Getting Started with Cornelius

1. Clone and configure

git clone https://github.com/Abilityai/cornelius.git
cd cornelius

# Create settings from template
cp .claude/settings.md.template .claude/settings.md

Edit .claude/settings.md and set your vault path. The default, VAULT_BASE_PATH=./Brain, points at the pre-seeded knowledge base that ships with the repo; point it at your own Obsidian vault (relative or absolute) to work on your notes instead:

VAULT_BASE_PATH=/Users/yourname/Documents/YourVault

2. Set up Local Brain Search

The core search engine — FAISS vector search that runs locally. It installs sentence-transformers (embedding model), faiss-cpu (vector search), and networkx (graph analytics).

cd resources/local-brain-search
python -m venv venv
source venv/bin/activate  # macOS/Linux
# venv\Scripts\activate   # Windows
pip install -r requirements.txt

# Index your vault
./run_index.sh

Indexing builds the FAISS index, a knowledge graph with explicit + semantic edges, and metadata for every note. The first run may take a few minutes depending on vault size; later runs reuse the embedding of every unchanged note.

3. Start and test

cd ../..
claude

In Claude Code:

/search-vault test

If that returns results, you're done.

# In resources/local-brain-search, with the venv active
./run_search.sh "knowledge management"
./run_connections.sh "Some Note Name"
./run_connections.sh --stats

Then, in Claude Code:

/search-vault knowledge management
/find-connections "Note Name"
/analyze-kb

When you add or change notes, re-index — from the terminal or from Claude Code:

cd resources/local-brain-search
source venv/bin/activate
./run_index.sh

# or, in Claude Code
/refresh-index

To update Cornelius itself, keep your settings across the pull:

cp .claude/settings.md .claude/settings.md.backup
git pull origin main
cp .claude/settings.md.backup .claude/settings.md

# Update dependencies
cd resources/local-brain-search && source venv/bin/activate
pip install -U -r requirements.txt

Uninstalling is deleting the project directory. Your vault stays untouched; the FAISS index lives inside the project directory.

MCP servers are optional. Core functionality — semantic search and connection discovery — runs on Local Brain Search, which needs only Python. MCP servers add extras like diagram generation and ebook processing.

Configure them in .mcp.json at the project root — a template ships at .mcp.json.template:

cp .mcp.json.template .mcp.json

{
  "mcpServers": {
    "mermaid-diagram": {
      "command": "npx",
      "args": ["-y", "@anthropic/mcp-mermaid-diagram"]
    },
    "ebook-mcp": {
      "command": "uvx",
      "args": ["ebook-mcp"]
    }
  }
}

Restart Claude Code and type /mcpto confirm the servers loaded. Smart Connections is deprecated — if it is still in your .mcp.json, you can safely remove it.

Trinity is the operating system for AI-native companies — open source and self-hosted. Each agent runs in an isolated Docker container with cron scheduling, real-time monitoring, and agent-to-agent delegation. Don't have an instance yet? See Deploying Trinity.

The fastest path is the trinity plugin from the abilities marketplace. From inside your Cornelius directory:

# Add the marketplace and install the plugin (one-time)
/plugin marketplace add abilityai/abilities
/plugin install trinity@abilityai

# Then, from inside Cornelius:
/trinity:connect    # one-time auth
/trinity:onboard    # deploy

On Trinity, Cornelius also gets the Brain Orb— a live 3D visualization of the knowledge base on the agent's Brain tab (requires a Trinity base image from 2026-07 or later, with the platform's Brain Orb flags enabled). See How It Works.

What the template declares

template.yaml is the Trinity-compatible agent configuration. The parts that matter when you deploy:

Scheduled playbooks run through /scheduled-run <skill>, the wrapper for cron automation that handles git sync before and after execution.

Full guides in the repo: QUICKSTART.md, INSTALL.md, and MCP-SETUP.md.