kbolt

kbolt indexes local notes, documentation, source code, HTML, and digital PDFs, then searches them with keyword, semantic, reranked, and deep retrieval modes.

Most users start with the local setup path. kbolt setup local downloads the default local models, starts managed llama-server processes, and writes provider bindings into your config.

Remote OpenAI-compatible providers are supported through index.toml. The quickstart documents the local model stack.

Start here

I want to...	Go to
Install kbolt	Install
Get one successful search	Quickstart
Add more directories or spaces	Add and organize content
Keep changed files searchable	Keep indexes fresh
Search better	Search effectively
Read indexed content behind search results	Read indexed content
Use kbolt from Claude Desktop	Use with Claude Desktop
Fix setup or freshness problems	Health and status
Check platform support	Platform support
Look up command details	CLI overview

Common paths

First local index

1. Install
2. Quickstart
3. Keep indexes fresh

Day-to-day search

1. Search effectively
2. Search modes
3. Read indexed content

Agent workflow

1. Use with Claude Desktop
2. Keep indexes fresh
3. MCP tools

Index management

1. Add and organize content
2. Spaces and collections
3. Exclude files

What kbolt does

• indexes Markdown, plaintext, HTML, digital PDFs, and source code from local directories
• groups content into spaces and collections
• searches with keyword, semantic, hybrid reranked, and deep retrieval modes
• reads indexed content through CLI and MCP tools
• keeps collections fresh automatically on macOS and Linux
• runs the default local model stack through managed llama-server processes

Guides solve user jobs. Reference pages give exact command and config details.