Token Compressor

Example Use Cases

• 1

  Reduce token usage and costs when sending prompts to LLMs.

• 2

  That require semantic preservation of prompt meaning including conditionals and negations.

• 3

  Integrate prompt compression seamlessly into their LLM workflows via MCP or CLI.

Description

Token Compressor is a two-stage pipeline that reduces the token count of prompts sent to large language models by rewriting prompts to their semantic minimum and validating the compression via embedding similarity. This reduces token usage and costs without losing the original intent or conditionality. It integrates with Ollama models locally and supports usage as a CLI tool, Claude Code hook, or MCP-compatible server.