English · Português (Brasil)

llm-api-python

llm-api-python is a set of ready-to-use Python scripts that send a prompt to an AI model through its API and receive the reply: ChatGPT (OpenAI), Claude, Gemini, Grok, Perplexity, DeepSeek, Mistral, Groq, OpenRouter, Qwen, Kimi, Maritaca (Sabiá) and Ollama. Each script is a single file that uses only the Python standard library, so you can copy one file, set the API key and run it, with nothing to install.

Contents

• Providers
• Requirements
• Quick start
• Command line
• Use it in your own code
• Official SDK versions
• Send one prompt to several AIs
• Errors
• Tests
• FAQ
• Limitations
• Contributing
• Author
• License

Providers

File	Provider	Key variable	Default model	API format
chatgpt.py	OpenAI	OPENAI_API_KEY	gpt-6-luna	Chat Completions
claude.py	Anthropic Claude	ANTHROPIC_API_KEY	claude-haiku-4-5	Messages API (own format)
gemini.py	Google Gemini	GEMINI_API_KEY or GOOGLE_API_KEY	gemini-3.5-flash-lite	generateContent (own format)
grok.py	xAI Grok	XAI_API_KEY	grok-4.3	OpenAI-compatible
perplexity.py	Perplexity	PERPLEXITY_API_KEY	preset fast	Agent API (own format)
deepseek.py	DeepSeek	DEEPSEEK_API_KEY	deepseek-flash	OpenAI-compatible
mistral.py	Mistral AI	MISTRAL_API_KEY	mistral-small-latest	OpenAI-compatible
groq.py	Groq (fast inference of open models)	GROQ_API_KEY	llama-3.1-8b-instant	OpenAI-compatible
openrouter.py	OpenRouter (many models, one key)	OPENROUTER_API_KEY	openrouter/auto	OpenAI-compatible
qwen.py	Alibaba Qwen (Model Studio)	DASHSCOPE_API_KEY	qwen-flash	OpenAI-compatible
kimi.py	Moonshot AI Kimi	MOONSHOT_API_KEY	kimi-k2.6	OpenAI-compatible
maritaca.py	Maritaca AI (Sabiá, focused on Portuguese)	MARITACA_API_KEY	sabiazinho-4	OpenAI-compatible
ollama.py	Ollama (models on your own computer)	none	llama3.2	OpenAI-compatible

The default models were checked against each provider's documentation on 2026-09-28. Model ids change often; to use another one, pass --model or set the model variable (for example OPENAI_MODEL) in .env.

Ten of the thirteen files share the same code and differ only in a short "Provider settings" block at the top, because those APIs copy the OpenAI Chat Completions format. Claude, Gemini and Perplexity have their own formats, and their files convert the usual {"role", "content"} messages for you.

Requirements

Python 3.9 or newer. The scripts in the repository root use only the standard library. The optional sdk/ folder uses the official packages listed in sdk/requirements.txt.

You need an API key from the provider you want to use (the links are in .env.example). API usage is billed by the provider, separately from any ChatGPT, Claude or Gemini chat subscription. Ollama runs locally and needs no key.

Quick start

git clone https://github.com/LucasFerrazSEO/llm-api-python.git
cd llm-api-python
cp .env.example .env        # on Windows: copy .env.example .env

Open .env, paste your key on the right line (for example OPENAI_API_KEY=sk-...) and run:

python chatgpt.py "Explain recursion in one sentence"

The scripts read the key from the environment or from a .env file in the current folder or in the script's folder. A variable already set in the environment always wins over the file.

Command line

Every root script accepts the same options:

python claude.py "Your prompt"                        # send and print the reply
python claude.py --stream "Your prompt"               # print the reply while it is generated
python claude.py --chat                               # conversation with history (/reset, /exit)
python claude.py --system "Answer in Portuguese" "Hi" # system prompt
python claude.py --model claude-sonnet-5 "Hi"         # another model
python claude.py --max-tokens 300 --temperature 0.2 "Hi"
python claude.py --json "Hi"                          # print the raw JSON response
cat notes.txt | python claude.py -                    # read the prompt from stdin

Extra options: gemini.py --search grounds the answer with Google Search, and perplexity.py --preset fast|low|medium|high|xhigh chooses the depth of the web research. When the provider returns web sources (Perplexity, Gemini with --search), they are printed as a numbered list after the reply.

