billion-context: Compress long coding-agent conversations

What it does

billion-context sits between a coding agent and its model API and compresses older conversation ranges into layered summaries. The model decides when and what to compress; the tool can later search those summaries or restore a range when exact details are needed. This is aimed at long-running coding sessions where past decisions and project context need to remain available. Its documentation lists client-specific launchers or plugins for agents including Claude Code, Codex, OpenCode, pi, Gemini CLI, and others, while also describing client limitations.

How to use it

Install the npm package, follow the instructions for the coding agent you use, and point that agent through the billion-context proxy. As the conversation grows, the agent can call the supplied compress, decompress, search_context, and acp_status tools; the proxy forwards the request to the model API.

Common use cases

• Compress a completed section of a long coding conversation and keep a summary available in the same session.
• Search compressed context for a prior decision when the agent needs it again.
• Restore a compressed message range when the exact earlier details matter to the next code change.