kbolt

kbolt is the command-line application for local-first indexing and retrieval over local notes and docs.

What it supports

• Register local directories as collections
• Search with keyword, semantic, reranked, and deep modes
• Organize collections into spaces
• Fetch source files with get, multi-get, and ls
• Check readiness with doctor and models list
• Serve the index over MCP with kbolt mcp

Typical flow

Install with Cargo:

cargo install kbolt

Or on macOS and Linux x86_64 with Homebrew:

brew install h3nock/kbolt/kbolt

If llama-server is not already available and you did not install via Homebrew, follow the official llama.cpp install guide.

Then run:

kbolt setup local
kbolt doctor
kbolt collection add /path/to/docs
kbolt search "your query"

kbolt collection add indexes immediately by default.

To enable deep search later:

kbolt local enable deep

This crate is the main user-facing package in the workspace. Most users should install and run kbolt, not the internal kbolt-core, kbolt-mcp, or kbolt-types crates directly.