Product

Retrieval you can trust

Ground combines keyword and vector search with optional RMH (recursive multi-hop) expansion so one surface can return neighboring context across files and packages—not only the top embedding hit.

Citations by default

Results carry paths, scores, and source names so humans and agents can open the same file in an editor and verify. Chat responses are generated with the same retrieved context and carry citations.

Your sources

Index Git repositories, documentation sites, PDFs, and public npm or PyPI packages. Global package catalog is available for quick start; tenants add their own sources for private or org-specific knowledge.

On-demand package search

Search many public packages without pre-indexing: fetch, chunk, and rank in one request when you need a symbol or pattern before committing it to a source.

MCP for IDEs and agents

Expose Ground as an MCP server so Cursor, Claude, and other hosts call search and tools with a stable contract. Configuration is a small JSON block; see the docs for the exact CLI and env vars.

Full MCP guide: docs.trygroundai.com/guides/mcp

HTTP API and SDKs

REST endpoints for search, chat, sources, jobs, keys, and quotas. Official TypeScript and Python clients wrap common flows; everything works with API keys suitable for server-side and agent runtimes.

Rover (deep GitHub)

Optional autonomous GitHub exploration for questions that need listing, reading, and globbing across a repo—planned as a separate mode from everyday hybrid search.