Tokens Are the New Compute Bill
Every agent step spends tokens: reading context, reasoning, calling tools and writing results. At enterprise scale, small inefficiencies compound into slow workflows and unpredictable invoices.
NOVA treats tokens as a budget, not a side effect. Each workflow run is planned, measured and routed to the cheapest model that can meet its quality bar.
Three Levers That Move the Number
Model routing: send extraction and classification to fast models, and reserve deep reasoning engines for the steps that genuinely need them.
Context budgeting: retrieve fewer, better chunks with hybrid search and reranking instead of stuffing entire documents into the prompt.
Result caching: reuse deterministic intermediate outputs across runs so repeated workflows do not pay twice for the same work.
“The cheapest token is the one your workflow never needed to send.”
Measuring the Gains
Teams that adopt routing and context budgeting typically see lower median latency and a flatter cost curve as usage grows, with no drop in evaluation scores.
