- Semantic code search using AI embeddings and vector similarity
- Post-filters semantic results using ripgrep patterns when provided
- Supports semantic metadata filtering (symbol types, complexity, exports, etc.)
- Returns code chunks ranked by semantic relevance to your query

Usage:
- `query` (required): Natural language description of what you're looking for
- `pattern` (optional): Regex pattern to filter semantic results (case-insensitive)
- `limit` (optional): Maximum results (default: 20)
- `include` (optional): File pattern filter (e.g. "*.ts", "src/**/*.tsx")
- `rerank` (optional): Compatibility flag. Remote-only mode keeps semantic ranking as-is
- `minScore` (optional): Minimum relevance score 0-1 (filters low-confidence results)
- `symbol` (optional): Filter by symbol name (e.g. "VectorStore")
- `maxPerFile` (optional): Max results per file (default: 1)
- `maxPerSymbol` (optional): Max results per symbol (default: 1)

Semantic Metadata Filters (optional):
- `symbolType`: Filter by code structure type - "function", "class", "method", "interface", "type", "variable", "namespace", "enum"
- `isExported`: Filter for exported symbols (true/false)
- `minComplexity`: Minimum cyclomatic complexity (e.g. 5 for moderately complex code)
- `maxComplexity`: Maximum cyclomatic complexity (e.g. 10 to avoid overly complex code)
- `hasDocumentation`: Filter for documented/undocumented code (true/false)
- `language`: Filter by language - "typescript", "javascript", "python", "java", "vue"
- `parentScope`: Filter by parent class/namespace (e.g. "VectorStore")
- `imports`: Filter by imported module name (e.g. "react")

How it works:
1. Semantic search finds conceptually relevant code chunks using AI embeddings
2. If `pattern` provided, ripgrep filters to only chunks containing the pattern
3. If `rerank` enabled, the current remote-only build keeps semantic ranking unchanged
4. Top results returned with full code content

When to use sensegrep vs grep:
- sensegrep: Conceptual searches ("find error handling", "authentication logic", "database queries")
- grep: Exact matches ("find TODO comments", "find imports of lodash")

Examples:
- Query: "error handling and retry logic" -> Finds code handling errors with retries
- Query: "user authentication flow", symbolType: "function", isExported: true -> Finds exported auth functions
- Query: "database connection", pattern: "pool" -> Finds DB code containing "pool"
- Query: "complex refactoring candidates", minComplexity: 10, hasDocumentation: false -> Finds complex undocumented code
- Query: "API rate limiting", symbolType: "class" -> Finds rate limiter classes

Note: Requires semantic index. Run `sensegrep index` first if you get an error about missing index.
