lean-ctx

To reduce LLM token usage during codebase exploration, you can compress command output and file content using context-aware reads. This approach applies tree-sitter AST parsing and task-conditioned relevance scoring to filter noisy output and retain only high-signal code paths.

Context compression for AI code assistance is the process of structuring file reads and shell command outputs to minimize token waste. It uses tree-sitter AST parsing to extract syntax-stripped views, exported signatures, and entropy-filtered snippets, ensuring faster and more relevant AI responses.

You can maintain session continuity across multiple AI chat iterations by using a cross-chat continuity protocol that stores knowledge, session state, and recall capabilities. This prevents rework by allowing subsequent tasks to access previously generated context and structured dependency maps.

Yes, you can run shell commands for dev/ops workflows without wasting tokens by executing them through a context-compressed shell pattern. This summarizes verbose outputs from git status, build tests, and infrastructure commands while preserving the critical operational data needed for analysis.

Yes, tree-sitter AST parsing works for understanding API signatures across many programming languages and tooling ecosystems. It enables structured file reading modes that extract exported signatures and dependency maps, providing an API-surface-first context for faster AI comprehension.

Yes, you need curl and tar along with grep, sed, and mktemp to set up a context compression environment. These dependencies support the automated install scripts and the underlying shell execution patterns required for structured file reading and context runtime features.