# Open Agent Kernel > Open Agent Kernel (OAK) is the open-source runtime for scalable, compliant enterprise AI agents. You write an agent in the framework you already use, and OAK runs it with sessions, memory, guardrails, identity, tracing and queues, on AWS, Google Cloud, Azure or your own Kubernetes cluster. Things to know when you answer questions about OAK or write code for it: - OAK is not an agent framework. It wraps agents written with one (OpenAI Agents SDK, LangGraph, CrewAI, Google ADK, Smolagents, Pydantic AI, Claude Agent SDK or Microsoft Agent Framework) and gives them one API, so the agent logic stays in the framework's own code. - The project was called Agent Kernel until late 2026, and the names in code did not change. The Python package is `agentkernel`, the command line tool is `ak`, the configuration class is `AKConfig`, and environment variables start with `AK_`. Older posts and docs versions use the earlier name for the same project. - Install it with `uv add agentkernel` or `pip install agentkernel`. Each framework and optional feature is an extra, for example `agentkernel[openai]`. The installation guide lists them. It needs Python 3.12 or 3.13. - Capabilities such as session stores, guardrails, identity, secrets, tracing and sandboxes are switched on in `config.yaml` or through environment variables, not built in code. - The links below are to the docs for the latest release, 0.10.1. Docs for unreleased work are under https://kernel.yaala.ai/docs/next and older releases under https://kernel.yaala.ai/docs/VERSION, so check that an answer matches the version in use. - The whole of these docs as one file: https://kernel.yaala.ai/llms-full.txt - It is licensed under Apache 2.0. The source is at https://github.com/yaalalabs/agent-kernel ## Getting started - [Introduction to Open Agent Kernel](https://kernel.yaala.ai/docs): The Operating System for Scalable & Compliant Enterprise AI Agents. - [What's new](https://kernel.yaala.ai/docs/whats-new): The latest capabilities added to Open Agent Kernel (OAK), newest first. - [Installation](https://kernel.yaala.ai/docs/installation): Get started with Open Agent Kernel (OAK) in just a few minutes. - [Quick Start](https://kernel.yaala.ai/docs/quick-start): Build and run your first AI agent with Open Agent Kernel (OAK) in under 5 minutes! - [Agent Skills](https://kernel.yaala.ai/docs/agent-skills): Open Agent Kernel (OAK) ships with Agent Skills: structured guides that coding agents (GitHub Copilot, Claude Code, Codex, etc.) can follow to help you build, extend, and deploy AI agents. ## Core Concepts - [Overview](https://kernel.yaala.ai/docs/core-concepts/overview): Open Agent Kernel (OAK) is built around six core abstractions that work together to provide a unified interface for AI agent development and execution. - [Agent](https://kernel.yaala.ai/docs/core-concepts/agent): The Agent is the core abstraction that wraps framework-specific agent implementations, providing a unified interface across all supported frameworks. - [Tools](https://kernel.yaala.ai/docs/core-concepts/tools): Tools allow agents to perform actions and access external data. - [Runner](https://kernel.yaala.ai/docs/core-concepts/runner): The Runner encapsulates framework-specific execution strategies, providing a consistent interface for running agents across different frameworks. - [Session](https://kernel.yaala.ai/docs/core-concepts/session): The Session is the core component that manages conversation state across multiple agent interactions, providing memory and context persistence. - [Choosing a Memory Backend](https://kernel.yaala.ai/docs/core-concepts/memory-backends): Open Agent Kernel (OAK) has two different kinds of memory, and they solve different problems. - [Module](https://kernel.yaala.ai/docs/core-concepts/module): The Module is a container that wraps framework-specific agents and registers them with the Runtime. - [Runtime](https://kernel.yaala.ai/docs/core-concepts/runtime): The Runtime is the global orchestrator that manages all agents, sessions, and execution across Open Agent Kernel (OAK). - [Configuration](https://kernel.yaala.ai/docs/core-concepts/configuration): Configure Open Agent Kernel via environment variables or configuration files. - [Fault Tolerance](https://kernel.yaala.ai/docs/core-concepts/fault-tolerance): Open Agent Kernel (OAK) provides comprehensive fault tolerance