OpenRouter Auto

Use any AI model through one SDK. TypeScript · React · Python · Go · Rust.

---

What is OpenRouter?

OpenRouter provides a single API to access models from Anthropic, OpenAI, Google, and dozens of other providers.

Why use OpenRouter:

• Access hundreds of models from major AI providers through one API
• Automatic failover and load balancing across providers
• No vendor lock-in — switch models with one line change

Get your API key:

1. Create an account at openrouter.ai
2. Add credits and generate an API key at openrouter.ai/settings/keys
3. Use your key with any SDK below

---

Why OpenRouter Auto?

Without this SDK	With OpenRouter Auto
Hardcode model IDs	Auto-fetch all 345+ models
Manual parameter config	Validation from model capabilities
No cost preview	Real-time cost estimation
Basic chat only	Streaming, reasoning, tools, vision, web search
One language	TypeScript · React · Python · Go · Rust
Duplicated config	Single registry/ JSON — zero drift

---

Install

TypeScript / React

npm install @openrouter-auto/core
# or for React components:
npm install @openrouter-auto/react

Python

pip install openrouter-auto

Go

import ora "github.com/faraz152/openrouter-auto-connect/go"

Rust

[dependencies]
openrouter-auto = { git = "https://github.com/faraz152/openrouter-auto-connect", subdirectory = "packages/rust" }

---

Quick Start

Every SDK follows the same three steps: init → fetch models → chat.

TypeScript:

import { OpenRouterAuto } from "@openrouter-auto/core";

const or = new OpenRouterAuto({ apiKey: process.env.OPENROUTER_API_KEY! });
await or.initialize();

const response = await or.chat({
  model: "openai/gpt-4o-mini",
  messages: [{ role: "user", content: "Hello!" }],
});
console.log(response.choices[0].message.content);

Python:

from openrouter_auto import create_openrouter_auto
from openrouter_auto.types import ChatRequest, ChatMessage

sdk = create_openrouter_auto({"api_key": "sk-or-v1-..."})
await sdk.fetch_models()

resp = await sdk.chat(ChatRequest(
    model="openai/gpt-4o-mini",
    messages=[ChatMessage(role="user", content="Hello!")],
))
print(resp.choices[0]["message"]["content"])

Go:

client, _ := ora.NewClient(ora.Options{APIKey: os.Getenv("OPENROUTER_API_KEY")})
client.FetchModels()

resp, _ := client.Chat(ora.ChatRequest{
    Model:    "openai/gpt-4o-mini",
    Messages: []ora.ChatMessage{{Role: "user", Content: "Hello!"}},
})
fmt.Println(resp.Content())

Rust:

let client = Client::new(Options { api_key: key, ..Default::default() })?;
client.fetch_models().await?;

let resp = client.chat(&ChatRequest::new(
    "openai/gpt-4o-mini",
    vec![ChatMessage::new("user", "Hello!")],
)).await?;
println!("{}", resp.content());

React:

import {
  OpenRouterProvider,
  ModelSelector,
  useOpenRouter,
} from "@openrouter-auto/react";

function App() {
  return (
    <OpenRouterProvider apiKey={process.env.REACT_APP_OPENROUTER_API_KEY!}>
      <Chat />
    </OpenRouterProvider>
  );
}

function Chat() {
  const { chat } = useOpenRouter();
  const [model, setModel] = useState<string | null>(null);

  return (
    <div>
      <ModelSelector value={model} onChange={setModel} />
      <button
        onClick={() =>
          chat({ model, messages: [{ role: "user", content: "Hello!" }] })
        }>
        Send
      </button>
    </div>
  );
}

---

Python CLI

The Python package includes a command-line tool for quick model management.

Command	Description
openrouter-auto setup	One-time API key setup
openrouter-auto models	List all 345+ models
openrouter-auto models --free	List free models only
openrouter-auto add <model>	Add and validate a model
openrouter-auto add <model> --temperature 0.7	Add with parameters
openrouter-auto test <model>	Test a saved model
openrouter-auto chat <model> "prompt"	Send a chat message
openrouter-auto chat <model> "prompt" --stream	Stream a response

export OPENROUTER_API_KEY=sk-or-v1-...

openrouter-auto models --free
openrouter-auto add anthropic/claude-3.5-sonnet --temperature 0.7
openrouter-auto chat anthropic/claude-3.5-sonnet "Hello!"

---

SDK API Reference

All five SDKs expose the same core methods. Names follow each language's conventions.

