tldr: openground is a tool that lets you give controlled access to documentation to AI agents. Everything happens on-device. Think of it as a local alternative to Context7.

I've been working on openground, an opensource and completely on-device tool that let's you give controlled documentation access to your coding agents. Solutions like Context7 provide a source of truth for docs, but their closed-source data ingestion and querying pose security/privacy risks. openground aims to give users full control over what content is available to their agents and how it is ingested. Find your library's documentation source (git repo, sitemap, or local), add it to openground via the CLI, and openground will use a local embedding model and vector db (lancedb) to store your docs. You can then use the CLI to install the MCP server to your agent to allow the agent to query the docs via hybrid BM25 full-text and vector search.

Using fastapi docs as an example, it's as simple as:

uv tool install openground openground add fastapi --source https://github.com/tiangolo/fastapi.git --docs-path docs/ openground install-mcp --opencode # or --claude-code or --cursor

And the agent will have access to locally query the fastapi docs when needed.

Features that I've implemented so far: - docs indexing from git repo, sitemap.xml, or local directories - index specific docs for specific library versions for git sources - efficient updating of existing docs so your machine is not embedding the same content every time you update

This is still an early version, so expect breaking changes. Upcoming features I am working on: - daemon style command that updates docs on a schedule - project specific access control from the MCP server - docs "registry" to allow pushing and pulling of documentation embeddings to and from S3 - lighter-weight package - better docs

Suggestions and PRs welcome! I'll also be around for any discussion.