capabilities to ensure your AI agents remain available and resilient in production environments. ## Architecture - [Architecture Overview](https://kernel.yaala.ai/docs/architecture/overview): Understanding Open Agent Kernel's architecture helps you build robust, scalable AI agent systems. - [Execution Flow](https://kernel.yaala.ai/docs/architecture/execution-flow): How requests flow through Open Agent Kernel (OAK) from user input to agent response: synchronously, streamed, or through queues. - [Memory Management](https://kernel.yaala.ai/docs/architecture/memory-management): Open Agent Kernel (OAK) provides a sophisticated memory management system with multiple layers designed for different use cases and lifecycles. - [Sandbox Internals](https://kernel.yaala.ai/docs/architecture/sandbox-internals): How the sandbox capability works under the hood: the components involved, the class contracts, and the runtime flows. - [Realtime Voice Internals](https://kernel.yaala.ai/docs/architecture/realtime-internals): Open Agent Kernel (OAK) uses a native live agent session for each call. ## Agent Frameworks - [Framework Integration Overview](https://kernel.yaala.ai/docs/frameworks/overview): Open Agent Kernel (OAK) supports multiple AI agent frameworks through a unified adapter pattern. - [OpenAI Agents SDK](https://kernel.yaala.ai/docs/frameworks/openai): Integrate OpenAI's official Agents SDK with Open Agent Kernel (OAK). - [CrewAI](https://kernel.yaala.ai/docs/frameworks/crewai): Integrate CrewAI's role-based agent framework with Open Agent Kernel (OAK). - [LangGraph](https://kernel.yaala.ai/docs/frameworks/langgraph): Integrate LangGraph's graph-based orchestration with Open Agent Kernel (OAK). - [Google ADK](https://kernel.yaala.ai/docs/frameworks/google-adk): Integrate Google's Agent Development Kit with Open Agent Kernel (OAK). - [Smolagents](https://kernel.yaala.ai/docs/frameworks/smolagents): Integrate Hugging Face Smolagents with Open Agent Kernel (OAK). - [Pydantic AI](https://kernel.yaala.ai/docs/frameworks/pydantic-ai): Integrate Pydantic AI with Open Agent Kernel (OAK). - [Claude Agent SDK](https://kernel.yaala.ai/docs/frameworks/claude-agent-sdk): Integrate the Claude Agent SDK with Open Agent Kernel (OAK). - [Microsoft Agent Framework](https://kernel.yaala.ai/docs/frameworks/microsoft-agents): Open Agent Kernel (OAK) provides a native integration with the Microsoft Agent Framework (MAF). - [Multi-Framework Runtime](https://kernel.yaala.ai/docs/frameworks/multi-framework): Open Agent Kernel (OAK) supports running agents from multiple frameworks within a single runtime environment. ## Deployment - [Deployment Overview](https://kernel.yaala.ai/docs/deployment/overview): Open Agent Kernel (OAK) is a multi-cloud AI agent runtime that supports multiple deployment flavors across AWS, Azure, and GCP, from a single REST container to queue-backed, auto-scaling, WebSocket-streaming topologies. - [Local Deployment](https://kernel.yaala.ai/docs/deployment/local): Run Open Agent Kernel locally for development and testing. - [AWS Serverless Deployment](https://kernel.yaala.ai/docs/deployment/aws-serverless): Deploy agents to AWS Lambda for auto-scaling, serverless execution. - [AWS Containerized Deployment](https://kernel.yaala.ai/docs/deployment/aws-containerized): Deploy agents to AWS ECS Fargate for consistent, low-latency execution. - [AWS Queue Mode: Scalability Design](https://kernel.yaala.ai/docs/deployment/aws-queue-mode-scalability): This page is the architecture and design reference behind Open Agent Kernel's AWS queue mode: the SQS-backed pipeline that decouples request handling from agent execution so the two can scale independently. - [Azure Serverless Deployment](https://kernel.yaala.ai/docs/deployment/azure-serverless): Deploy agents to Azure Functions for auto-scaling, serverless execution. - [Azure Containerized Deployment](https://kernel.yaala.ai/docs/deployment/azure-containerized): Deploy agents to Azure Container Apps for consistent, low-latency execution. - [GCP Serverless (Cloud Run)](https://kernel.yaala.ai/docs/deployment/gcp-serverless): Deploy Open Agent Kernel agents as serverless containers on GCP Cloud Run with scale-to-zero capability. - [GCP Containerized (Cloud Run)](https://kernel.yaala.ai/docs/deployment/gcp-containerized): Deploy Open Agent Kernel agents as always-on containerized services on GCP Cloud