Core Methods

Method	TypeScript	Python	Go	Rust	Description
Initialize	initialize()	fetch_models()	FetchModels()	fetch_models()	Fetch all 345+ models from OpenRouter
Chat	chat(req)	chat(req)	Chat(req)	chat(&req)	Send a chat completion request
Stream	streamChat(req)	stream_chat(req)	StreamChat(req)	stream_chat(&req)	Stream a chat response (SSE)
Add Model	addModel(id, params)	add_model(id, params)	AddModel(id, params)	add_model(id, params)	Save a model config with validation
Get Models	getModels()	get_models()	GetModels()	get_models()	Return cached model list
Filter	filterModels(opts)	filter_models(opts)	FilterModels(opts)	filter_models(&opts)	Filter by price, provider, context, etc.
Cost	calculateCost(...)	calculate_cost(...)	CalculateCost(...)	calculate_cost(...)	Estimate cost for a given token count
Events	on(event, fn)	on(event, fn)	On(event, fn)	on(event, fn)	Subscribe to SDK events

Filter Options

or.filterModels({
  freeOnly: true, // free models only
  search: "claude", // search by name or ID
  provider: "anthropic", // filter by provider
  minContextLength: 100000, // minimum context window
  maxPrice: 0.001, // max price per token
});

Events

Event	Payload	When
models:updated	{ count }	After models are fetched
model:added	{ modelId }	After a model config is saved
chat:success	{ model, tokens }	After a successful chat
chat:error	{ model, error }	After a failed chat
error	{ code, message }	Any SDK error

---

Features

Streaming + StreamAccumulator

All SDKs include a StreamAccumulator that collects streamed chunks into a final response — content, reasoning, and tool calls reassembled automatically.

const acc = new StreamAccumulator();
for await (const chunk of or.streamChat(request)) {
  acc.push(chunk);
  process.stdout.write(acc.content);
}
const response = acc.toResponse();

acc = StreamAccumulator()
async for chunk in sdk.stream_chat(request):
    acc.push(chunk)
    print(acc.content, end="", flush=True)
response = acc.to_response()

Reasoning Models

Models like DeepSeek-R1 and MiniMax M2.7 emit a separate reasoning field. StreamAccumulator captures both.

const resp = await or.chat({
  model: "deepseek/deepseek-r1",
  messages: [{ role: "user", content: "Solve: 120km in 2h, speed?" }],
  reasoning: { effort: "high" },
});
// resp.choices[0].message.reasoning  → chain-of-thought
// resp.choices[0].message.content    → final answer

Tool Calling

Standard OpenAI-compatible function definitions. Works with streaming too.

const resp = await or.chat({
  model: "openai/gpt-4o-mini",
  messages: [{ role: "user", content: "Weather in Tokyo?" }],
  tools: [
    {
      type: "function",
      function: {
        name: "get_weather",
        description: "Get weather for a city",
        parameters: {
          type: "object",
          properties: { city: { type: "string" } },
          required: ["city"],
        },
      },
    },
  ],
  tool_choice: "auto",
});

Web Search

Built-in helpers to add server-side web search to any request.

import { enableWebSearch } from "@openrouter-auto/core";

const resp = await or.chat(
  enableWebSearch({
    model: "openai/gpt-4o-mini",
    messages: [{ role: "user", content: "Latest AI news?" }],
  }),
);

from openrouter_auto import enable_web_search

resp = await sdk.chat(enable_web_search(ChatRequest(
    model="openai/gpt-4o-mini",
    messages=[ChatMessage(role="user", content="Latest AI news?")],
)))

Vision / Multimodal

Pass images to any vision-capable model.

const resp = await or.chat({
  model: "openai/gpt-4.1-mini",
  messages: [
    {
      role: "user",
      content: [
        { type: "text", text: "Describe this image." },
        {
          type: "image_url",
          image_url: { url: "https://example.com/img.png" },
        },
      ],
    },
  ],
});

Provider Routing & Fallbacks

Control which provider handles requests and define fallback chains.

const resp = await or.chat({
  model: "openai/gpt-4o-mini",
  messages: [{ role: "user", content: "Hello!" }],
  provider: { order: ["OpenAI", "Azure"], allow_fallbacks: true },
  models: ["openai/gpt-4o-mini", "openai/gpt-4.1-nano"],
  route: "fallback",
});

Cost Estimation

Real-time cost calculation including reasoning tokens.

const cost = or.calculateCost("openai/gpt-4o-mini", 1000, 500);
console.log(`Total: $${cost.totalCost}`);

cost = calculate_cost(model, prompt_tokens=1000, completion_tokens=500)
print(f"Total: ${cost.total_cost:.6f}")

