Agent Router: An API aggregation gateway that provides unified invocation of multiple large language models.

AgentRouter is a unified Large Language Model (LLM) API aggregation gateway designed for developers, whose core goal is to solve the problem of cumbersome key management and high subscription cost of multi-model platforms. The platform adopts a unified interface compatible with the OpenAI standard, and only needs an API key and a base URL to access and seamlessly switch to the world's most cutting-edge mainstream LLMs, including Claude, GPT, DeepSeek, and GLMs, etc. AgentRouter is mainly developer-friendly, and supports quick login and no complicated registration via GitHub. AgentRouter is developer-friendly, supports quick login via GitHub, eliminates complicated registration process, and provides generous free call credits (up to $200) to new users. The platform is perfectly adapted to the mainstream AI-assisted programming tools on the market (such as Roo Code, Cline, Claude Code, etc.), which is the ideal infrastructure for individual developers and startup teams to compare models, prototype development and AI-assisted coding.

Function List

• Provide a unified API interface of 100% compatible with OpenAI format, realizing zero code modification and seamless switching access of cross-vendor models.
• Support one-click call to the world's mainstream large language models (including Claude series, GPT series, DeepSeek, GLM, etc.).
• Automatic adaptation and seamless access to all major AI programming assistance plug-ins (native support for Roo Code, etc.), Cline, Claude Code, Kilo Code, etc.).
• With a single API key and unified credit pool billing, the cumbersome operation of cross-platform registration, multiple subscriptions and multi-bill management is completely eliminated.
• Minimalist developer onboarding process, supports one-click authorization to log in via GitHub account and instantly generates call keys.
• Highly rewarding newbie API testing benefits for new users and developers ($100 credit for basic signup, up to $200 and above for referral links).

Using Help

I. AgentRouter Core Concepts and Preparedness
The first principle of AgentRouter's design is to “simplify”. In traditional AI application development or AI programming assistance scenarios, if you want to compare the logical reasoning ability of GPT-5 with the coding ability of Claude-3.5, you usually need to register OpenAI and Anthropic accounts, bind different credit cards, and maintain two completely different sets of API call logic.
The AgentRouter breaks down this barrier by acting as a central “API gateway”. You can orchestrate the vast matrix of models behind the AgentRouter as the only network request node.

Step 1: Account Registration and Benefit Amount Collection

1. Open a browser and visit the official AgentRouter website (https://agentrouter.org）。
2. On the home page click on Login via GitHub Authorized login. Currently, the platform is focused on the developer community, and to simplify the process, a GitHub account is used by default for secure and fast login.
3. Once authorization is complete, the system automatically creates a developer account for you and jumps you directly to the console dashboard (Dashboard).
4. About the free amount: For new users, a $100 test credit is issued by default. If you visit and sign up through a developer's Referral Link, you will automatically be issued a large credit (usually between $200 and $225). This credit is free of any platform collateral and can be used to directly call all the top big models supported within the platform.

Step 2: Generate an Exclusive API Key

1. After successfully logging in and entering the Dashboard.
2. Click on Access in the left navigation menu API Keys(Key Management) page.
3. strike (on the keyboard) Create New Key(Create New Key) button. To facilitate subsequent management, the system allows you to set an identifying name for the key (for example:VSCode-RooCode-Key 或 Python-Test-Key）。
4. After the key is generated (in sk- (a long string beginning withCopy it immediately and keep it safe in a password manager or local secure file. Due to strict privacy and security policies, once this pop-up is closed, you will not be able to view the full contents of this key again.

II. Detailed Configuration Guidelines in Major AI Programming Aids
The most popular scenario for AgentRouter is as the underlying model “engine” for major AI code editors. Below is a nanny-level configuration process for a few popular tools:

1. Configuration process in Roo Code (formerly RooProvider) / Cline
Roo Code and Cline are the most powerful Agent plug-ins available for the VS Code editor. By plugging into AgentRouter, you can experience top-notch code automation at zero cost:

• Open the settings panel: Open the Roo Code or Cline plugin in VS Code and click on the gear icon (⚙️) in the upper right corner to enter the settings.
• Switch Service Provider: in API Provider（In the (API Provider) drop-down list, select the OpenAI Compatible(compatible with OpenAI services). Do not select Anthropic or OpenAI directly.
• Configure the base node: in Base URL(Base request address) input box, be sure to fill it in accurately:https://agentrouter.org/v1 (Note: must contain /v1 suffix to match the API routing specification).
• key injection: Paste the API key you just generated in the AgentRouter console into the API Key in the field.
• Specify the model name: in Model ID（In the (Model ID) input box, manually type in the model designator you want to use. For example, to call the Claude model, which is currently the most coded model, enter claude-sonnet-4-5-20250929If you want to use the GPT series, you can fill in the gpt-5。
• Once the configuration is complete, enter the programming requirements directly into the dialog box and enjoy the charm of multi-model collaborative work.

2. Seamless integration in Claude Code (Anthropic's official CLI tool)
If you are used to using the command line Claude Code Fully automated system operation, with environment variables that allow for “steal-and-swap” access:

• Open your operating system terminal (Terminal).
• Enter the following command to configure a global temporary environment variable that forces requests to be redirected to the AgentRouter gateway:

  export ANTHROPIC_BASE_URL="https://agentrouter.org/v1"
  export ANTHROPIC_API_KEY="您的AgentRouter专属API密钥"
  

• After the setup is complete, execute the normal claude The wakeup command is sufficient, and the system will consume your free credits on AgentRouter by default without incurring an official bill.

