Deterministic node IDs are SHA-256 hashes of project::kind::qualified_name to ensure stable IDs across scans for reliable incremental updates and graph diffs.

The system resolves references across languages in two steps: Tree-sitter for precise syntax and positions, followed by language-specific resolvers (e.g., Ty for Python, TypeScript, Go gopls, Rust Analyzer) to create real, edge-backed connections like CALLS, HAS_TYPE, and INHERITS_FROM.

GraphLens supports multiple usage modes: as a library, via CLI (indexing, querying, visualization, Neo4j export, and model context protocol for LLMs), and in CI with strict validation to prevent degraded graphs.

GraphLens is not a database, UI, or embedded execution engine; it focuses on producing a robust cross-language graph IR that other tools can consume for navigation, dependency analysis, and LLM prompting.

GraphLens is an open-source, polyglot code-analysis framework that converts a multi-language codebase into a single, typed graph IR to enable cross-language analysis and tooling.

Edges are made accurate rather than based on simple name matching, enabling precise cross-reference mapping across languages and even cross-language boundaries via boundary nodes and communicates_with edges.

The project uses a plugin architecture, where each language adapter is a separate package registered through a core adapter registry, allowing easy addition of new languages without altering the core.

Its three-layer architecture comprises Language Adapters to parse source files, the GraphLens IR for nodes and typed edges, and a Graph Backend for persistence or querying, with adapters designed as pure data producers that do not perform filesystem or database writes.

Boundaries normalize endpoints so that, for example, a Python FastAPI route and a TypeScript fetch to the same endpoint share the same boundary ID, enabling merged graphs and cross-language analysis.