How RoGenie saves tokens (up to 11×)
What "up to 11× more usage" means - discounted model rates, minified code context and smart context compression - and how the estimate is made.
On this page
RoGenie says you get up to 11× more usage than using the same models at official list prices, which is up to 91% lower cost. That is RoGenie's maximum estimate, and it comes from three levers working together. Actual savings vary by model, project and workload.
Lever 1: discounted model rates
RoGenie's rates for the same models are discounted against the providers' official list prices. Nothing to turn on: every built-in model's multiplier already reflects it.
Lever 2: minified code, same line numbers
When RoGenie reads your scripts, the model can get a minified copy with comments, indentation and extra spaces removed, typically about a quarter fewer tokens. Every line keeps its line number, so edits land on the exact lines of your real script.
- Your scripts are never stored minified. What you see in Studio and the IDE is always your real code.
- This is marked experimental. Switch it in Settings › Agent › Minified code context.
Lever 3: smart context compression
Large JSON tool outputs, like scene reads, search results and property dumps, are structurally compressed before they reach the model. This cuts token cost on long builds while preserving answers: about 97% fidelity, benchmark-backed.
Switch it in Settings › Agent › Smart context compression.
Other ways RoGenie keeps cost down
- Targeted diffs. RoGenie edits the exact lines that need to change instead of rewriting whole scripts.
- Automatic summarizing. Long chats are compacted so each step re-sends less. See Chat memory and summarizing.
- Efficient pacing. The default optimization mode favours fewer, bigger replies, which cost the least overall. See Agent settings.
- Cheap defaults. Auto (intelligent & cheap) is the default model, and simple sub-agent roles run on it unless you choose otherwise.
How the estimate is made
The "up to 11×" figure compares one build task on one model, paid two ways: at the model's official list price with no savings, and through RoGenie with all three levers. It is a maximum, not a promise: a project with little code to read or few large tool outputs saves less, and the model you choose matters most of all.