CodexA

Appearance

Quick Start ​

Get from zero to semantic code search in under 60 seconds.

1. Initialize & Index ​

bash
cd /path/to/your-project
codexa init --index  # Creates .codexa/ and builds index in one step
codexa doctor        # Verify everything is healthy

Or step by step:

bash
codexa init          # Creates .codexa/ directory (auto-detects best model for your RAM)
codexa index .       # Index the entire codebase

Model Profiles

Pick an embedding model tier to match your hardware:

bash
codexa init --profile fast       # Tiny model, <1 GB RAM
codexa init --profile balanced   # Default, ~2 GB RAM
codexa init --profile precise    # Best quality, ~4 GB RAM

Compare models on your codebase: codexa models benchmark

2. Search Your Code ​

bash
# Semantic search (natural language)
codexa search "authentication middleware"

# Get JSON output for tooling
codexa search "error handling" --json

3. Explore Symbols ​

bash
# Explain a function or class
codexa explain process_payment

# Get rich context
codexa context process_payment

# Trace dependencies
codexa deps src/api/handlers.py

4. AI-Powered Analysis ​

Requires LLM Configuration

Set up an LLM provider in .codexa/config.json first. See Installation.

bash
# Ask questions about your code
codexa ask "How does the authentication flow work?"

# Multi-turn conversation
codexa chat

# AI code review
codexa review src/api/auth.py

# Autonomous investigation
codexa investigate "Find all security vulnerabilities"

5. Quality & Metrics ​

bash
# Analyze code quality
codexa quality src/

# Find risky hotspots
codexa hotspots

# Impact analysis (what breaks if I change this?)
codexa impact

# Enforce quality gates in CI
codexa gate

6. VS Code Integration ​

Add CodexA as a Copilot instruction source for your project:

bash
mkdir -p .github

Create .github/copilot-instructions.md referencing CodexA commands. See the AI Agent Setup for the full integration guide.

Configure VS Code settings:

json
{
  "github.copilot.chat.codeGeneration.instructions": [
    { "file": ".github/copilot-instructions.md" }
  ]
}

Now Copilot Chat in Agent mode will automatically use CodexA for code understanding.

7. Start the Bridge Server (Optional) ​

For persistent IDE or agent connections:

bash
codexa serve --port 24842

Agents can then call http://127.0.0.1:24842/tools/invoke directly.

What's Next? ​