Sense launches local AI code intelligence tool — gives coding agents structural understanding of your codebase without sending code to the cloud

Sense (version 0.23.0, MIT license) is a ~40 MB single binary for macOS, Linux, and WSL that indexes your codebase locally and exposes four read-only tools to AI agents: Graph (traces who calls what), Search (finds code by meaning, not just keywords), Blast Radius (shows what breaks if you change a symbol), and Conventions (detects your team's coding patterns in naming, error handling, and structure). Your AI agent automatically calls these tools during reasoning — you never interact with Sense directly.

Instead of forcing AI to read 30 files to answer 'who calls this function?' or hallucinating dependency chains, Sense uses Tree-sitter (a code parser) and bundled ONNX embeddings (mathematical models that understand code meaning) to give AI agents the structural knowledge a senior engineer carries. No API keys, no network calls, no external dependencies — everything runs offline on your machine and your code never leaves it.

For developers and teams using AI coding assistants (Claude, Cursor, or any MCP client), this eliminates a major source of AI error: the agent now makes fewer wrong assumptions about code structure, writes code that matches your team's conventions rather than generic style, and avoids breaking hidden dependencies. For AI agent builders, Sense reduces token usage and reasoning loops by turning vague file-search results into precise structural facts your AI can trust.