Search your codebase by meaning, not by string match.

Mind RAG is the semantic search engine we use ourselves to navigate KB Labs. It works at the concept level, returns source citations and confidence scores, and gives AI agents a first-class tool that doesn't hallucinate.

Search accuracy across query complexity

EASY0.63
MEDIUM0.78
HARD0.70
Average: 7.0 / 10

Three modes for every workflow

Choose the depth of analysis that fits your task.

Instant
~30s

Fast lookup for specific classes, functions, or interfaces. Minimal LLM calls.

  • Find a class definition
  • Locate a specific file
  • Quick reference checks
  • Simple lookups
🔄AutoRecommended
~60s

Balanced mode that auto-detects query complexity and adjusts search depth.

  • Understand how features work
  • Explore implementation details
  • Balanced speed and depth
  • Default for most queries
🧠Thinking
~90s

Deep multi-step reasoning for complex architectural questions and cross-module analysis.

  • Architectural deep-dives
  • Cross-module analysis
  • Compare implementations
  • End-to-end flow tracing

Works from your terminal

Terminal
$ pnpm kb mind rag-query --text "How does hybrid search work?" --agent
// Response (JSON)
{
"confidence": 0.87,
"mode": "auto",
"sources": [
"packages/mind-engine/src/search/hybrid-search.ts",
"packages/mind-engine/src/search/rrf-merger.ts"
],
"answer": "Hybrid search combines BM25 keyword..."
}

Hybrid Search

Combines vector embeddings with BM25 scoring via Reciprocal Rank Fusion for results that understand meaning, not just keywords.

Anti-Hallucination

Built-in verification cross-checks every answer against actual source code with confidence scoring and citation tracking.

Agent-Ready

First-class tool for AI agents with structured JSON output, confidence scores, and source file citations.

Incremental Indexing

Configurable scopes with include/exclude patterns. Only re-indexes changed files for fast updates.

Search your codebase by meaning, not by string match.

Semantic code search built for engineers and AI agents — with confidence scores, source citations, and verification against hallucinations.