balance queries, and local records.
Agent ops authors who need automated daily reports, weekly reports, and budget reminders.