Use Dragonfly with Cursor
Access 30+ Chinese AI models (DeepSeek, Qwen, Kimi, GLM, Doubao) directly in Cursor. One-time setup, 2 minutes.
Setup
1. Get your API key
Sign up at dragonfly-api.com and create an API key from the Dashboard.
2. Open Cursor Settings
Cmd/Ctrl + Shift + J → Models tab
Or: Cursor menu → Settings → Models
3. Add Dragonfly as a provider
Click "+ Add Model" and configure:
- Provider: OpenAI Compatible
- Base URL:
https://dragonfly-api.com/v1 - API Key: Your
sk-df-...key
4. Add models
Add the models you want to use. Enter the full model ID:
| Model ID | Best For | Cost |
|---|---|---|
deepseek/deepseek-chat | General coding, chat | $0.27/1M |
deepseek/deepseek-reasoner | Complex reasoning, math | $0.55/1M |
qwen/qwen3-235b-a22b | Multilingual, large context | $1.00/1M |
moonshot/kimi-k2 | Long-context coding | $1.00/1M |
doubao/doubao-1.5-pro-256k | Cheapest option | $0.30/1M |
5. Select and use
In the Cursor chat or Cmd+K panel, select your Dragonfly model from the model dropdown. Done.
Recommended Setup
For the best cost/performance balance:
- Chat (Cmd+L):
deepseek/deepseek-chat— great coding ability, ultra cheap - Inline Edit (Cmd+K):
doubao/doubao-1.5-pro-256k— fast, cheapest - Complex tasks:
deepseek/deepseek-reasoner— chain-of-thought reasoning
Tips
- Streaming works — responses stream in real-time, same as built-in models
- Function calling works — Cursor's agent features work with DeepSeek and Qwen
- Switch models anytime — use the model dropdown to switch between Dragonfly models and built-in ones
- Cost tracking — check your usage at dragonfly-api.com/dashboard/usage
Cost Comparison
Using Cursor with Dragonfly vs built-in models:
| Scenario | Built-in (Claude) | Dragonfly (DeepSeek V3) | Savings |
|---|---|---|---|
| 100 chat messages/day | ~$3/day | ~$0.08/day | 97% |
| Heavy refactoring session | ~$5 | ~$0.15 | 97% |
| Monthly (moderate use) | ~$60 | ~$2 | 97% |
Troubleshooting
"Model not found" — Make sure you're using the full model ID (e.g., deepseek/deepseek-chat not just deepseek-chat).
Slow responses — Chinese models route through our gateway. First response may take 1-2s longer than direct connections. Subsequent tokens stream at normal speed.
"Unauthorized" — Check your API key. Make sure it starts with sk-df-.