fleetfluid 0.1.6

AI Agent Functions for ETL/Data Processing

FleetFluid

AI Agent Functions for Data Processing with Dual-Mode Support

FleetFluid is a Python library that simplifies data transformation by letting you use AI-powered functions without writing them from scratch. Instead of building functions, you invoke ready-made, agent-based functions that handle tasks like text cleaning, information extraction, translation, labeling, anonymization, and more—just by specifying what you need in natural language.

🚀 Dual-Mode Architecture

FleetFluid supports two execution modes:

• 🔓 Open Source Mode: Uses PydanticAI agents directly on your machine
• ☁️ Cloud Mode: Uses cloud computation for enterprise-grade performance and reliability

The same Python interface works in both modes - just change the initialization to switch between them!

Installation

pip install fleetfluid

Quick Start

Open Source Mode

from fleetfluid.core import FleetFluid

# Initialize with PydanticAI (open source mode)
ff = FleetFluid(
    model="openai:gpt-4",
    temperature=0.7,
    max_tokens=1000
)

# AI Transformation
result = ff.ai("Rewrite this in a more technical tone", "The data processing pipeline needs optimization.")
print(result)

# Multi-label Classification
result = ff.label("Database query is slow", ["Performance Issue", "Feature Request", "Bug Report"], multiple=True)
print(f"Labels: {result.labels}")
print(f"Confidence: {result.confidence_scores}")

Cloud Mode

from fleetfluid.core import FleetFluid

# Initialize with API key (cloud mode)
# A key with free tier could be obtained on https://fleetfluid.io
ff = FleetFluid(
    api_key="your_fleetfluid_api_key",
    api_endpoint="https://api.fleetfluid.io"
)

# Same interface, different backend!
result = ff.label("Database query is slow", ["Performance Issue", "Feature Request", "Bug Report"])
print(f"Label: {result.label}")

Environment-Based Configuration

# Set environment variables
export FLEETFLUID-API-KEY="your_api_key"
export FLEETFLUID_API_ENDPOINT="https://api.fleetfluid.io"

# Automatically detects cloud mode
ff = FleetFluid()  # No parameters needed!

🔄 Mode Switching

You can switch between modes at runtime:

# Start in open source mode
ff = FleetFluid(model="openai:gpt-4")

# Switch to cloud mode
ff.switch_to_cloud_mode("new_api_key", "https://api.fleetfluid.io")

# Switch back to open source mode
ff.switch_to_open_source_mode("anthropic:claude-3-sonnet", temperature=0.3)

📋 Complete Function Reference

Initialization

Open Source Mode

# Basic initialization
ff = FleetFluid()

# Custom model and parameters
ff = FleetFluid(
    model="anthropic:claude-3-sonnet",
    temperature=0.7,
    max_tokens=1000,
    top_p=0.9
)

Cloud Mode

# Explicit configuration
ff = FleetFluid(
    api_key="your_api_key",
    api_endpoint="https://api.fleetfluid.io"
)

# Environment-based
ff = FleetFluid()  # Uses FLEETFLUID-API-KEY env var

Core Methods

All methods work identically in both modes:

ff.label(text, labels, multiple=False)

Label text using AI with structured output.

# Single label
result = ff.label("Hello world", ["greeting", "statement", "question"])
print(f"Label: {result.label}")
print(f"Confidence: {result.confidence}")

# Multiple labels
result = ff.label("Hello world", ["greeting", "statement", "question"], multiple=True)
print(f"Labels: {result.labels}")
print(f"Confidence Scores: {result.confidence_scores}")

ff.ai(prompt, data)

Apply AI transformation to data.

result = ff.ai("Make this more formal", "hey there, what's up?")
print(result)  # "Hello, how are you doing?"

ff.extract(extraction_type, text)

Extract specific information from text.

skills = ff.extract("skills", "Python developer with ML experience")
print(skills)  # ["Python", "machine learning"]

ff.anonymize(text)

Anonymize personal information.

anonymized = ff.anonymize("My name is John, email: [e-mail]")
print(anonymized)  # "My name is [NAME], email: [EMAIL]"

ff.describe(features, style="natural")

Generate descriptions from features.

description = ff.describe(
    {"color": "blue", "size": "large"}, 
    style="marketing"
)
print(description)  # "A stunning large blue item..."

Async Versions

All methods have async counterparts:

# Async versions for use in async contexts
result = await ff.label_async("Hello world", ["greeting", "statement"])
result = await ff.ai_async("Make formal", "hey there")
result = await ff.extract_async("skills", "Python developer")

🔧 Configuration

Environment Variables

Variable	Description	Default
FLEETFLUID-API-KEY	API key for cloud mode	None
FLEETFLUID_API_ENDPOINT	API endpoint for cloud mode	https://api.fleetfluid.io

Open Source Mode Parameters

All PydanticAI parameters are supported:

• model: Model identifier
• temperature: Creativity level (0.0-1.0)
• max_tokens: Maximum response length
• top_p: Nucleus sampling parameter
• frequency_penalty: Frequency penalty
• presence_penalty: Presence penalty

Cloud Mode Parameters

• api_key: Your fleetfluid cloud API key
• api_endpoint: Custom API endpoint (optional)

🔒 Security

• Open Source Mode: API keys stay on your machine
• Cloud Mode: Uses Bearer token authentication
• HTTPS: All API communications are encrypted
• No Data Logging: Your data is not stored or logged
• OpenAI Integration: In Cloud mode, OpenAI GPT-4 model is used via OpenAI API Platform and FleetFluid inherits all the terms and conditions and privacy policy of OpenAI at https://openai.com/en-GB/policies/row-privacy-policy/

📦 Requirements

• Python 3.8+
• PydanticAI (for open source mode)
• httpx (for cloud mode)
• API key for your chosen model provider (open source mode) or FleetFluid (cloud mode)

🚀 Quick Reference

Initialization Cheat Sheet

Mode	Code	Use Case
Open Source	FleetFluid(model="gpt-4")	Development, open source
Open Source	FleetFluid(model="claude-3", temperature=0.7)	Custom AI parameters
Cloud	FleetFluid(api_key="key")	Production, enterprise
Cloud	FleetFluid(api_key="key", api_endpoint="https://api.fleetfluid.io")	Custom API endpoint
Auto	FleetFluid()	Uses environment variables

Function Quick Reference

Function	Description	Example
ff.label(text, labels)	Single label classification	ff.label("Hello", ["greeting", "statement"])
ff.label(text, labels, multiple=True)	Multiple label classification	ff.label("Hello", ["greeting", "statement"], multiple=True)
ff.ai(prompt, data)	AI transformation	ff.ai("Make formal", "hey there")
ff.extract(type, text)	Information extraction	ff.extract("skills", "Python developer")
ff.anonymize(text)	Text anonymization	ff.anonymize("My name is John")
ff.describe(features, style)	Feature description	ff.describe({"color": "blue"}, "marketing")

Environment Variables

# For cloud mode
export FLEETFLUID-API-KEY="your_key"
export FLEETFLUID_API_ENDPOINT="https://api.fleetfluid.io"

# For open source mode (standard AI provider keys)
export OPENAI_API_KEY="your_openai_key"
export ANTHROPIC_API_KEY="your_anthropic_key"

Mode Detection Priority

1. Constructor Parameters (highest)
2. Environment Variables
3. Default Values (lowest)

Pro Tip: api_key parameter always wins over model parameter!

📄 License

MIT License. See LICENSE file for details.