Requesty
Integrations

Works with the stack you already have.

An OpenAI compatible API behind one base URL. The OpenAI SDK, LangChain, PydanticAI and the Vercel AI SDK work without a wrapper.

Speak to founders
OpenAI compatiblestreaming, tool calls, structured outputs600+ models
client.py
2 lines changed
client = OpenAI(
    - base_url="https://api.openai.com/v1",
    + base_url="https://router.requesty.ai/v1",
    - api_key=OPENAI_API_KEY,
    + api_key=REQUESTY_API_KEY,
)

# everything else in the file is unchanged
SDKs

The SDK you already use

Requesty speaks the OpenAI API. The official Python and JavaScript packages work untouched, and there is no Requesty SDK to keep current.

python
same endpoint
from openai import OpenAI

client = OpenAI(
    base_url="https://router.requesty.ai/v1",
    api_key=REQUESTY_API_KEY,
)
typescript
import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://router.requesty.ai/v1",
  apiKey: process.env.REQUESTY_API_KEY,
});
curl
curl https://router.requesty.ai/v1/chat/completions \
  -H "Authorization: Bearer $REQUESTY_API_KEY" \
  -d '{"model": "anthropic/claude-opus-5", "messages": [...]}'
The OpenAI SDK guide in the docs
Compatibility

Nothing in your code has to change

Streaming, tool calls and structured outputs keep the shape your application already sends. Switching models becomes a string change.

  • 600+ models, one shapeAcross 30+ providers
  • Any languageIf it speaks HTTP, it works
Streaming in the docsStructured outputs in the docs
compatibility
OpenAI shaped
Chat completionsstreaming and non-streaming
Tool and function callingunchanged request shape
Structured outputsJSON schema enforced across models
Embeddingssame endpoint, same SDK
Images and PDFsmultimodal models, one API
Reasoning effortunified across providers
Frameworks

A guide for the framework you use

Each one is a base URL and a key, with a docs page to copy from.

OpenAI SDKSwap the base URL in the Python or JS packageLangChainUse Requesty in chains and agentsPydanticAITyped responses against 600+ modelsVercel AI SDKThe official Requesty provider packageHaystackDrop into Haystack pipelinesLlamaIndex TSUse with the TypeScript buildPython RequestsPlain HTTP calls, no SDKAxiosPlain HTTP calls from JavaScript
Agent harnesses

The tools your team codes in

Coding agents point at the same endpoint as your application. The models a developer reaches for in the editor are the ones you already approved and can already see.

integrations / agent harnesses
15 tools
Claude CodeTwo environment variables and the CLI is on the gatewayClaude CoworkThe same key, for the desktop workspaceClinePick Requesty as the provider in the panelRoo CodeChoose Requesty, then any approved modelVS CodeSwitch models without leaving the editorGitHub CopilotBYOK endpoint in Copilot ChatCodexPoint the CLI at the base URLOpenCodeAdd a provider block to the config fileOpenClawOne base URL for 600+ modelsPiSet the endpoint and key in the configAnthropic SDKsRequesty as the backend for the agent SDKsDeepSeekThe harness, on any model you allown8nWorkflow automation through one endpointLibreChatSelf-hosted chat UI on the gatewayOpenWebUIMulti-model chat for the whole team
The coding agents page, in fullApprove which models they may use

Change the base URL, keep the code

One endpoint for 600+ models, and your SDK never knows the difference.

Speak to founders