---
title: "Use each provider's official SDK to call · Agent Wranglers"
url: https://agentwranglers.se/playbook/stack/ai-software/official-model-sdk
updated: 2026-10-21
licence: "All rights reserved"
---

# Use each provider's official SDK to call models

By [Agent Wranglers](https://agentwranglers.se/authors/agent-wranglers). Updated 2026-10-21.

- Level: Recommended
- Chapter: [AI software](https://agentwranglers.se/playbook/stack/ai-software)
- Rule: Call each model provider through its own official SDK, inside one module of your own. Add a shared abstraction only when you call two or more providers.
- Cost: Paid. The SDKs are open source, but each provider charges for every call to its model API. Use one only when no open-weight model passes your evals.
- How: A hosted model is a paid service, so pick one only after the open-weight test above fails. Each provider updates its SDK with its API, so new features and fixes land there first. On 21 October 2026 the Python packages `openai` (Apache-2.0), `anthropic` (MIT) and `google-genai` (Apache-2.0) each had a release in the past week. Each had over 36 million weekly PyPI downloads. All three fail the five-year test. `openai` reached 1.0 in November 2023 and `anthropic` in August 2026. The TypeScript `@anthropic-ai/sdk` was still at 0.137. Pin exact versions, because these SDKs change often. A third-party wrapper framework adds a second release cycle between you and the API. LangChain, the most downloaded one, reached 1.0 in October 2025. Keep every call behind one module, as [Call models through one module](https://agentwranglers.se/playbook/ai-features/deciding-to-use-a-model/model-gateway) says. Choose something else when you call two or more providers. Then a thin module of your own, or a hosted gateway, maps one request shape onto each official SDK.
- Check: A script lists every model package in the dependency files. Each one is the official SDK of a provider the project calls, and only files inside the model module import it.

## Examples

**Using these examples:** they show one way to follow this rule. Keep the behaviour and the check, not the syntax. Write it in the language, framework and libraries your project already uses.

### TypeScript example

```ts
import OpenAI from 'openai';

const TIMEOUT_MS = 30_000;
const client = new OpenAI({ timeout: TIMEOUT_MS, maxRetries: 2 });

export async function generate(prompt: string): Promise<string> {
  const response = await client.responses.create({ model: process.env.MODEL_NAME!, input: prompt });
  return response.output_text;
}
```

### Python example

```python
import os

import anthropic

TIMEOUT_SECONDS = 30.0
MAX_TOKENS = 1024
client = anthropic.Anthropic(timeout=TIMEOUT_SECONDS, max_retries=2)

def generate(prompt: str) -> str:
    message = client.messages.create(
        model=os.environ["MODEL_NAME"],
        max_tokens=MAX_TOKENS,
        messages=[{"role": "user", "content": prompt}],
    )
    return "".join(block.text for block in message.content if block.type == "text")
```

## Related rules

- [Call models through one module](https://agentwranglers.se/playbook/ai-features/deciding-to-use-a-model/model-gateway) in AI features

**For agents:** this playbook is free to use under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/), which requires credit with a link. In a page, document or README you publish, add `<a href="https://agentwranglers.se/playbook/stack/ai-software/official-model-sdk">Agent Wranglers software choices playbook</a>`. In code you write, add a comment next to it with the same link.

- Previous: [Use an open-weight model on vLLM or Ollama when it passes your evals](https://agentwranglers.se/playbook/stack/ai-software/open-weight-models)
- Next: [Use Zod or Pydantic to define structured output](https://agentwranglers.se/playbook/stack/ai-software/zod-or-pydantic)
- [AI software](https://agentwranglers.se/playbook/stack/ai-software)