caveman

You can reduce token usage by applying output compression that condenses surrounding prose while preserving exact code blocks, API names, error strings, and commands. This ensures the technical payload remains intact during long chat interactions.

Adaptive output compression offers multiple levels—lite, full, ultra, and wenyan—allowing you to match the verbosity reduction to your specific needs. You can activate a specific level by commanding the assistant to enable it.

Yes, compression mode persists across chat sessions using memory persistence guidelines. The active mode is remembered continuously until you explicitly issue a deactivation command to disable it.

Yes, output compression maintains exact codes, API names, error messages, and commands while compressing the surrounding explanatory text. This guarantees that essential technical content is never altered during the compression process.

You activate verbosity reduction by telling the assistant to enable compression with a specific level, such as "caveman full". Deactivation is handled through explicit user commands processed in real time.

You should avoid using output compression when your chat requires fully detailed, uncompressed prose explanations rather than condensed technical payloads. If preserving the original narrative flow is strictly necessary, compression will alter the surrounding text.