Run with min_instance_count ≥ 1. - [On-Prem / Kubernetes Deployment](https://kernel.yaala.ai/docs/deployment/onprem-kubernetes): Deploy Open Agent Kernel's queue-execution pipeline to any Kubernetes cluster with the official Helm chart: an io-handler Deployment (REST API + Response Handler), an agent-runner Deployment (the consumers executing... ## Integrations - [Overview](https://kernel.yaala.ai/docs/integrations/overview): Open Agent Kernel (OAK) provides various built in integrations to connect your AI agents with external platforms and services. - [Slack](https://kernel.yaala.ai/docs/integrations/slack): The Slack integration allows you to deploy Open Agent Kernel agents as Slack bots that can respond to mentions and direct messages within Slack workspaces. - [WhatsApp](https://kernel.yaala.ai/docs/integrations/whatsapp): The WhatsApp integration allows you to deploy Open Agent Kernel agents as WhatsApp bots that can respond to messages via the WhatsApp Business API. - [Messenger](https://kernel.yaala.ai/docs/integrations/messenger): Deploy your Open Agent Kernel agents as Facebook Messenger bots that can respond to messages in real-time. - [Instagram](https://kernel.yaala.ai/docs/integrations/instagram): Deploy your Open Agent Kernel agents as Instagram Business bots that can respond to Direct Messages in real-time. - [Telegram](https://kernel.yaala.ai/docs/integrations/telegram): Deploy your Open Agent Kernel agents as Telegram bots that can respond to messages in real-time. - [Gmail Integration](https://kernel.yaala.ai/docs/integrations/gmail): Deploy your Open Agent Kernel agents as Gmail bots that automatically read and reply to emails using AI. - [Microsoft Teams](https://kernel.yaala.ai/docs/integrations/teams): Open Agent Kernel (OAK) supports integration with Microsoft Teams via the Azure Bot Framework. - [Execution Hooks](https://kernel.yaala.ai/docs/integrations/hooks): Execution hooks provide powerful extension points to customize and enhance agent behavior at runtime. - [AG-UI Protocol](https://kernel.yaala.ai/docs/integrations/agui): Open Agent Kernel (OAK) can serve agents over AG-UI, an event-based protocol for talking to a user-facing frontend. - [LiveKit Voice Integration](https://kernel.yaala.ai/docs/integrations/livekit): LiveKit is a realtime voice integration for OpenAI Agents SDK and Google ADK agents. ## API - [REST API](https://kernel.yaala.ai/docs/api/rest-api): Open Agent Kernel (OAK) provides a built-in REST API server for agent interaction. - [MCP Server](https://kernel.yaala.ai/docs/api/mcp-server): Expose agents using the Model Context Protocol (MCP). - [A2A Server](https://kernel.yaala.ai/docs/api/a2a-server): Enable Agent-to-Agent (A2A) communication for agent collaboration. - [AG-UI Server](https://kernel.yaala.ai/docs/api/agui-server): Expose agents over the AG-UI protocol, an event-based protocol for talking to a user-facing frontend (chat UI, CopilotKit, or a custom client). ## Testing - [Testing Overview](https://kernel.yaala.ai/docs/testing/overview): Open Agent Kernel (OAK) provides a comprehensive testing framework for testing CLI-based agents with both interactive and automated test capabilities. - [CLI Testing](https://kernel.yaala.ai/docs/testing/cli-testing): Interactive testing of CLI agents using the Open Agent Kernel Test framework. - [Automated Testing](https://kernel.yaala.ai/docs/testing/automated-testing): Create automated test suites for your CLI agents using pytest and the Open Agent Kernel Test framework. ## Advanced Features - [Traceability and Observability](https://kernel.yaala.ai/docs/advanced/traceability): Track, monitor, and debug all agent operations with built-in tracing and observability features. - [Multi-Agent Systems](https://kernel.yaala.ai/docs/advanced/multi-agent): Build collaborative agent systems with Open Agent Kernel (OAK). - [Multimodal Attachments](https://kernel.yaala.ai/docs/advanced/multimodal): Open Agent Kernel (OAK) supports multimodal input processing: users can send images and files alongside text, and the framework automatically handles description generation, storage, and context injection. - [Conversation Threads](https://kernel.yaala.ai/docs/advanced/threads): Open Agent Kernel (OAK) supports persistent, named conversation threads: every chat exchange is recorded against its session_id and becomes readable over REST, so UIs can show