III. Accessibility Guidelines in Standalone Code Projects (Python / Node.js)
For independent developers who need to bring AI into their own development projects (e.g., intelligent customer service, knowledge base QA systems), AgentRouter is fully compatible with the OpenAI protocol specification, so you can complete the migration with just a two-line code replacement.

1. Python SDK access to standardized code
Make sure you have the latest version of the official OpenAI libraries installed in your virtual environment (running the pip install openai）。

from openai import OpenAI
# 初始化客户端，将 Base URL 和 API Key 指向 AgentRouter
client = OpenAI(
base_url="https://agentrouter.org/v1",  # 关键替换点：指向中转网关
api_key="sk-your-agentrouter-api-key"   # 您的 AgentRouter 密钥
)
def generate_code_snippet(prompt):
try:
response = client.chat.completions.create(
# 在此处无缝切换您想测试的模型，例如 deepseek-chat, gpt-5
model="claude-sonnet-4-5-20250929", 
messages=[
{"role": "system", "content": "你是一个资深的全栈工程师。"},
{"role": "user", "content": prompt}
],
temperature=0.7
)
return response.choices[0].message.content
except Exception as e:
return f"接口请求失败: {str(e)}"
print(generate_code_snippet("请用 Python 编写一段快速排序算法"))

2. Node.js access to standard code
Front-end or back-end full-stack developers can directly use the @openai/openai npm package call:

import OpenAI from 'openai';
// 实例化 OpenAI 客户端
const openai = new OpenAI({
baseURL: 'https://agentrouter.org/v1', // 核心变更项
apiKey: process.env.AGENTROUTER_API_KEY // 建议将密钥存放在 .env 文件中
});
async function main() {
const completion = await openai.chat.completions.create({
messages: [{ role: 'user', content: '简述第一性原理在编程中的应用。' }],
model: 'gpt-5', // 通过修改字符串轻松切换大模型厂商
});
console.log(completion.choices[0].message.content);
}
main();

IV. Model Selection Specifications and Platform Considerations

• Model ID Strict Matching: Calls to the API must be made using the exact model designator that is publicized on the platform. Some version-specific designators may have a date suffix (e.g. claude-3-5-haiku-20241022). Passing in the wrong model name will trigger a 404 or illegal request error.
• Exception handling mechanism: Despite the sizable free quota provided by the platform, the gateway layer may set dynamic Rate Limits for high-frequency concurrency in order to prevent the interface from being maliciously attacked or abused. If you run code or use an IDE that encounters 429 Too Many Requests Reporting errors, it is recommended to add Retry in the call logic or reduce the frequency of instruction sending appropriately.
• Privacy and data desensitization: As an API gateway service, the user's Prompt and generated context need to be relayed and routed through AgentRouter's server. If you are developing enterprise-level applications or dealing with extremely sensitive commercial, medical, or personal data, it is recommended to desensitize the core data before initiating a request. For regular programming assistance, prototyping and open source project development, it is safe to use.

application scenario

1. AI programming assistance with deep IDE integration
   Perfect as a “model backend” for smart programming tools like Roo Code, Cline, Cursor, etc., developers don't need to pay expensive official standalone subscriptions to use top-tier code models like Claude 3.5 Sonnet or GPT, dramatically reducing the cost of day-to-day coding assistance.
2. Performance Comparison and Evaluation of Multiple Language Models
   For AI researchers and developers who need to evaluate the performance of different grand models under specific tasks, simply change the model Passing the parameters, we can compare the response quality and latency performance of different models (e.g., GPT, Claude, DeepSeek) in parallel under one interface.
3. Rapid application prototyping for startup teams
   With the $100-200 in free API credits provided by the platform's registration, startup teams and independent developers can quickly validate the business concept (POC) and functional logic of an AI product at zero cost, eliminating the hassle of configuring infrastructure in the early stages.
4. Unified billing and team API scheduling management
   When complex AI Agent projects need to invoke multiple specialized models at the same time (e.g., GPT for logic judgment, Claude for long text reading), by accessing the gateway, developers only need to maintain a single key to complete the whole process of concurrent invocations, dramatically reducing the management complexity of billing and authentication.

QA

1. Is AgentRouter completely free?
   Not completely free, but comes with generous free benefits. The platform is a non-profit interface gateway that gives new users a base quota of $100 for signing up, and around $200 in calling credits for signing up through a referral link. Until these free credits are utilized, you can call all the major top models absolutely free of charge.
2. What large language models are supported?
   The platform aggregates the most mainstream advanced big models on the market, including but not limited to Anthropic's Claude series (e.g., Sonnet 4.5, Haiku 3.5), OpenAI's GPT series (e.g., GPT-4o, GPT-5), and the excellent home-made DeepSeek depth-seeking models, as well as Smart Spectrum GLM models, etc. Please refer to the console for the latest list of models.
3. How do I replace AgentRouter in my original OpenAI SDK-based code?
   There is no need to rewrite core business code at all. All you need to do is to set up the Base URL modify to https://agentrouter.org/v1and will API Key Modify the key you generated for the platform to seamlessly use all the functionality as if it were an official interface.
4. Is my code and cue word data safe?
   As a request relay gateway, your data will necessarily flow through the platform's routing servers before being forwarded to the final model vendor. While the platform is primarily used to assist developers in reducing costs, it lacks a detailed whitepaper on data retention. Desensitization is recommended when dealing with highly confidential, commercial production-grade sensitive data, while general development, code writing, and personal learning experiments are not a cause for undue concern.