Skip to content

CONFIGURATION.md

Latest commit

 

History

History
150 lines (110 loc) · 3.79 KB

CONFIGURATION.md

File metadata and controls

150 lines (110 loc) · 3.79 KB

Configuration

All the knobs for AI providers, models, and environment settings.

AI Providers

Claude CLI (Recommended - No API Key!)

If you have Claude Code installed, just use it:

AI_PROVIDER=claude
AI_MODEL=haiku

The index generator spawns claude --model haiku under the hood. You're already authenticated, so no extra setup.

Models:

  • haiku - Fast, cheap, perfect for keyword extraction (default)
  • sonnet - Better quality if you want it
  • opus - Overkill but hey, your call

OpenAI

AI_PROVIDER=openai
AI_API_KEY=sk-...
AI_MODEL=gpt-4o-mini

Anthropic (Direct)

AI_PROVIDER=anthropic
AI_API_KEY=sk-ant-...
AI_MODEL=claude-3-haiku-20240307

Ollama (Local)

AI_PROVIDER=ollama
AI_MODEL=llama3.2
AI_BASE_URL=http://localhost:11434/v1

Free, local, no API key. Just need Ollama running.

OpenRouter

AI_PROVIDER=openrouter
AI_API_KEY=sk-or-...
AI_MODEL=anthropic/claude-3-haiku

Access to tons of models through one API.

Custom Endpoint

AI_PROVIDER=custom
AI_API_KEY=your-key
AI_BASE_URL=https://your-api.com/v1
AI_MODEL=whatever-model

Works with anything OpenAI-compatible.

Full Environment Reference

# =============================================================================
# REQUIRED: Pick one provider
# =============================================================================
AI_PROVIDER=claude           # claude, openai, anthropic, ollama, openrouter, custom

# =============================================================================
# PROVIDER-SPECIFIC
# =============================================================================
AI_API_KEY=your-key          # Not needed for claude provider
AI_MODEL=haiku               # Model name (provider-specific)
AI_BASE_URL=                 # Custom endpoint URL (optional)

# =============================================================================
# OPTIONAL: Tuning
# =============================================================================
AI_FALLBACK_MODELS=gpt-4o-mini,gpt-3.5-turbo   # Try these if primary fails
AI_RATE_LIMIT_RPM=20                            # Requests per minute limit
AI_MAX_RETRIES=3                                # Retry failed requests
AI_TIMEOUT=60                                   # Request timeout in seconds
AI_LANGUAGES=english                            # Keyword extraction languages

# =============================================================================
# SKILL PATHS
# =============================================================================
CLAUDE_SKILLS_PATH=/custom/path:/another/path   # Additional skill directories

Custom Skill Paths

By default, skills are discovered from:

  1. ./skills/ - Project-specific
  2. ./.claude/skills/ - Project-specific (alt)
  3. ~/.claude/skills/ - User global
  4. System paths - Platform-specific

Add custom paths via environment:

Windows:

$env:CLAUDE_SKILLS_PATH = "C:\MySkills;D:\TeamSkills"

Linux/macOS:

export CLAUDE_SKILLS_PATH="/path/to/skills:/another/path"

Testing Your Setup

# Test AI connection
python src/skill_activator.py --test-ai

# Debug mode (shows full request/response)
python src/skill_activator.py --test-ai --debug

# Generate index for a skills folder
python src/index_generator.py ./skills

Troubleshooting

"Claude CLI not found"

  • Make sure Claude Code is installed and claude is in your PATH
  • Try running claude --version to verify

"API key invalid"

  • Double-check your key in .env
  • Make sure there are no extra spaces or quotes

"Model not found"

  • Check the model name for your provider
  • OpenRouter models need the full path like anthropic/claude-3-haiku

"Rate limited"

  • Set AI_RATE_LIMIT_RPM to a lower value
  • Or just wait a bit and retry