lean-ctx
An open-source local context layer for AI coding agents, mainly helping developers working on large repos and noisy command output produce lower-token coding sessions and tool-call context.
Tool overview
Based on the available evidence, lean-ctx looks worth trying for developers who already rely heavily on AI coding agents and are willing to configure a local workflow, but it is not yet proven enough to treat as a universally validated cost-optimization default. The strongest adoption signals come from the official GitHub repo, its clearly stated positioning around Cursor, Claude Code, and Codex workflows, and frequent release activity. By contrast, reposts on X, roundup posts, and package-port mentions are better treated as proof of attention than proof of repeatable performance.
What it appears to do in practice is not replace the model, and it is not best understood as a generic RAG system, IDE, or hosted inference gateway. A better analogy is a local context-shaping or compression layer for coding agents. The evidence supports capabilities such as deciding what the agent can read, compressing reads, caching context, and controlling memory writes to reduce noisy inputs like file contents and shell output.