LLM Providers

Supported providers, default models, and operational behavior.

DragonClaw supports two API patterns:

• OpenAI-compatible chat endpoints

• Anthropic Messages API

Supported providers

• DeepSeek — deepseek-chat

• Qwen — qwen-max

• Kimi — moonshot-v1-128k

• GLM — glm-4-flash

• OpenAI — gpt-4o

• Anthropic — Claude models

• OpenRouter — any routed model

• Local — Ollama, vLLM, LM Studio, and similar servers

Why Chinese providers fit DragonClaw

• Lower token cost

• Better Chinese fluency

• Lower latency in Asia-Pacific

• Fewer geoblocking issues

Local model support

Point DragonClaw to any OpenAI-compatible endpoint:

llm:
  provider: local
  baseUrl: http://localhost:11434/v1
  model: qwen2.5:32b
  apiKey: not-needed

Reliability features

All LLM calls include:

• 60-second timeout

• 3 retries with backoff on retryable failures

• Structured provider-aware errors

• Logging for provider, model, latency, and response size