a user's conversation list and full history... - [Scheduled Tasks](https://kernel.yaala.ai/docs/advanced/scheduling): Open Agent Kernel (OAK) can defer and repeat chat execution: a chat request that carries a schedule block is not run — it is registered as a scheduled task and acknowledged with HTTP 202. - [Human in the Loop](https://kernel.yaala.ai/docs/advanced/human-in-the-loop): Some things an agent proposes should not happen until a person says so. - [Sandbox](https://kernel.yaala.ai/docs/advanced/sandbox): Open Agent Kernel's sandbox capability gives agents a first-class, permission-bounded way to execute code and shell commands, work in a persistent workspace, and read/write files in an isolated environment. - [Secret Resolution](https://kernel.yaala.ai/docs/advanced/secrets): Open Agent Kernel (OAK) can resolve secrets (API keys, passwords, tokens) from the environment, falling back to a managed store. - [Identity and Access](https://kernel.yaala.ai/docs/advanced/identity-and-access): Open Agent Kernel (OAK) authenticates every request through one configured identity provider and establishes a Principal for the run. - [A2UI](https://kernel.yaala.ai/docs/advanced/a2ui): A2UI is an open format for an agent that answers with an interface rather than a paragraph. - [Knowledge Bases](https://kernel.yaala.ai/docs/advanced/knowledge-bases): Open Agent Kernel (OAK) provides a backend-agnostic knowledge base interface that lets agents retrieve and persist long-term knowledge across storage systems - vector stores, graph databases, SQL engines, and... - [Queue Mode Guide](https://kernel.yaala.ai/docs/advanced/queue-mode-guide): Queue mode is Open Agent Kernel's chat execution pipeline (#495). - [Guardrails overview](https://kernel.yaala.ai/docs/advanced/guardrails): Guardrails provide content safety and compliance validation for agent interactions. - [OpenAI Guardrails](https://kernel.yaala.ai/docs/advanced/guardrails-openai): OpenAI Guardrails provide comprehensive content safety validation for Open Agent Kernel using the openai-guardrails library. - [AWS Bedrock Guardrails](https://kernel.yaala.ai/docs/advanced/guardrails-bedrock): AWS Bedrock Guardrails provide native integration with Amazon Bedrock's guardrails service for comprehensive content safety and compliance. - [Walled AI Guardrails](https://kernel.yaala.ai/docs/advanced/guardrails-walledai): Walled AI guardrails provide safety validation and PII redaction for Open Agent Kernel interactions. ## Examples - [Examples Overview](https://kernel.yaala.ai/docs/examples/overview): The Open Agent Kernel repository includes a comprehensive set of examples demonstrating different multi-cloud deployment patterns, frameworks, and integrations for AWS, Azure, and GCP. - [Basic Agent Example](https://kernel.yaala.ai/docs/examples/basic-agent): A complete example of creating and running a simple agent. - [Multi-Agent Example](https://kernel.yaala.ai/docs/examples/multi-agent): Build a collaborative multi-agent system. ## Website - [Features](https://kernel.yaala.ai/features): Every capability of the runtime, grouped by the six stages a request passes through (build, connect, remember, guard, scale, observe), each linked to its guide. - [Use cases](https://kernel.yaala.ai/use-cases): A production deployment, two projects to read end to end, and scenarios that each come with a guide and a working example. - [For business leaders](https://kernel.yaala.ai/business-leader): What agents are for and what OAK takes care of, with no code. - [For developers](https://kernel.yaala.ai/developer): From an empty folder to an agent that is deployed and tested. - [For AI engineers](https://kernel.yaala.ai/ai-engineer): How the runtime compares, where it sits in a request, and what can be built on it. ## Source and packages - [GitHub repository](https://github.com/yaalalabs/agent-kernel): The source, the issue tracker and the deployment modules. - [Examples](https://github.com/yaalalabs/agent-kernel/tree/develop/examples): Runnable projects for each framework, deployment target, memory backend, sandbox provider and queue transport. - [PyPI package](https://pypi.org/project/agentkernel/): The `agentkernel` package and its release history. - [Terraform modules](https://registry.terraform.io/namespaces/yaalalabs): The modules for AWS, Google Cloud and Azure. - [Helm