ai-extract 0.0.5

Extract structured data from any AI. Fast. Simple. Reliable.

ai-extract

Extract structured data from any AI response. Fast. Simple. Reliable.

ai-extract is a fast, lightweight Python library for extracting JSON from AI outputs. It handles markdown fences, surrounding text, and multiple JSON blocks.

Features

• Fast: Uses orjson for 10x faster JSON parsing
• Reliable: Multiple extraction strategies with automatic fallback
• Simple API: One function for most use cases
• CLI Tool: Quick extraction from command line
• Zero Config: Works out of the box

Installation

pip install ai-extract

Quick Start

from ai_extract import extract_json

# Extract from messy AI output
text = """
Here's the JSON you requested:
```json
{"name": "John", "age": 30}

Hope this helps! """

data = extract_json(text) print(data) # {'name': 'John', 'age': 30}

## Usage Examples

### Basic Extraction

```python
from ai_extract import extract_json

# Pure JSON
data = extract_json('{"key": "value"}')

# JSON in text
data = extract_json('The result is: {"success": true}')

# JSON in markdown fence
data = extract_json('```json\n{"data": [1,2,3]}\n```')

Multiple JSON Blocks

# Get first JSON (default)
data = extract_json('{"a": 1} and {"b": 2}')
# Returns: {'a': 1}

# Get largest JSON
data = extract_json('{"a": 1} and {"b": 2, "c": 3}', strategy="largest")
# Returns: {'b': 2, 'c': 3}

# Get all JSON blocks
data = extract_json('{"a": 1} and {"b": 2}', strategy="all")
# Returns: [{'a': 1}, {'b': 2}]

Error Handling

from ai_extract import extract_json, ExtractError

# Raise exception (default)
try:
    data = extract_json("no json here")
except ExtractError as e:
    print(f"Error: {e.message}")
    print(f"Type: {e.error_type}")

# Return None instead
data = extract_json("no json here", raise_on_error=False)
# Returns: None

Get Extraction Metadata

from ai_extract import extract_json_with_metadata

result = extract_json_with_metadata('```json\n{"key": "value"}\n```')

print(result.success)        # True
print(result.data)           # {'key': 'value'}
print(result.method)         # ExtractionMethod.MARKDOWN_FENCE

CLI Usage

# From argument
ai-extract '{"key": "value"}'

# From file
ai-extract -f response.txt

# From stdin
echo '{"key": "value"}' | ai-extract

# Pretty print
ai-extract -f response.txt --pretty

# Get all JSON blocks
ai-extract -f response.txt --all

# Show metadata
ai-extract -f response.txt --verbose

API Reference

extract_json(text, *, strategy="first", raise_on_error=True)

Extract JSON from text.

Parameters:

• text (str): Text containing JSON
• strategy (str): How to handle multiple JSON blocks
  • "first": Return first valid JSON (default)
  • "largest": Return largest JSON structure
  • "all": Return list of all JSON blocks
• raise_on_error (bool): Raise ExtractError on failure (default: True)

Returns: Parsed JSON data, or list if strategy="all", or None if raise_on_error=False

extract_json_with_metadata(text, *, strategy="first")

Extract JSON with detailed metadata.

Returns: ExtractResult with fields:

• success (bool): Whether extraction succeeded
• data (Any): Parsed JSON data
• raw_json (str): Raw JSON string before parsing
• method (ExtractionMethod): How JSON was found
• candidates_found (int): Number of JSON candidates found
• error (ExtractError): Error details if failed

Extraction Methods

The library tries multiple strategies in order:

1. Direct Parse - Try parsing entire input as JSON
2. Markdown Fence - Extract from ```json blocks
3. Brace Matching - Find balanced {...} or [...]
4. Heuristic - Pattern matching after "Here's the JSON:" etc.

License

MIT License - see LICENSE for details.

Author

Rahul Vishwakarma (@rvv-karma)