📖 Table of Contents

• What is Future AGI?
• Installation
• Authentication
• 30-Second Examples
• Quick Start
  • Dataset Management
  • Prompt Workbench
  • Knowledge Base (RAG)
• How It Works
• Core Use Cases
• Real-World Use Cases
• Why Choose Future AGI?
• Supported Integrations
• Documentation
• Language Support
• Support & Community
• Contributing
• Testimonials
• Roadmap
• Troubleshooting & FAQ

🚀 What is Future AGI?

Your agent passed every eval. Then it hallucinated a refund policy that doesn't exist. Future AGI gives you the tools to catch that — datasets, prompt versioning, knowledge bases, evaluations, and guardrails. One SDK, one feedback loop.

# Get started in 30 seconds
pip install futureagi
export FI_API_KEY="your_key"
export FI_SECRET_KEY="your_secret"

👉 Get Free API Keys • View Live Demo • Read Quick Start Guide

✨ Key Features

• 🎯 Evaluations — 50+ metrics, LLM-as-judge, and custom rubrics powered by the Critique AI agent
• ⚡ Guardrails — Real-time safety checks with sub-100ms latency
• 📊 Datasets — Programmatically create, version, and manage training and test datasets
• 🎨 Prompt Workbench — Version control, A/B testing, and deployment labels for prompts
• 📚 Knowledge Base — Document management and retrieval for RAG applications
• 📈 Analytics — Model performance, token costs, and behavior insights
• 🤖 Simulate — Test your AI system against realistic scenarios before users hit it
• 🔍 Observability — OpenTelemetry-native tracing across 50+ frameworks

📦 Installation

Python

pip install futureagi

TypeScript/JavaScript

npm install @future-agi/sdk
# or
pnpm add @future-agi/sdk

Requirements: Python >= 3.10 | Node.js >= 14

🔑 Authentication

Get your API credentials from the Future AGI Dashboard:

export FI_API_KEY="your_api_key"
export FI_SECRET_KEY="your_secret_key"

Or set them programmatically:

import os
os.environ["FI_API_KEY"] = "your_api_key"
os.environ["FI_SECRET_KEY"] = "your_secret_key"

🎯 Quick Start

📊 Dataset Management

Create and manage datasets with built-in evaluations:

from fi.datasets import Dataset
from fi.datasets.types import (
    Cell, Column, DatasetConfig, DataTypeChoices,
    ModelTypes, Row, SourceChoices
)

# Create a new dataset
config = DatasetConfig(name="qa_dataset", model_type=ModelTypes.GENERATIVE_LLM)
dataset = Dataset(dataset_config=config)
dataset = dataset.create()

# Define columns
columns = [
    Column(name="user_query", data_type=DataTypeChoices.TEXT, source=SourceChoices.OTHERS),
    Column(name="ai_response", data_type=DataTypeChoices.TEXT, source=SourceChoices.OTHERS),
    Column(name="quality_score", data_type=DataTypeChoices.INTEGER, source=SourceChoices.OTHERS),
]

# Add data
rows = [
    Row(order=1, cells=[
            Cell(column_name="user_query", value="What is machine learning?"),
        Cell(column_name="ai_response", value="Machine learning is a subset of AI..."),
        Cell(column_name="quality_score", value=9),
    ]),
    Row(order=2, cells=[
            Cell(column_name="user_query", value="Explain quantum computing"),
        Cell(column_name="ai_response", value="Quantum computing uses quantum bits..."),
        Cell(column_name="quality_score", value=8),
    ]),
]

# Push data and run evaluations
    dataset = dataset.add_columns(columns=columns)
    dataset = dataset.add_rows(rows=rows)

# Add automated evaluation
dataset.add_evaluation(
    name="factual_accuracy",
    eval_template="is_factually_consistent",
    model="gpt-4o-mini",
    required_keys_to_column_names={
        "input": "user_query",
        "output": "ai_response",
        "context": "user_query",
    },
    run=True
)

print("✓ Dataset created with automated evaluations")

🎨 Prompt Workbench

Version control and A/B test your prompts:

from fi.prompt import Prompt, PromptTemplate, ModelConfig

# Create a versioned prompt template
template = PromptTemplate(
    name="customer_support",
    messages=[
        {"role": "system", "content": "You are a helpful customer support agent."},
        {"role": "user", "content": "Help {{customer_name}} with {{issue_type}}."},
    ],
    variable_names={"customer_name": ["Alice"], "issue_type": ["billing"]},
    model_configuration=ModelConfig(model_name="gpt-4o-mini", temperature=0.7)
)

