ask2api

ask2api is a minimal Python CLI tool that turns natural language prompts into structured API-style JSON responses using LLM.

It allows you to define a JSON Schema and force the model to answer strictly in that format.

Why ask2api?

Because LLMs are no longer just chatbots, they are also programmable API engines.

ask2api lets you use them that way. 🚀

Key features:

• Minimal dependencies
• CLI first
• Prompt → API behavior
• No markdown, no explanations, only valid JSON
• Vision modality support
• Designed for automation pipelines and AI-driven backend workflows

Installation

pip install ask2api

Set your API key:

export ASK2API_API_KEY="your_api_key"
# Or you can pass OpenAI key
# export OPENAI_API_KEY="your_api_key"

Provider Support

ask2api supports both OpenAI (and OpenAI-compatible) and Anthropic (Claude) models.

Using OpenAI (default)

By default, ask2api uses OpenAI. No additional configuration is needed:

export OPENAI_API_KEY="sk-..."
ask2api -p "What is 2+2?" -e '{"result": 1}'

Using Anthropic Claude

To use Anthropic's Claude models, set the provider explicitly:

export ASK2API_PROVIDER="anthropic"
export ANTHROPIC_API_KEY="sk-ant-..."
ask2api -p "What is the capital of France?" -e '{"country": "string", "city": "string"}'

You can also customize the model:

export ASK2API_PROVIDER="anthropic"
export ASK2API_API_KEY="sk-ant-..."
export ASK2API_MODEL="claude-opus-4-5"
ask2api -p "Analyze carbon" -e '{"symbol": "string", "atomic_number": 1}'

Anthropic's Claude models support vision as well:

export ASK2API_PROVIDER="anthropic"
export ANTHROPIC_API_KEY="sk-ant-..."
ask2api -p "What's in this image?" -e '{"description": "string"}' -i photo.jpg

Usage

Text-only prompts

Instead of asking:

"Where is the capital of France?"

and receiving free-form text, you can do this:

ask2api -p "Where is the capital of France?" -sf schema.json

Or pass an example directly without a schema file:

ask2api -p "Where is the capital of France?" -e '{"country": "string", "city": "string"}'

And get a structured API response:

{
  "country": "France",
  "city": "Paris"
}

ask2api -p "Analyze carbon element" -e '{
  "symbol": "element symbol",
  "atomic_number": 1234,
  "atomic_weight": 12.34,
  "is_metal": true,
  "isotopes": ["name of the isotope"],
  "properties": {
    "melting_point": 1234.5,
    "boiling_point": 2345.6,
    "magnetic": true
  }
}'

{
  "symbol": "C",
  "atomic_number": 6,
  "atomic_weight": 12.011,
  "is_metal": false,
  "isotopes": [
    "C-12",
    "C-13",
    "C-14"
  ],
  "properties": {
    "melting_point": 3550,
    "boiling_point": 4827,
    "magnetic": false
  }
}

Vision modality

You can also analyze images and get structured JSON responses:

ask2api -p "Where is this place?" -sf schema.json -i https://upload.wikimedia.org/wikipedia/commons/6/64/Lesdeuxmagots.jpg

curl -sSL -o place.jpg "https://upload.wikimedia.org/wikipedia/commons/6/64/Lesdeuxmagots.jpg"
ask2api -p "Where is this place?" -sf schema.json -i ./place.jpg

wget -O place.jpg "https://upload.wikimedia.org/wikipedia/commons/6/64/Lesdeuxmagots.jpg"

How it works

1. You define the desired output structure using a JSON Schema.
2. The schema is passed to the model:
   • For OpenAI: using the json_schema structured output format
   • For Anthropic: using tool calling with forced tool use
3. The system prompt enforces strict JSON-only responses.
4. For vision tasks, images are automatically encoded (base64 for local files) or passed as URLs.
5. The CLI prints the API-ready JSON output.

The model is treated as a deterministic API function, regardless of provider.

Example schema

Create a file named schema.json:

{
  "type": "object",
  "properties": {
    "country": { "type": "string" },
    "city": { "type": "string" }
  },
  "required": ["country", "city"]
}

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

MIT