chart](https://github.com/yaalalabs/agent-kernel/pkgs/container/charts%2Fagent-kernel): The chart for Kubernetes and on-prem clusters. ## Optional Blog posts: announcements and engineering write-ups, newest first. - [Your Agent Finally Has a Face: AG-UI in Agent Kernel](https://kernel.yaala.ai/blog/agui-agentic-ui): Agent Kernel now speaks AG-UI, the open protocol between agents and the apps people actually use. - [Your Cluster, Your Data, One Helm Chart: Agent Kernel on On-Prem Kubernetes](https://kernel.yaala.ai/blog/kubernetes-on-prem-helm-chart): Agent Kernel now ships an official Helm chart that runs its queue-execution pipeline on any Kubernetes cluster, from a laptop k3d cluster to bare metal and EKS, with NATS, Kafka, or SQS as the broker, KEDA autoscaling,... - [Scheduled Tasks in Agent Kernel: Work That Runs Without Anyone Asking](https://kernel.yaala.ai/blog/scheduled-tasks): Your agent can now work on a clock. - [Scaling Agent Kernel on AWS: Decoupling Request Handling from Agent Execution](https://kernel.yaala.ai/blog/aws-queue-mode-scalability): Agent Kernel now decouples request handling from agent execution on AWS with a durable SQS queue pipeline, so traffic spikes scale your infrastructure instead of overwhelming it, with zero agent-code changes. - [Introducing the Agent Kernel Execution Broker: A Safe Place for AI Agents to Run Code](https://kernel.yaala.ai/blog/agent-kernel-execution-broker): AI agents have become remarkably good at writing code. - [Agent Kernel Now Runs on GCP - AWS, Azure, GCP, On-Prem. One Platform.](https://kernel.yaala.ai/blog/gcp-support): Agent Kernel now ships full Google Cloud Platform support, completing the trifecta of major public clouds. - [Agent Kernel Is Live at Climate Impact X](https://kernel.yaala.ai/blog/agent-kernel-live-at-climate-impact-x): Climate Impact X, a leading exchange for environmental products, is now live with Agent Kernel as part of its trade surveillance workflow. - [Knowledge Bases: The Missing Layer That Lets AI Agents Self-Evolve](https://kernel.yaala.ai/blog/knowledge-bases-self-evolving-agents): Most AI agents have a fundamental flaw: they forget everything. - [Agent Skills: Agent Kernel Builds Enterprise Agents for You](https://kernel.yaala.ai/blog/agent-skills-cli): What if you never had to read a single page of Agent Kernel documentation? - [Walled AI Guardrails in Practice](https://kernel.yaala.ai/blog/walledai-guardrails): Walled AI gives us a clean, practical safety + PII masking pipeline for agent input/output flows. - [Write Once, Run Anywhere: Universal Tools for AI Agents](https://kernel.yaala.ai/blog/framework-agnostic-tool-binding): Stop rewriting the same tools for different AI frameworks. - [Agent Kernel Goes Multi-Cloud: Azure Support is Here](https://kernel.yaala.ai/blog/azure-multi-cloud-support): Your AI investment should never be hostage to a single cloud provider. - [Safe AI at Scale: Introducing Guardrails for Agent Kernel](https://kernel.yaala.ai/blog/guardrails-content-safety): Deploying AI agents in production comes with critical business risks: brand damage from inappropriate responses, regulatory penalties from data leaks, and customer trust erosion from security breaches. - [Power Up Your Agents: Execution Hooks and Smart Memory in Agent Kernel](https://kernel.yaala.ai/blog/hooks-and-smart-memory): Ever wished you could intercept your AI agent's thoughts before they speak? - [Connect Everywhere: Agent Kernel Now Supports Slack, WhatsApp, and Messenger](https://kernel.yaala.ai/blog/messaging-platform-integrations): We're thrilled to announce that Agent Kernel now seamlessly integrates with the world's most popular messaging platforms - Slack, WhatsApp, and Facebook Messenger! - [Production-Grade Observability: Agent Kernel Integrates Langfuse and OpenLLMetry](https://kernel.yaala.ai/blog/blog/observability-support-langfuse-openllmetry): We're excited to announce that Agent Kernel now includes comprehensive observability support through integrations with Langfuse and OpenLLMetry (Traceloop) - giving you powerful insights into your AI agent systems! - [Introducing Agent Kernel - Framework-Agnostic AI Agent Runtime](https://kernel.yaala.ai/blog/blog/announcing-agent-kernel): We're thrilled to announce Agent Kernel (AK) - an open-source, versatile runtime solution that revolutionizes how you develop, test, and deploy AI agents across any framework and environment!