Llmtrim operates as a local proxy that intelligently compresses your LLM API requests before they reach the provider, significantly reducing token usage and associated costs. By sitting between your AI tools and the LLM service, it identifies and removes redundant or wasteful tokens from prompts, history, tool output, and code, ensuring you receive identical answers for a smaller bill. The tool guarantees that compression never increases your bill or breaks a request; if a step doesn't save tokens or the provider rejects a compressed request, the original is sent. Everything runs locally on your machine, with no data ever sent externally, ensuring privacy and security.
主な機能
01Functions as a local proxy for transparent LLM API request compression.
02Achieves significant token savings (e.g., -31% input, -74% output) without altering model responses.
03133 GitHub stars
04Offers multiple interfaces including a proxy, CLI, MCP server, and libraries for Python, Ruby, Swift, Kotlin, JS/TS, and WASM.
05Guarantees request integrity by never increasing costs or breaking requests, reverting to the original if compression is unsuccessful.
06Ensures privacy and security by running entirely locally, with no data ever sent externally.
ユースケース
01Optimizing prompts, conversation history, and raw tool outputs for more efficient token utilization.
02Reducing operational costs for LLM API interactions when using AI agents and development tools.
03Embedding token compression directly into applications using its multi-language libraries for improved cost-efficiency.