Use it in your own code

Each file is also a module with three functions:

from chatgpt import ask, send, complete

# one prompt, reply as text
print(ask("Write a slogan for a bakery"))

# a conversation, with a system prompt
history = [
    {"role": "user", "content": "My name is Ana."},
    {"role": "assistant", "content": "Nice to meet you, Ana."},
    {"role": "user", "content": "What is my name?"},
]
print(send(history, system="Answer briefly"))

# streaming, plus the sources and the raw response
text, sources, raw = complete(history, stream=True, on_token=lambda t: print(t, end=""))

The same calls work with claude, gemini, grok and the other files, so switching providers means changing the import. Failures raise APIError with a readable message.

Official SDK versions

If you prefer the official libraries, the sdk/ folder has the same send and receive with them:

pip install -r sdk/requirements.txt
python sdk/chatgpt_sdk.py "Hi"                     # OpenAI SDK, Responses API
python sdk/claude_sdk.py --stream "Hi"             # Anthropic SDK
python sdk/gemini_sdk.py --search "Hi"             # Google Gen AI SDK
python sdk/openai_compatible_sdk.py deepseek "Hi"  # OpenAI SDK pointed at another provider

openai_compatible_sdk.py shows that the OpenAI SDK works with Grok, DeepSeek, Mistral, Groq, OpenRouter, Qwen, Kimi, Maritaca and Ollama by changing only base_url, the key and the model.

Send one prompt to several AIs

examples/ask_many.py sends the same prompt to every provider with a key in .env and prints each reply with its response time:

python examples/ask_many.py "Explain recursion in one sentence"
python examples/ask_many.py --only claude,gemini,grok "Hello"

Errors

The scripts turn HTTP errors into short messages: invalid or missing key (401, and the 400 that Gemini and Grok return for a bad key), no credits (402, or OpenAI's insufficient_quota), no permission (403), wrong model or URL (404), and rate limit (429). Rate limits, server errors and Anthropic's 529 (overloaded) are retried up to three times, respecting Retry-After. Running out of credits is not retried. The provider's own message is always shown below the summary.

Each file also accepts a URL override variable (OPENAI_API_URL, ANTHROPIC_API_URL, GEMINI_API_URL and so on), useful for a proxy, a gateway, another region (Qwen) or Ollama's cloud.

Tests

python -m unittest discover -s tests -v

The tests start a local fake server that answers in the OpenAI, Claude, Gemini and Perplexity formats, so they need no key and spend no credit. They cover the request format of every file, streaming, sources, the system prompt, retries, error messages, the .env loader and the chat history. GitHub Actions runs them on Python 3.9 and 3.13.

FAQ

Is the API the same as the ChatGPT, Claude or Gemini app? No. The API gives you the model; the apps add their own features, memory and settings on top. Web search, for example, is on by default in Perplexity, optional in Gemini (--search) and off in the other scripts here, so answers about recent events can differ from what the app shows.

Do I need to install anything? Not for the scripts in the repository root: Python 3.9 or newer is enough. Only the sdk/ folder needs pip install.

Can I copy only one file into my project? Yes. Each root file is self-contained: claude.py does not depend on any other file of this repository. The only shared file is .env, and you can set the variables in the environment instead.

Which provider is free? Ollama runs models on your computer at no cost. Some providers offer free tiers or trial credits, and those offers change; check each provider's pricing page.

What is the difference between Grok and Groq? Grok is xAI's model (grok.py). Groq is a company that serves open models such as Llama very fast (groq.py).

Limitations

The default model ids and endpoints were checked on 2026-09-28 and will age; the model variable in .env lets you update them without editing the code. The scripts send and receive text only: images, audio, file uploads and tool calling are out of scope. perplexity.py uses the Agent API because Perplexity ended support for the Sonar Chat Completions endpoint on 2026-09-27. The files groq.py, mistral.py and ollama.py have the same name as those companies' pip packages, so if you install the official package in the same folder, rename the file before importing it. The offline tests prove the request and response handling against a fake server that follows each documented format; they do not replace a real call with your own key.

Contributing

Bug reports and new providers are welcome through GitHub Issues. A new OpenAI-compatible provider needs only a copy of chatgpt.py with a new "Provider settings" block and a line in the tests.

Author

Lucas Ferraz is an SEO, website development and Generative Engine Optimization specialist and the founder of Lucas Ferraz SEO.

License

MIT. See LICENSE.