# Create and version the template
client = Prompt(template)
client.create()  # Create v1
client.commit_current_version("Initial version", set_default=True)

# Assign deployment labels
client.assign_label("Production", version="v1")

# Compile with variables
compiled = client.compile(customer_name="Bob", issue_type="refund")
print(compiled)
# Output: [
#   {"role": "system", "content": "You are a helpful customer support agent."},
#   {"role": "user", "content": "Help Bob with refund."}
# ]

A/B Testing Example:

import random
from openai import OpenAI
from fi.prompt import Prompt

# Fetch different variants (returns Prompt instances)
variant_a = Prompt.get_template_by_name("customer_support", label="variant-a")
variant_b = Prompt.get_template_by_name("customer_support", label="variant-b")

# Randomly select and use
selected = random.choice([variant_a, variant_b])
compiled = selected.compile(customer_name="Alice", issue_type="refund")

# Send to your LLM provider
openai = OpenAI(api_key="your_openai_key")
response = openai.chat.completions.create(model="gpt-4o", messages=compiled)
print(f"Using variant: {selected.template.name}")
print(f"Response: {response.choices[0].message.content}")

📚 Knowledge Base (RAG)

Manage documents for retrieval-augmented generation:

from fi.kb import KnowledgeBase

# Initialize client
kb_client = KnowledgeBase(
    fi_api_key="your_api_key",
    fi_secret_key="your_secret_key"
)

# Create a knowledge base with documents
kb = kb_client.create_kb(
    name="product_docs",
    file_paths=["manual.pdf", "faq.txt", "guide.docx"]
)

print(f"✓ Knowledge base created: {kb.kb.name}")
print(f"  Files uploaded: {len(kb.kb.files)}")

# Update with more files
updated_kb = kb_client.update_kb(
    kb_name=kb.kb.name,
    file_paths=["updates.pdf"]
)

# Delete specific files
kb_client.delete_files_from_kb(file_names=["updates.pdf"])

# Clean up
kb_client.delete_kb(kb_ids=[kb.kb.id])

🎯 Core Use Cases

🔥 Why Choose Future AGI?

🔌 Supported Integrations

Future AGI works seamlessly with your existing AI stack:

LLM Providers
OpenAI • Anthropic • Google Gemini • Azure OpenAI • AWS Bedrock • Cohere • Mistral • Ollama • vLLM

Frameworks
LangChain • LlamaIndex • CrewAI • AutoGen • Haystack • Semantic Kernel

Vector Databases
Pinecone • Weaviate • Qdrant • Milvus • Chroma • FAISS

Observability
OpenTelemetry • Custom Logging • Trace Context Propagation

📚 Documentation

• Complete Documentation
• Dataset SDK Guide
• Prompt Workbench Guide
• Knowledge Base Guide
• API Reference

🤝 Language Support

🆘 Support & Community

• 📧 Email: [email protected]
• 💼 LinkedIn: Future AGI Company
• 🐦 X (Twitter): @FutureAGI_
• 📰 Substack: Future AGI Blog

🤝 Contributing

We welcome contributions! Here's how to get involved:

• 🐛 Report bugs: Open an issue
• 💡 Request features: Start a discussion
• 🔧 Submit PRs: Fork, create a feature branch, and submit a pull request
• 📖 Improve docs: Help us make our documentation better

See CONTRIBUTING.md for detailed guidelines.

🌟 Testimonials

"Future AGI cut our evaluation time from days to minutes. The automated critiques are spot-on!"
— AI Engineering Team, Fortune 500 Company

"The prompt versioning alone saved us countless headaches. A/B testing is now trivial."
— ML Lead, Healthcare Startup

"Sub-100ms guardrails in production. Game changer for our customer-facing AI."
— CTO, E-commerce Platform

📊 Roadmap

• Datasets with automated evaluations
• Prompt workbench with versioning
• Knowledge base for RAG
• Real-time guardrails (sub-100ms)
• Multi-language SDK (Python + TypeScript)
• Bulk Annotations for Human in the Loop
• On-premise deployment toolkit

❓ Troubleshooting & FAQ

pip install futureagi --upgrade

echo $FI_API_KEY
echo $FI_SECRET_KEY

client.assign_label("Development", version="v1")
client.assign_label("Staging", version="v2")
client.assign_label("Production", version="v3")

Need more help? Check our complete FAQ or join our community.

📄 License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.