Switch cloud platform later?
Bedrock / Azure / GoogleRewrite
LangGraph · CrewAI · OpenAIYou build it
Agent KernelOne config change
Step 01: Analogy
Guardrails, Traceability, Memory and Other Enterprise Features for Your Agent Code.
Agent Kernel is a unified, capable runtime for the AI agents. Its pluggable architecture allows you to attach capabilities to your effortlessly.
01
Your Agent Logic
Domain-Specific Agent CodeBusiness RulesPrompts & Tools
02
Agent Kernel Core
Agent ModuleAgent WrapperFramework-Specific RunnerSession ManagerRuntimePre / Post Hooks
03
Framework Adapters
LangGraphOpenAI AgentsCrewAIGoogle ADKSmolagentsLiveKitBring your own [advanced]
04
Storage & Memory
In-MemoryRedisDynamoDBCosmosDBFirestore
05
Knowledge Bases
ChromaDBNeo4jStarburstSQLDB
06
Observability
LangFuseOpenLLMetry
07
Execution Surface
AWS LambdaECSAzure FunctionsContainer AppsGCP Cloud RunGCP Cloud Run Functions
08
Interfacing
CLIMCPA2AREST API
09
Channels
SlackTeamsWhatsAppTelegramMessengerInstagramGmailRedisDynamoDB
Step 02: Stand Out
What Makes Agent Kernel Stand Out
| Bedrock AgentCore / Azure AI Foundry / Google Vertex AIBedrock / Azure / Google | LangGraph · CrewAI · OpenAI Agents etcLangGraph · CrewAI · OpenAI | Agent Kernel | |
|---|---|---|---|
| Switch cloud platform later? | Rewrite | You build it | One config change |
| Multi-framework agent execution? | Not possible | Not possible | Run in one runtime |
| Out of Box integrations?(i.e. Slack / Teams / REST / A2A / MCP) | Partial | DIY | Built-in |
| Sessions, memory, observability? | Proprietary | DIY | Built-in and Pluggable |
| Open-source / no licensing? | Apache 2.0 | ||
| Knowledge bases? | Proprietary | DIY | Built-in and Pluggable |
| Lift-and-shift an existing agent? | Rewrite | Rewrite | Wrap & ship |
Multi-framework agent execution?
Bedrock / Azure / GoogleNot possible
LangGraph · CrewAI · OpenAINot possible
Agent KernelRun in one runtime
Out of Box integrations?(i.e. Slack / Teams / REST / A2A / MCP)
Bedrock / Azure / GooglePartial
LangGraph · CrewAI · OpenAIDIY
Agent KernelBuilt-in
Sessions, memory, observability?
Bedrock / Azure / GoogleProprietary
LangGraph · CrewAI · OpenAIDIY
Agent KernelBuilt-in and Pluggable
Open-source / no licensing?
Bedrock / Azure / Google
LangGraph · CrewAI · OpenAI
Agent KernelApache 2.0
Knowledge bases?
Bedrock / Azure / GoogleProprietary
LangGraph · CrewAI · OpenAIDIY
Agent KernelBuilt-in and Pluggable
Lift-and-shift an existing agent?
Bedrock / Azure / GoogleRewrite
LangGraph · CrewAI · OpenAIRewrite
Agent KernelWrap & ship
Step 03: Features
Focus on Agent Logic.
We Handle the Rest.
All Enterprise Features Available Free And Open-Source
Build & Interface
CLI for Prototyping
Easy interfacing your agents on your laptop via Agent Kernel's command line interface.
REST API Server
FastAPI-based server out of the box. No boilerplate. Just run your agent.
Native MCP and A2A support
Expose agents as tools (MCP) and enable agent-to-agent collaboration (A2A). Makes integration with external AI systems straightforward.
Multi-Framework Support
Run OpenAI Agents, LangGraph, CrewAI, Google ADK, Smolagents, LiveKit side-by-side. Keep one runtime across teams while using the best framework per use case.
Runtime & Extensibility
Pluggable Session & Memory
Start local in-memory, scale to Redis, DynamoDB, or Cosmos DB in production. Switch via config, not code rewrites.
Execution Hooks
Pre/post hooks for RAG injection, input validation, response moderation, analytics.
Multimodal Support
In-built framework-neutral multimodal support across all integration channels. Handle files/images cleanly and keep sessions lightweight. Additional voice and video support via LiveKit.