Error Handling

All SDKs map HTTP errors to typed error codes with user-friendly messages and retry guidance.

try {
  await or.chat({ model: "bad-model", messages: [] });
} catch (err) {
  console.log(err.code); // "MODEL_NOT_FOUND"
  console.log(err.retryable); // false
}

Error Code	HTTP	Retryable	Tip
INVALID_API_KEY	401	No	Check your key at openrouter.ai/keys
RATE_LIMITED	429	Yes	Wait a few seconds before retrying
MODEL_NOT_FOUND	404	No	Refresh the model list
INSUFFICIENT_CREDITS	402	No	Add credits at openrouter.ai/credits
PROVIDER_ERROR	502	Yes	Try a different model
NETWORK_ERROR	—	Yes	Check your internet connection
TIMEOUT	408	Yes	Try again or use a different model

---

React Components

Four ready-to-use components. All state flows through useOpenRouter().

Component	Purpose
<ModelSelector />	Searchable model dropdown with pricing
<ModelConfigPanel />	Parameter form with validation + test button
<CostEstimator />	Live cost breakdown by token count
<ErrorDisplay />	Error message with tips and retry

import {
  OpenRouterProvider,
  ModelSelector,
  ModelConfigPanel,
  CostEstimator,
  ErrorDisplay,
} from "@openrouter-auto/react";

<OpenRouterProvider apiKey="sk-or-v1-...">
  <ModelSelector value={model} onChange={setModel} showPricing />
  <ModelConfigPanel modelId={model} onSave={handleSave} showTestButton />
  <CostEstimator modelId={model} showTextInput />
  <ErrorDisplay onRetry={retry} onDismiss={dismiss} />
</OpenRouterProvider>;

---

Storage Options

All SDKs support pluggable storage via options.storage (or storageType).

Adapter	Runtimes	Persistence	Key
Memory	All	None (default)	"memory"
localStorage	Browser (TS/React)	Per-domain	"localStorage"
File	Node.js, Python, Go, Rust	.openrouter-auto.json	"file"

new OpenRouterAuto({
  apiKey: "...",
  storageType: "file",
  configPath: "./.openrouter-auto.json",
});

---

Configuration

Option	Type	Default	Description
apiKey	string	—	Required. Your OpenRouter API key
baseUrl	string	https://openrouter.ai/api/v1	API base URL
storageType	string	"memory"	"memory" / "localStorage" / "file"
configPath	string	.openrouter-auto.json	Path for file storage
autoFetch	bool	true	Auto-fetch models on init
fetchInterval	number	3600000	Model refresh interval (ms)
cacheDuration	number	3600000	Cache TTL (ms)
enableTesting	bool	true	Test models on add
testPrompt	string	"Say hello"	Prompt used for model tests
onError	function	—	Global error callback
onEvent	function	—	Global event callback

---

Project Structure

openrouter-auto-connect/
├── packages/
│   ├── registry/          # Single source of truth (shared by all SDKs)
│   │   ├── parameters.json
│   │   ├── errors.json
│   │   ├── cost.json
│   │   └── platform-params.json
│   ├── core/              # TypeScript SDK        (52 tests)
│   ├── react/             # React components
│   ├── python/            # Python SDK + CLI      (54 tests)
│   ├── go/                # Go SDK                (22 tests)
│   └── rust/              # Rust SDK              (21 tests)
├── examples/
│   ├── react-basic.tsx
│   └── python-basic.py
├── QUICKSTART.md
├── PROJECT_SUMMARY.md
└── README.md

The registry/ folder holds parameter definitions, error codes, cost tiers, and platform-allowed params as JSON. Each SDK embeds these at build time — zero data duplication across languages.

---

Build & Test

# TypeScript (all packages)
npm install && npm run build && npm test          # 52 tests

# Python
cd packages/python
pip install -e ".[dev]"
pytest tests/ -q                                  # 54 tests

# Go
cd packages/go
go test ./...                                     # 22 tests

# Rust
cd packages/rust
cargo test                                        # 21 tests

# Live E2E (real API, Python)
cd packages/python
export OPENROUTER_API_KEY=sk-or-v1-...
python live_test.py                               # 18 checks

---

Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes — follow the parity rule: features added to one SDK must be added to all
4. Add tests
5. Submit a PR

---

License

MIT — see LICENSE

---

OpenRouter · Get API Key · Quick Start · Full Architecture