omniroute-compression

To compress token-heavy LLM request payloads, you can apply stacked mode combining RTK and Caveman compression to shrink terminal output and chat history, reducing token size by up to 95% to fit context windows.

The best way to reduce API costs for large chat history and terminal outputs is using stacked compression pipelines, which apply Caveman prose filtering and RTK intent preservation to achieve 78–95% token savings.

Yes, you can filter MCP accessibility-tree snapshots by collapsing repeated lines and truncating verbose text using configurable thresholds, reducing browser snapshot token sizes by 60–80%.

Yes, you need OmniRoute configured with OMNIROUTE_URL and OMNIROUTE_KEY to use token compression, and you can control compression settings at runtime via the /api/settings/compression and /api/compression/preview endpoints.

RTK compression preserves developer intent when shrinking build and test outputs by intelligently filtering terminal, build, test, and git data to achieve 60–90% token reduction without losing critical debugging context.

A limitation of using Caveman compression for chat history inputs is that it achieves about 46% token reduction, which is lower than stacked mode, requiring configurable intensity adjustments for longer sessions.