Ship & Secure
Guardrails and Content Safety
Input and output protection in the same runtime pipeline. Supports policy checks for safety, PII handling, and jailbreak defense.
Cloud Deployment
Pre-built Terraform modules for AWS Lambda, ECS, Azure Functions, Container Apps, GCP Cloud Run, GCP Cloud Run Functions.
Reliability
Built for resilient cloud deployments with health checks and failover patterns.
Integrate & Observe
Messaging Integrations
Slack, WhatsApp, Instagram, Telegram, Gmail, Teams, Messenger plug and play.
Testing Framework
pytest-integrated test runner. Write deterministic automated test scenarios for your AI agents like any other code.
Observability
Langfuse and OpenLLMetry tracing with one config line. No manual instrumentation. Trace requests, latency, tool calls, and token behavior.
Step 04: Frameworks
Use The Framework You Prefer
Choose a supported framework that fits your team, while Agent Kernel gives you a consistent production-ready layer for deployment, APIs, sessions, and integrations.
Installation:
| 1 | # 1. Install the CLI |
| 2 | pip install agentkernel[openai] |
Basic Usage:
| 1 | from agents import Agent as OpenAIAgent |
| 2 | from agentkernel.cli import CLI |
| 3 | from agentkernel.openai import OpenAIModule |
| 4 | |
| 5 | agent = OpenAIAgent( |
| 6 | name="assistant", |
| 7 | instructions="You are a helpful assistant.", |
| 8 | ) |
| 9 | |
| 10 | OpenAIModule([agent]) |
| 11 | |
| 12 | if __name__ == "__main__": |
| 13 | CLI.main() |
Step 05: Build Agents
How Agent Kernel Helps You Build Production-Ready Compliant AI Agents
Building AI Agents
Guardrails
Multi-Agent System
Agent Memory
Supervisor Agent
General Agent
Travel Expert Agent
Booking Agent
Destination Discovery Tool
Localized Knowledge Query Tool
Flight Booking Tool
Hotel Booking Tool
Language Detection and Translation Tool
Guardrails
Step 06: Complete Picture
How Agent Kernel Fits In
You write your AI agent's logic. Agent Kernel handles everything else: the infrastructure, the cloud deployment, memory, knowledge bases, hooks, observability & traceability, LLM cost tracking, and the integrations so your agent is live and talking to real users in days.
Your Agent Logic
OpenAILangGraphCrewAIGoogle ADKSmolagentsLiveKit
Agent Kernel Runtime
Framework Adapters
Session & Memory
Execution Nodes
Knowledge Bases
Messaging
Observability
Guardrails
Integrations
AWS Lambda
AWS ECS
Azure Functions
Google Apps
Docker
Step 07: Operating System Depth
Why Agent Kernel is a Powerful
Operating System
Three-Layer Memory
Three Memory Layers, Zero Context Chaos
Keep conversations coherent, enrich each request, and carry useful session data forward without bloating the model context window.
Conversational State (Session)
- The agent remembers the conversation naturally across turns.
- This is the memory that keeps chat continuity.
Volatile Cache (Per Request)
- A scratchpad for one request only.
- Great for RAG snippets, temporary file reads, and intermediate results.
- Auto-clears after every response.
Non-Volatile Cache (Session)
- Session memory for supporting data that should persist.
- Perfect for user preferences, auth context, and running counters.
- Available to tools/hooks without spending LLM tokens.
Execution Hook Pipeline
Control Every Request Without Rewriting Agent Logic
Use hooks before and after agent execution to enforce safety, enrich inputs, and polish outputs.
Pre-Execution Hooks
- Validate and enrich input before the LLM runs.
- Typical uses: guardrails, redaction, RAG injection, request shaping.
Agent Execution
- Your existing framework logic runs as-is.
Post-Execution Hooks
- Apply output checks and final formatting before returning to users.
- Typical uses: moderation, compliance notes, analytics, audit logging.
Early Termination
- Stop unsafe or invalid requests early and return a controlled response.
- Useful for input guardrails, rate limits, cached shortcuts, and circuit breakers.
✦Continue Exploring
Go deeper with Agent Kernel
Explore the platform capabilities and real-world workflows behind secure, production-ready AI agents.
Ready to Ship Your
First Agent?
Free, open-source, Apache 2.0. No licensing costs, no vendor lock-in. Join hundreds of developers building production AI agents with Agent Kernel.
Ask AI