使用方法
- 复制安装命令,在自己的智能体环境安装前先检查源码包。
- 从原始文档核对所需工具、凭据和支持的宿主。
- 说明输入材料与期望交付物,并对照来源检查输出。
npx skills add https://github.com/mnfst/awesome-free-llm-apis --skill free-llm-apis开始前需了解
- 安装不包含模型订阅、API 额度或访问文件的权限。
- 仓库星数和安装量反映活跃情况,不等于独立质量或安全测试。
仓库原始内容
按源仓库内容呈现,保留标题、案例、代码、表格、链接以及原文引用的演示图片。
Free LLM API Setup
Help users pick a free LLM provider and configure their API key. Every provider here has a permanent free tier, no credit card needed.
Provider Selection
Ask the user what matters most, then recommend accordingly:
| Priority | Best picks |
|---|---|
| Fastest inference | Groq (LPU hardware; ~1,000 RPD on chat models) |
| Largest model selection | NVIDIA NIM (100+ models), Cloudflare Workers AI (60+), Hugging Face (thousands, small credit) |
| Strongest proprietary models | Google Gemini, Cohere (non-commercial only) |
| Free frontier open-weight models | OpenRouter (:free models), Kilo Code, NVIDIA NIM |
| European provider / EU hosting | OVHcloud AI Endpoints (FR, EU data centers), LLM7.io (UK) |
| No signup or key required | OVHcloud AI Endpoints (2 RPM/IP), Kilo Code (200 req/hr/IP), LLM7.io (anonymous tier) |
| Privacy-sensitive prompts | Avoid Gemini free tier outside the EEA, Switzerland and the UK, OpenRouter/Kilo free routes, NVIDIA trial endpoints |
Provider categories
<!-- providers:start --> Provider APIs -- run by the companies that train the models:
- Aion Labs, Cohere, Google Gemini, Z AI (Zhipu AI)
- See references/provider-apis.md for setup instructions.
Inference providers -- third-party platforms hosting open-weight models:
- AxForge, Cloudflare Workers AI, Gonka Broker, Groq, Kilo Code (no key needed), LLM7.io (no key needed), ModelScope, NVIDIA NIM, OpenRouter, OVHcloud AI Endpoints (no key needed), SiliconFlow
- See references/inference-providers.md for setup instructions.
<!-- providers:end -->
Workflow
- Ask what models, rate limits, or features the user cares about. If they already know which provider they want, skip to step 3.
- Match their priorities against the table above. Suggest 1-2 options with a short reason.
- Load the right reference file and walk through the setup steps for that provider (API key, code example, env var).
- Offer a quick test script or curl command so they can confirm the key works.
Quick Test Template
After setup, use this to verify any provider:
python
from openai import OpenAI
client = OpenAI(api_key="KEY", base_url="BASE_URL")
response = client.chat.completions.create(
model="MODEL_NAME",
messages=[{"role": "user", "content": "Say hello in one sentence."}],
max_tokens=50
)
print(response.choices[0].message.content)Replace KEY, BASEURL, and MODELNAME with values from the provider's setup guide.
Key Notes
- All endpoints work with the OpenAI SDK unless noted.
- RPM = requests per minute. RPD = requests per day.
- Google Gemini's free tier works in the EEA, UK, and Switzerland, but its terms require paid services if you ship an app to users there.
- Free tiers, model lists, and limits change often. The reference files are generated from
data.json; check the provider's own docs before relying on a limit. - "Limits undocumented" = the provider doesn't publish rate limits.
- Don't hardcode API keys. Use environment variables or a secrets manager.