Mantic
lastmod 2026-01-19created 2026-01-191 edits
A structural code search engine for AI agents
About
- A Structural code search engine developed to solve the slow context search of AI code assistants
- Infers intent by analyzing file structure and metadata, achieving search speeds under 500ms
- Local-First design that works without embeddings, vector databases, or external APIs
- Delivers consistent search performance even in large monorepos (e.g., Chromium with 480,000 files)
- Git-based file scanning prioritizes tracked files, and applies Deterministic Scoring
- Runs as an MCP (Model Context Protocol) server, supporting integration with Claude Desktop, Cursor, VS Code, and more
- Impact Analysis feature supports analyzing the Blast Radius of code changes
- Scoped Search option (--path) can restrict search to a specific directory
- Provides fine-grained CLI options, such as searching code, test, and config files separately
- Search latency of around 0.3–0.4 seconds, up to 25x faster than vector search
- Reduces token usage by up to 63% and minimizes unnecessary file access
- Environment variable-based configuration: controls file count, timeout, ignore patterns, and more
- Dual-license structure of AGPL-3.0 and a commercial license
- Free for internal use and open-source integration
- A separate license is required for integration into commercial IDEs and SaaS