Quickstart

What you'll accomplish

• Add local documents as knowledge base sources.
• Build RAG vectors for fast semantic search.
• Optionally build a KAG graph for multi-hop queries.
• Confirm Conduit search works locally.
• Connect the Conduit MCP server to an AI client.
• Validate end-to-end with a few test prompts.

0) Before you start

• Make sure you've completed Install and setup: /docs/install.
• If anything looks off, use /docs/troubleshooting.
• If conduit status isn't healthy, go back to Install first.

conduit status
conduit doctor

1) Add knowledge base sources

Conduit indexes local folders and files into your private KB. Nothing is uploaded to a cloud service.

Add at least one source folder (you can add multiple):

# A documentation folder
conduit kb add ~/Documents/my-project --name "My Project Docs"
 
# A code repository
conduit kb add ~/code/my-repo --name "My Repo"

If you want to confirm they were added:

conduit kb list

2) Build RAG vectors (KB sync)

This is the "first value" step. It builds full-text and vector indexes.

conduit kb sync

First sync time depends on file count and size. For a reset (use sparingly):

conduit kb sync --rebuild-vectors

3) Verify locally with search

Run a few searches and confirm you get file paths and snippets back.

# Hybrid search (default)
conduit kb search "how does authentication work"
 
# Semantic search
conduit kb search "user login security" --semantic
 
# Keyword search
conduit kb search "OAuth2 client_id" --fts5

Look for top matches and citations (file paths + snippets). If results look empty, re-run conduit doctor and check /docs/troubleshooting.

4) Optional: Build KAG (knowledge graph)

KAG enables entity/relationship reasoning and auditable multi-hop answers, but it takes longer and uses more compute/storage. Use it only when you truly need structured graph reasoning.

# Build the graph from your indexed KB
conduit kb kag-sync
 
# Query the graph
conduit kb kag-query "Kubernetes"
conduit kb kag-query "authentication" --entities OAuth,JWT --max-hops 2

KAG build time scales with corpus size; expect the first run to take longer. For deeper guidance, see /docs/kag.

5) Confirm the MCP server exists (local)

Conduit exposes your KB over MCP so AI clients can query it. Verify the MCP status:

conduit mcp status

You should see the KB MCP server listed and ready. The MCP server itself is launched by the client using:

conduit mcp kb

6) Wire Conduit into an AI client (Claude Desktop)

Add the MCP server config to Claude Desktop, then restart the app.

Config file (macOS): ~/Library/Application Support/Claude/claude_desktop_config.json

Add or merge this snippet into mcpServers:

{
  "mcpServers": {
    "conduit-kb": {
      "command": "conduit",
      "args": ["mcp", "kb"]
    }
  }
}

If you use another client (Claude Code, Cursor, VS Code), Conduit can auto-configure via conduit mcp configure. See /docs/mcp for client-specific guidance.

7) Test it in the AI client

Try these prompts to validate end-to-end behavior:

1. "Search my KB for authentication and cite sources."
2. "Summarize what the KB says about API rate limits; include file names."
3. If KAG is enabled: "Using the knowledge graph, show relationships for Kubernetes."

8) Next steps

• Install: /docs/install
• CLI commands: /docs/cli
• MCP setup: /docs/mcp
• KAG: /docs/kag
• Troubleshooting: /docs/troubleshooting
• Admin guide: /docs/admin

If you get stuck, run conduit doctor and check Known Issues.