Reduce chat-history tokens while keeping the facts the next turn needs

Classify old turns

Mark a turn as active requirement, durable fact, completed work, obsolete attempt or raw evidence. Active requirements and durable facts should remain explicit. Completed work can often become a concise state summary.

Summarize with structure

A useful summary states the goal, current state, decisions, unresolved problems, exact identifiers and next action. Avoid vague phrases such as “continue as before” because the model may no longer have the earlier context.

Use a quality gate

Test follow-up questions that depend on old facts. Confirm that tool calls, names and constraints remain correct. Varion can compact eligible history and use passthrough when preservation checks fail.

Measurement checklist

1. Choose a representative completed task, not an artificial one-line prompt.
2. Record the selected model, provider input, cached input, output, retries and final result.
3. Change one optimization mechanism at a time so the cause remains visible.
4. Verify required identifiers, tool calls, code changes or business fields.
5. Keep passthrough available when the reduced request does not pass.

How Varion fits

Varion Token Engine is a gateway and testing platform for reducing eligible input-token waste across supported AI traffic. It reports original and provider-bound input, keeps provider charges separate, and does not claim that every request can be reduced. New verified users receive 100,000 processed input tokens and 50 local test runs.