Features

Handle documentation 100x larger than your context window. Recursive decomposition breaks complex queries into focused sub-queries, reducing hallucination and grounding responses in actual documentation.

Share coding standards, best practices, and prompt templates across your team. Ensure consistent LLM responses that follow your organization's guidelines.

Understand when to use keyword, semantic, or hybrid retrieval. Tune recall quality, latency, and grounding for the query type you actually have.

Preserve context across compaction events with PreCompact and SessionStart hooks. Your LLM remembers what it learned, even after context window resets.

Run reindexing, health checks, and memory hygiene from GitHub Actions, cron, or database schedulers without adding a proprietary jobs layer.

Connect your repository once and keep documentation current automatically. Every push triggers re-indexing of changed files.

Autonomous agent runner with observe-think-act loops, sub-LLM orchestration, and sandboxed code execution (Docker, local, WebAssembly). MCP server integration for Claude Code with 7 tools.

Monitor documentation coverage, chunk quality, and search performance. Get real-time health scores, latency percentiles, tool usage breakdowns, and actionable recommendations.