Isolated Agents SDK¶

Production-ready SDK for running AI agents in isolated containers

What is Isolated Agents SDK?¶

The Isolated Agents SDK is a production-ready framework for running AI agents in secure, isolated containers. It provides:

🔒 Security-first design with container isolation
🎯 Simple API with Pythonic decorators
🔌 Pluggable architecture with adapter pattern
🌐 Framework agnostic - works with any AI framework
🚀 Production-ready with comprehensive monitoring
📦 Cross-platform support (Linux, macOS, Windows)

Quick Example¶

from isolated_agents_sdk import run_agent, Policy, NetworkPolicy

def my_agent():
    """Your agent code runs in an isolated container."""
    from langchain_openai import ChatOpenAI
    from pathlib import Path

    llm = ChatOpenAI(model="gpt-4")
    result = llm.invoke("Explain quantum computing")

    # Save output
    Path("/output/response.txt").write_text(result.content)

# Run agent in isolated container
result = run_agent(
    agent=my_agent,
    working_dir="./workspace",
    host_output_path="./output",
    policy=Policy(
        network=NetworkPolicy(disabled=False),
        allowed_env_vars=["OPENAI_API_KEY"],
        pip_packages=["langchain-openai"],
    )
)

print(result.artifacts["response.txt"])

Key Features¶

🔒 Security & Isolation¶

Container isolation - Each agent runs in its own container
Network policies - Fine-grained network access control
Resource limits - CPU, memory, and timeout constraints
Read-only filesystem - Immutable container environment
Audit logging - Complete audit trail of all operations

🎯 Developer Experience¶

Simple API - Intuitive Python interface
Decorator support - Clean, Pythonic code
Type safety - Full type hints and validation
Error handling - Comprehensive error messages
Hot reload - Fast development iteration

🔌 Extensibility¶

Adapter pattern - Pluggable container runtimes (Podman, Docker, K8s)
Storage adapters - Local, S3, Azure, GCS
Logger adapters - File, database, cloud logging
Custom adapters - Easy to extend

🌐 Framework Support¶

Works with any AI framework:

LangChain - Full support with examples
CrewAI - Multi-agent orchestration
AutoGPT - Autonomous agents
LlamaIndex - RAG and document processing
Haystack - NLP pipelines
Semantic Kernel - Microsoft's AI framework
Custom frameworks - Bring your own

🚀 Production Features¶

Composability - Chain agents together
Multimodal outputs - Text, images, audio, video
Validation - Output validation and testing
Telemetry - Real-time monitoring
Auto-scaling - Horizontal scaling support
Cross-platform - Linux, macOS, Windows

Use Cases¶

🕷️ Web Scraping¶

Run web scraping agents with network isolation and rate limiting.

@isolated_agent
@network(allowed_endpoints=["example.com:443"])
def scraper():
    import requests
    from bs4 import BeautifulSoup
    # Scraping logic...

📊 Data Analysis¶

Process sensitive data in isolated environments.

@isolated_agent
@resources(memory_mb=4096, cpu_cores=4.0)
def analyzer():
    import pandas as pd
    # Analysis logic...

🤖 Multi-Agent Systems¶

Orchestrate multiple agents with different permissions.

@chain(agents=[researcher, writer, editor])
def content_pipeline(topic: str):
    """Complete content creation pipeline."""
    pass

🔐 Secure Code Execution¶

Execute untrusted code safely in containers.

@isolated_agent
@policy(network=NetworkPolicy(disabled=True))
def code_executor(code: str):
    exec(code)  # Safe in isolated container

Architecture¶

graph TB
    A[Your Application] --> B[Isolated Agents SDK]
    B --> C[Container Runtime Adapter]
    C --> D[Podman]
    C --> E[Docker]
    C --> F[Kubernetes]
    B --> G[Storage Adapter]
    G --> H[Local]
    G --> I[S3]
    G --> J[Azure]
    B --> K[Audit Logger]
    K --> L[File]
    K --> M[Database]
    K --> N[CloudWatch]

Why Isolated Agents SDK?¶

Security¶

Traditional AI agents run in your application's process, with full access to: - ❌ Your filesystem - ❌ Your network - ❌ Your environment variables - ❌ Your credentials

Isolated Agents SDK runs each agent in a container with: - ✅ Isolated filesystem - ✅ Controlled network access - ✅ Limited resources - ✅ No access to host credentials

Reliability¶

Resource limits prevent agents from consuming all resources
Timeouts prevent infinite loops
Crash isolation - one agent crash doesn't affect others
Audit logging for debugging and compliance

Flexibility¶

Framework agnostic - use any AI framework
Language agnostic - Python, Node.js, Go, Rust, Java
Pluggable - swap container runtimes, storage, logging
Composable - chain agents together

Getting Started¶

Installation¶

pip install isolated-agents-sdk

Prerequisites¶

Python 3.11+
Podman or Docker

Quick Start¶

Install the SDK
```
pip install isolated-agents-sdk
```

Write your first agent

from isolated_agents_sdk import run_agent, Policy

def hello_agent():
    print("Hello from isolated container!")

result = run_agent(agent=hello_agent, policy=Policy())

Run it
```
python my_agent.py
```

See the Getting Started Guide for more details.

Documentation¶

Getting Started - Installation and first steps
Adapter Architecture - Understanding the SDK
Architecture - System design
Examples - 81+ working examples
Extending Adapters - Implementation guides

Examples¶

LangChain Agent¶

from isolated_agents_sdk import isolated_agent, network

@isolated_agent
@network(enabled=True)
def langchain_agent():
    from langchain_openai import ChatOpenAI
    llm = ChatOpenAI(model="gpt-4")
    result = llm.invoke("Explain AI safety")
    return result.content

Multi-Agent Pipeline¶

from isolated_agents_sdk import chain

@chain(agents=[researcher, writer, editor])
def content_pipeline(topic: str):
    """Research → Write → Edit"""
    pass

result = content_pipeline("AI Safety")

Data Analysis¶

from isolated_agents_sdk import isolated_agent, resources

@isolated_agent
@resources(memory_mb=4096, cpu_cores=4.0)
def analyze_data():
    import pandas as pd
    df = pd.read_csv("/workspace/data.csv")
    return df.describe()

See all examples for more.

Community¶

GitHub - Source code and issues
Discord - Community chat
Twitter - Updates and news
Forum - Discussions

Contributing¶

We welcome contributions! Please see our GitHub repository for more details.

License¶

Isolated Agents SDK is licensed under the MIT License.

Support¶

Documentation: docs.isolated-agents.dev
Issues: GitHub Issues
Discord: Join our community
Email: support@isolated-agents.dev

Made with ❤️ by the Isolated Agents team

Built with IBM BOB IBM BOB Logo

GitHub • Documentation • Discord • Twitter