smart-context

Extract only relevant context from your codebase graph instead of loading entire files. Smart Context applies graph-based filtering on signatures, references, and dependencies to achieve ~72% token savings while maintaining ~92% accuracy, using configurable depth levels (0-5) and relation scopes.

Context-level extraction selectively loads information at different depths—Level 0 through Level 5—from a codebase graph. This strategy identifies only signatures, call relationships, and dependencies relevant to your target function or class, minimizing unnecessary token consumption.

Yes. Query the codebase graph to load only related nodes showing where your target function or class is referenced. This partial loading strategy reveals call chains and dependencies without loading entire files, saving tokens and focus.

Selective extraction from codebase graphs delivers approximately 72% token reduction with around 92% accuracy. Results depend on configuring appropriate depth levels and relation scopes to match your analysis scope.

Yes. Smart Context requires a codebase-graph as input—a prerequisite that maps signatures, references, and dependencies across your project. This graph structure enables the selective, token-efficient extraction of relevant context.

Level 0 captures minimal direct information; higher levels (up to Level 5) progressively include related signatures, transitive dependencies, and call chains. Adjust depth to balance token budget against comprehensiveness for your analysis task.