Cog

Semantic code search for AI assistants. Powered by Nomic embeddings on Apple Silicon.

What it does

• Search code by meaning - Find functions that "handle authentication" not just keyword matches
• Analyze structure - Extract functions, classes, and dependencies from any file
• Generate embeddings - 768-dim vectors on Metal GPU (M1/M2/M3)

Requirements

• macOS with Apple Silicon
• Python 3.12+
• Node.js 18+
• uv

Install

git clone https://github.com/vitorcalvi/cog.git
cd cog

# Python engine
cd packages/core && uv sync

# MCP server
cd ../mcp && npm install && npm run build

Index your code

cd packages/core
uv run cog-index ./your/project

Connect to OpenCode

Edit ~/.config/opencode/opencode.json:

{
  "mcp": {
    "cog": {
      "type": "local",
      "command": ["node", "/ABSOLUTE/PATH/TO/cog/packages/mcp/dist/index.js"],
      "environment": {
        "COG_CORE_DIR": "/ABSOLUTE/PATH/TO/cog/packages/core",
        "COG_DB_PATH": "/ABSOLUTE/PATH/TO/cog/packages/core/cog_memory"
      },
      "enabled": true
    }
  }
}

Replace /ABSOLUTE/PATH/TO with your actual path.

Use in OpenCode

> Search my codebase for user authentication
> Analyze the structure of src/api.py
> Find database connection functions

Tools

Tool	Description
search_code	Semantic search - finds code by meaning
analyze_structure	Extract functions and classes from a file
generate_embedding	Get 768-dim vector for any text

Test it works

Package	Tests	Coverage
Python Core	60	100%
TypeScript MCP	35	98.7%

Run tests:

cd packages/core
uv run pytest
cd packages/mcp
npm test

License

MIT © Vitor Calvi