Agent Kernel

Why Agent Kernel?

Most agent frameworks help you build a prototype. Agent Kernel is the platform layer that gets you to production — with the governance, portability, and operational maturity that enterprises actually require.

⭐ If Agent Kernel is solving real problems for you, please star the repo — it's the single best way to help us grow.

🚀 Quick Start

Requirements: Python 3.12 – 3.13.x

pip install agentkernel

Build a multi-agent system that runs on OpenAI Agents SDK today and LangGraph tomorrow — same code:

from agentkernel.cli import CLI
from agentkernel.openai import OpenAIModule
from agents import Agent

math_agent = Agent(
    name="math",
    handoff_description="Specialist agent for math questions",
    instructions="You provide help with math problems.",
)

general_agent = Agent(
    name="general",
    handoff_description="Agent for general questions",
    instructions="You provide assistance with general queries.",
)

triage_agent = Agent(
    name="triage",
    instructions="You determine which agent to use based on the user's question.",
    handoffs=[general_agent, math_agent],
)

module = OpenAIModule([triage_agent, math_agent, general_agent])

if __name__ == "__main__":
    CLI.main()

That's it. Same code deploys to AWS Lambda, ECS, Azure Functions, or Container Apps with a single Terraform module. 👉 Get Started

✨ Features

🧩 Works with the Frameworks You Already Use

No rewrites. No re-learning. Bring your agents — Agent Kernel handles the platform layer.

🛡️ Compliance & Governance — First-Class, Not an Afterthought

Enterprises can't ship agents they can't audit. Agent Kernel makes compliance the default:

• Guardrails — OpenAI and AWS Bedrock guardrails for PII detection, jailbreak prevention, content moderation.
• Pre/Post Execution Hooks — Inject policy checks, RAG context, redaction, or moderation around every agent call.
• Full Traceability — Every agent action, tool call, and LLM invocation logged with configurable verbosity.
• Observability — LangFuse and OpenLLMetry tracing with a single config line.
• Data Residency — Pick your cloud, your region, your storage backend. Your data stays where you need it.

📦 Sandboxed Code Execution

Let agents run code and shell commands in an isolated, permission-bounded environment — the platform handles it, your agent code stays clean.

• Enable it in config — agents automatically gain code/command/file tools and the usage guidance is injected into their prompt.
• Pluggable providers — local_subprocess (dev) and docker (container-isolated) today, with e2b, daytona, kubernetes, and cloud-native backends planned; bring your own via a dotted path.
• Workload profiles — per-call, per-session, or shared lifetimes; each with its own permission policy (network egress, filesystem, CPU/memory, timeout) enforced fail-closed.
• Per-user identity — run sandboxed code under the invoking user's identity, not one shared agent identity, via a pluggable principal resolver.

Learn more →

🧠 Memory, Sessions & Knowledge Bases

💬 Messaging Channels — Out of the Box

Slack • WhatsApp • Microsoft Teams • Telegram • Gmail • Messenger • Instagram

Build once. Ship to every channel your users live on. No bespoke bot code.

🔌 Open Protocols

• MCP (Model Context Protocol) — Connect agents to external tools, data sources, and services. Optionally expose your agents as MCP tools.
• A2A (Agent-to-Agent) — Native message passing, handoffs, and coordination between agents in a shared ecosystem.

⚡ Built-In Execution Modes

• Synchronous — Classic request/response.
• WebSocket / Async — Streaming, long-running, real-time agent interactions.
• Stateless MCP — Lightweight agent-as-tool deployments.

☁️ Deploy Anywhere

Same agent code. Pick your runtime. Full Terraform modules included.

🛠 Agent Skills — Supercharge Your Coding Assistant

Install Agent Kernel skills and Claude, Copilot, Cursor, or Windsurf become experts at building production agents — no hallucinated APIs.

ak skill install

See the use-cases/ directory for complete end-to-end examples built using these skills — each starting from a SPEC.md and generating a fully deployed agent.

🧪 Testing — pytest-Integrated

pytest tests/

Built-in fuzzy, semantic, and fallback comparison modes. CI/CD ready. Test agent behavior, not just code.

👥 Who's It For?

• AI Startups — Go from prototype to production in days, not quarters.
• Software Product Companies — Add intelligent agents to your SaaS without framework lock-in.
• Software Services Companies — Ship client agent solutions in weeks, not months.
• Domain Experts — Build production AI products without a full platform engineering team.
• Regulated Enterprises — Get compliance, traceability, and data residency by default.

🌐 Affiliations

Agent Kernel is proud to be:

• A member of the Linux Foundation
• A member of the Agentic AI Foundation

💬 Community & Support

• 📖 Documentation: kernel.yaala.ai/docs
• 💬 Discord: Join the community
• 🐛 Issues: GitHub Issues
• 📦 PyPI: pip install agentkernel
• ☁️ Terraform Registry: Official modules
• 🛠 Developer Guide: DEVELOPER_GUIDE.md
• 🤖 Contributing with AI agents: AGENTS.md

License

Unless otherwise specified, all content — including all source and documentation files in this repository — is:

Copyright (c) 2025-2026 Yaala Labs.

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.

SPDX-License-Identifier: Apache-2.0