ranxianglei/billion-context

A context-compression plugin for month-long single sessions (billions of tokens) ,small context windows (a 100K context is enough) and token savings (5x fewer tokens) .上下文压缩插件，月级长会话会话， 兼顾小窗口(100k上下文足矣)省token(省5倍token) . billion-context is all you need

What it solves

billion-context is a context-compression proxy that prevents long-lived AI coding sessions from exceeding their model's context window. It reduces token usage (up to 5x fewer tokens) and prevents session degradation or death by compressing consumed conversation history into layered, high-fidelity summaries.

How it works

The tool acts as a proxy between an AI agent (such as Claude Code, Aider, or Cursor) and the model API (Anthropic, OpenAI, etc.). It parses requests and responses, applying the acp-kernel compression logic.

Instead of hard truncation, it injects four specific context-management tools into the conversation that the model can call itself:

  • compress: Folds a range of messages into a detailed summary.
  • decompress: Restores a compressed range when exact details are needed.
  • search_context: Performs keyword searches across summaries and visible messages.
  • acp_status: Provides an overview of context usage and compressible ranges.

Who it’s for

Developers using AI coding agents for long-term projects where conversation history quickly exhausts the context window, leading to increased costs and loss of model coherence.

Highlights

  • Incremental and Reversible: Summaries are written in small ranges and can be decompressed on demand.
  • Prefix-Cache Friendly: Designed to keep the cache prefix intact to maintain performance.
  • Broad Compatibility: Supports a wide array of clients via native plugins, a launcher, or simple URL rewrites.
  • Model-Driven: The model decides when and what to compress rather than relying on a fixed limit.

Related

  • Project
    headroomlabs-ai/headroom

    A local context compression layer for AI agents that reduces prompt and output tokens for tool outputs, logs, and RAG results without losing critical information.

  • Project
    Compresr-ai/Context-Gateway

    A proxy that sits between AI agents and LLM APIs to compress conversation history in the background, preventing delays when context limits are are reached.

  • Project
    rohitg00/agentmemory

    A persistent memory layer for AI coding agents that captures and compresses session data to eliminate the need for re-explaining project context across sessions and tools.

  • Project
    beilusaiying/always-accompany

    A multi-purpose AI agent framework focused on context and attention mechanisms, providing layered memory and dynamic retrieval for long-term companionship, coding, and work.

  • Dispatch
    Anthropic Effective Context Engineering for AI Agents

    Anthropic introduces context engineering as a strategy for curating the optimal set of tokens to maximize LLM performance and overcome attention scarcity and context rot.