I built a "context OS" that stops AI agents from drowning in your codebase

Short summary

ContextOS is an offline CLI that intelligently selects relevant files from large codebases for AI agent context, saving 80%+ tokens through keyword matching, import graph analysis, and git churn scoring. It integrates with Claude via MCP, includes 980 tests with 96% coverage, and is Apache 2.0 open-source.

• •Solves context overflow by ranking files by relevance (keyword, import centrality, AST overlap, git churn)
• •Fully offline with MCP integration for Claude; no telemetry or accounts
• •Production-ready: 980 tests, 96% coverage, supports Claude, Cursor, Codex, Aider export formats

Generated with AI, which can make mistakes.