tokenmark - LLM cost analyzer
Pricing
Pay per usage
Pay-per-event LLM cost analysis. POST a JSONL log or array of LLM call entries; get back per-day/per-model spend, top costly calls, and rule-based route recommendations. Local-first alternative: tokenmark npm package.
What to do. 'analyze' computes a full report from a log (paid). 'compute_cost' returns the USD cost of a single call (paid, micro-priced). 'list_pricing' returns the current pricing table (free).
For 'analyze': array of call log objects with fields {provider, model, prompt_tokens, completion_tokens, [timestamp, cache_*, user_id, error]}. Either this or 'jsonl' must be supplied.
For 'analyze': raw JSONL log (one call entry per line). Either this or 'entries' must be supplied.
For 'analyze': how many top-costly calls to include in the report. Default 10, max 100.
For 'compute_cost': LLM provider.
For 'compute_cost': model name (e.g., claude-haiku-4-5, gpt-5, gemini-2.5-flash, llama-3.3-70b-versatile, deepseek-v3.1).
For 'compute_cost': output token count.
Optional. For 'compute_cost' if you used Anthropic prompt caching.
