juyterman1000/entroly

Cut AI context cost without trusting the compressor. Every reduction is reversible, byte-exact recoverable, and carries an auditable receipt. Local-first, works through proxy, MCP, SDK, or agent wrapper.

What it solves

Entroly addresses the high cost, latency, and "distraction" (reduced accuracy) associated with sending large amounts of codebase context to AI coding assistants. It prevents AI models from becoming overwhelmed by irrelevant data while ensuring that no critical information is permanently lost during the compression process.

How it works

Entroly acts as a context assurance layer between the user's code and the AI. It identifies the specific parts of the codebase relevant to a query, selects the necessary evidence, and compresses the remaining content. Crucially, it generates a "receipt" for every selection, providing a handle to recover the original bytes character-for-character if needed. It can be deployed as a CLI tool, a proxy, an MCP server, or via SDKs, allowing it to integrate with existing tools like Cursor, Claude Code, and GitHub Copilot without requiring changes to the agent architecture.

Who it’s for

It is designed for developers using AI coding assistants and engineers building their own AI applications who want to reduce token costs and improve model performance by optimizing the context window.

Highlights

  • Recoverable Compression: Omitted content is not deleted but stored so it can be restored exactly as it was.
  • Auditable Receipts: Every decision to keep or omit text is documented in a receipt for transparency.
  • Token Efficiency: Reduces input tokens to lower API bills and minimize model distraction.
  • Session Rescue: Automatically compacts bulky tool output when a session approaches the provider's context limit to prevent request rejection.
  • Local-First: Offers local verification and simulation tools (entroly verify-claims and entroly simulate) to measure potential savings without an API key.

Related

  • Project
    headroomlabs-ai/headroom

    A local context compression layer for AI agents that reduces prompt and response tokens for JSON, code, and text while preserving critical information.

  • Project
    yvgude/lean-ctx

    A context engineering layer for AI coding agents that compresses inputs and shell outputs to reduce token costs and extend context windows.

  • Project
    larsderidder/context-lens

    A local proxy that captures LLM API calls from AI coding tools to provide a detailed visual breakdown of context window composition and token costs.

  • Project
    jia-gao/leanctx

    A drop-in prompt compression library that reduces LLM input-token costs by 10-40% using loss-tolerance routing to protect critical code and tool calls while compressing prose.

  • Project
    jfrog/boost

    Boost is a command wrapper that compresses noisy shell output into structured context for AI coding agents, reducing token costs without sacrificing task performance.