Skip to main content

AI Cost Tracking

The Organization → Finance tab lets you track your organization's real AI model consumption.

What is tracked​

Every model call records its consumption (tokens, pages or duration depending on the pricing) along with the corresponding cost, calculated at the model's rate at the time of the call — so historical cost stays immutable even if the model's rate is changed afterwards. Four model families are tracked:

  • AI models (LLM) — agents, supervisor, validation and JSON extraction, but also the switch, "AI judgment" interceptors, the authoring assistant (agent drafting help, regex generator), chat memory, the benchmark and RAG evaluation judges, and the knowledge bases' AI calls (contextualization, enrichment, section summaries, query decomposition, filter extraction, reranking).
  • OCR models — a job's support documents, knowledge base document extraction, and OCR previews (pipeline editor, knowledge base extraction test). Tesseract, built in and free, isn't counted.
  • Embedding models — indexing of knowledge bases and documents, and the vectorization of every search (estimated tokens).
  • Decision models — switch, interceptors, annotation filter, reranking, memory pre-filter, judges.

Background calls with no user (scheduled knowledge base reindex) count in the summary and the per-model breakdown, without showing in the per-user view. Only the model forms' "Test" buttons aren't counted: they check a configuration not saved yet, so with no rate.

The same records feed Statistics (tokens) and each user's daily token quota.

Available views​

  • Summary — total cost, number of pages processed, cost per page, input/output tokens over the period.
  • By user — consumption breakdown per account.
  • By portfolio — breakdown per pipeline, with cost per page for each pipeline.
  • By model — breakdown per AI model used, with dedicated views per OCR, embedding and decision model.
  • By category — separates normal production usage, pipeline testing, and benchmarking, so real product cost isn't muddied by qualification activity.

There is currently no cost breakdown by team.

Currency and pricing​

The display currency (ISO code, e.g. EUR) is configured globally for the organization — see Application Settings. Each model's rate (price per million tokens, input and output) is set directly from this Finance tab, and applies to all subsequent calls.