# Connect clients

Source: https://docs.getworg.com/connect-clients

Worg speaks the OpenAI Chat Completions API, so most clients only need a base URL, a key and a model ID. Every example on this page was run before it was published.

## The three settings

- Base URL: `https://api.getworg.com/v1`
- API key: `$WORG_API_KEY`
- Model: `deepseek-v4.1-flash:cached`

Enter the base URL exactly as shown, with `/v1` and nothing after it. Clients add `/chat/completions` themselves. Read the key from an environment variable rather than writing it into a config file. Any ID from `GET /v1/models` works as the model, and so does `combo/<name>`.

## OpenAI SDK

Pass the base URL and key when you create the client. The rest of your code stays the same.

Python:

```
import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.getworg.com/v1",
    api_key=os.environ["WORG_API_KEY"],
)

response = client.chat.completions.create(
    model="deepseek-v4.1-flash:cached",
    messages=[{"role": "user", "content": "Say hello in Indonesian."}],
    max_tokens=512,
)
print(response.model)
print(response.choices[0].message.content)
```

TypeScript:

```
import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.getworg.com/v1",
  apiKey: process.env.WORG_API_KEY,
});

const response = await client.chat.completions.create({
  model: "deepseek-v4.1-flash:cached",
  messages: [{ role: "user", content: "Say hello in Indonesian." }],
  max_tokens: 512,
});
console.log(response.model);
console.log(response.choices[0].message.content);
```

## LiteLLM

Put `openai/` in front of the model ID. It tells LiteLLM to use the OpenAI format, and LiteLLM removes it before sending the request.

Python:

```
import os
import litellm

response = litellm.completion(
    model="openai/deepseek-v4.1-flash:cached",
    api_base="https://api.getworg.com/v1",
    api_key=os.environ["WORG_API_KEY"],
    messages=[{"role": "user", "content": "Say hello in Indonesian."}],
    max_tokens=512,
)
print(response.model)
print(response.choices[0].message.content)
```

## LangChain

Use `ChatOpenAI` from `langchain-openai` with Worg's base URL.

Python:

```
import os
from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    model="deepseek-v4.1-flash:cached",
    base_url="https://api.getworg.com/v1",
    api_key=os.environ["WORG_API_KEY"],
    max_tokens=512,
)
reply = llm.invoke("Say hello in Indonesian.")
print(reply.response_metadata["model_name"])
print(reply.content)
```

## Vercel AI SDK

Use the OpenAI-compatible provider. Streaming with `streamText` works the same way.

TypeScript:

```
import { createOpenAICompatible } from "@ai-sdk/openai-compatible";
import { generateText } from "ai";

const worg = createOpenAICompatible({
  name: "worg",
  baseURL: "https://api.getworg.com/v1",
  apiKey: process.env.WORG_API_KEY,
});

const { text, response } = await generateText({
  model: worg("deepseek-v4.1-flash:cached"),
  prompt: "Say hello in Indonesian.",
  maxOutputTokens: 512,
});
console.log(response.modelId);
console.log(text);
```

If you use `@ai-sdk/openai` instead, call `openai.chat("deepseek-v4.1-flash:cached")`. Plain `openai("…")` calls the Responses API, which Worg does not have, and gets 404.

## Coding agents

OpenCode, in `opencode.json`. Then pick the model as `worg/deepseek-v4.1-flash:cached`:

opencode.json:

```
{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "worg": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Worg",
      "options": {
        "baseURL": "https://api.getworg.com/v1",
        "apiKey": "{env:WORG_API_KEY}"
      },
      "models": {
        "deepseek-v4.1-flash:cached": {
          "name": "DeepSeek V4.1 Flash",
          "limit": {
            "context": 1048576,
            "output": 8192
          }
        }
      }
    }
  }
}
```

Hermes Agent, in `config.yaml`:

config.yaml:

```
model:
  default: deepseek-v4.1-flash:cached
  provider: custom
  base_url: https://api.getworg.com/v1
  key_env: WORG_API_KEY
```

Other tools that accept an OpenAI-compatible base URL, a key and a model ID take the same three settings. This page lists only the tools we have run against Worg.

> Give the agent its own key
>
> Coding agents send many requests and sometimes leave out `max_tokens`. Worg then holds the model's whole output limit for each request, so a key with a very small budget gets `402` even though each answer costs little. The error message states the hold. Create a separate key for the agent, with a budget that covers that hold, and revoke it when the work is done. See [Agent context](https://docs.getworg.com/ai-agents#agent-key).

## What does not work

Worg has two endpoints: `POST /v1/chat/completions` and `GET /v1/models`. A client that needs anything else gets 404:

- The Anthropic Messages API. Claude Code and other tools built on it cannot use Worg.
- The OpenAI Responses API, and clients that call only it.
- Embeddings, audio, image generation, files and batches.

Next: [Troubleshooting](https://docs.getworg.com/troubleshooting) · [Agent context](https://docs.getworg.com/ai-agents) · [Models and compatibility](https://docs.getworg.com/models)