Skip to main content

​
What is Ground?

Ground is an evidence-grounded retrieval engine designed to give LLMs reliable, citable context from your codebase and documentation. Unlike traditional search, Ground:
  • Tracks provenance: Every result includes a citation with source, path, line numbers, and version
  • Monitors freshness: Sources have staleness budgets and results show when data may be outdated
  • Enforces trust policies: Configure how stale content is handled and set minimum evidence thresholds
  • Detects conflicts: When multiple sources disagree, Ground surfaces the conflict explicitly

​
Why Ground?

LLMs often hallucinate when answering questions about specific codebases or APIs. Ground solves this by:
  1. Indexing your sources (Git repos, documentation sites, OpenAPI specs, npm/PyPI packages)
  2. Providing grounded retrieval with hybrid search (vector + full-text)
  3. Surfacing quality signals so you know when to trust the answer

Quick Start

Get Ground integrated into your workflow in under 2 minutes

Concepts

Learn the core concepts: sources, jobs, search, citations

SDKs

Official SDKs for TypeScript and Python

API Reference

Explore the REST API for integrating Ground

​
Key Features

  • Git Repositories: Index code files with language-aware chunking
  • Documentation Sites: Crawl HTML docs via sitemap or URL list
  • OpenAPI Specs: Parse and index API operations with schema tracking
  • Packages: Official support for npm and PyPI registries
Combines vector similarity (semantic search) with full-text search (keyword matching) for optimal retrieval.
Every result includes:
  • Source name and ID
  • File path or URL
  • Line numbers (for code)
  • Version reference (commit SHA or doc version)
Configure:
  • Staleness budgets (how old is too old?)
  • Source priorities (weight OpenAPI higher for API questions)
  • Refusal thresholds (refuse to answer with insufficient evidence)
When the same API endpoint is defined differently in multiple sources, Ground detects and surfaces the conflict.