LLM Capabilities at a Predictable Cost

Features

1. 

   03.1

   Decision Steps

   A decision model answers yes or no, picks from a fixed list, or gives a worded score, priced far below a chat model. One call answers every question about the same input.

   • Ask yes-or-no questions cheaply.
   • Pick one option from a list.
   • Score inputs on a worded scale.
   • Ask several questions in one call.
   • Get a confidence with each answer.
2. 

   03.2

   Prices Before the Spend

   Model and vendor prices are quoted in credits before anything runs. Every synchronous call reports what it charged, and each run’s spend is itemised by step.

   • See prices before anything runs.
   • Compare models by credit price.
   • Know what each call charged.
   • Read costs from a response header.
   • See each run’s cost by step.
3. 

   03.3

   A Ceiling per Kind of Spender

   Separate ceilings cover design-time work, each signed-in user, and anonymous public calls per day, with an early warning. At a ceiling the project turns read-only.

   • Cap what building and testing costs.
   • Cap spend for each signed-in user.
   • Cap anonymous public calls per day.
   • Get warned before a limit hits.
   • Stop spending at the ceiling.
4. 

   03.4

   Models Bound to Tasks

   Each step names a task, such as extraction or summarisation. One setting moves every step on that task to another model.

   • Move every extraction step at once.
   • Run simple tasks on cheaper models.
   • Upgrade a model in one setting.
   • Try a new model project-wide.
5. 

   03.5

   Repeat Work From Cache

   In live runs, a step that has already answered the same inputs is served from its saved result at no charge. A page or prompt already answered costs compute only.

   • Skip paying for repeated answers.
   • Rerun flows without re-asking models.
   • Reuse pages you already fetched.