Skip to content

Latest commit

 

History

History
71 lines (48 loc) · 2.48 KB

CLAUDE.md

File metadata and controls

71 lines (48 loc) · 2.48 KB

CLAUDE.md

Claude Code guidance for working with the llmoxy repository.

Quick Reference

Direct Moonshot Access (Recommended):

  • kimi - Kimi K2 via Moonshot's /anthropic endpoint
  • kimi-turbo - Kimi K2 Turbo Preview via Moonshot's /anthropic endpoint

Proxy-based (Fallback):

  • kimi-or - Kimi K2 via OpenRouter
  • dsr1 - DeepSeek R1 via OpenRouter
  • qwen - Qwen3 Coder via OpenRouter

Management:

  • llmoxy --start/--stop/--restart - Proxy control
  • llmoxy --status - Health check and model listing
  • llmoxy --logs - View container logs
  • llmoxy --generate-aliases - Update aliases from config.yaml

Key Files & Their Purpose

  • llmoxy: Main executable (bash script)
  • config.yaml: LiteLLM model configurations for proxy-based models
  • aliases.sh: Shell aliases (auto-generated + manual direct access)
  • .env: API keys (OPENROUTER_API_KEY, MOONSHOT_API_KEY)
  • docker-compose.yml: LiteLLM proxy container definition

Architecture Notes

Direct Access: kimi/kimi-turbo bypass the proxy entirely, connecting directly to https://api.moonshot.ai/anthropic with environment variable overrides for better caching.

Proxy Access: Other models route through local LiteLLM proxy (localhost:4000) which translates to OpenRouter APIs.

Environment Variable Pattern: llmoxy temporarily sets ANTHROPIC_AUTH_TOKEN, ANTHROPIC_BASE_URL, and ANTHROPIC_MODEL when launching Claude Code, then cleans up on exit.

Development Commands

# Test proxy health
curl http://localhost:4000/health

# View available proxy models
curl http://localhost:4000/models

# Check container status
docker ps | grep llmoxy

# Follow container logs
docker compose logs -f litellm

# Regenerate aliases after config changes
llmoxy --generate-aliases && source aliases.sh

Key Functions (llmoxy script)

  • get_full_model_name() - Maps model nicknames to full paths via config.yaml parsing
  • run_claude_with_model() - Core launcher with environment setup
  • generate_aliases() - Auto-generates aliases.sh from config.yaml
  • proxy_start/stop/restart() - Docker Compose management
  • should_regenerate_aliases() - Detects when aliases need updating

Troubleshooting

Missing API keys: Models will appear in listings but fail at request time if their required API key is missing from .env

Proxy won't start: Check Docker status, port 4000 availability, and API key validity

Aliases not working: Run source aliases.sh after any config changes