⚡AgentBreeder Cloud is coming — managed infrastructure, zero DevOps, automatic governance.Learn more ↓

Build agents.
Deploy anywhere.
Govern automatically.

One YAML file. Any framework. Any cloud. Governance, RBAC, cost tracking and audit trail — automatic on every deploy.

$pip3 install agentbreeder
Get started →★  Star on GitHub
agent.yamlready to deploy
# identity
name: customer-support-agent
version: 1.0.0
team: customer-success
framework: langgraph
# model config
model:
  primary: ollama/gemma3
  fallback: gpt-4o
  temperature: 0.7
# run locally
deploy:
  cloud: local
  runtime: docker
✦Prompts

versioned · cached

variables · v3

◈RAG

vector · graph

hybrid search

⬡MCP

4 servers · auto-discover

tool registry

⇄A2A

agent-to-agent

JSON-RPC · auth

terminal
$ agentbreeder deploy
✓ Validating · Building · Deploying
✓ Prompts cached · RAG indexed · MCP wired
→ http://localhost:8080/support-v1
Works with every major framework
LangGraphCrewAIClaude SDKGoogle ADKOpenAI AgentsCustom

Agent Architect

From idea to deployed agent

Not sure which framework, model, or RAG setup is right? Run /agent-build in Claude Code — it interviews you, recommends the full stack, and scaffolds a production-ready project.

agent-architect-demo
Advisory Interview
Generated Project
StepStarting...

Agent for All

No matter your role, you ship faster.

Business users drag and drop. Engineers write YAML. Researchers use the full SDK. All three compile to the same pipeline, with the same governance, to every cloud.

Starting…

agentbreeder
No Code
Low Code
Full Code
Prompts
RAG
MCP
A2A
All 6 Frameworks — Every Tier
✓LangGraph
✓OpenAI Agents
✓Claude SDK
✓CrewAI
✓Google ADK
✓Custom
All 6 Deployment Targets — Every Tier
✓Local
✓Cloud Run
✓ECS Fargate
✓App Runner
✓Azure
✓Kubernetes
Tier Mobility — No Lock-in
No Code → view YAML → Low Code
Low Code → eject → Full Code
All tiers → same 8-step pipeline → auto governanceagentbreeder eject --to sdk

Deploy Anywhere

One command. Every cloud.

agentbreeder deploy runs the same 8-step atomic pipeline regardless of target — from Local Docker Compose to AWS ECS Fargate, GCP Cloud Run, and Azure Container Apps.

agentbreeder-deploy
Deploy Pipeline
Deployment Targets
pending
Local
Docker Compose
localhost:8080
pending
Cloud Run
GCP · Serverless
agent-abc.a.run.app
pending
ECS Fargate
AWS · Serverless
ecs.us-east-1.aws
pending
App Runner
AWS · Managed
apprunner.aws.com
pending
Azure
Container Apps
agent.azurecontainerapps.io
pending
Kubernetes
EKS / GKE / AKS
k8s.cluster.local
StatusStarting…
✓
Registered in org registry
✓
RBAC enforced
✓
Cost attributed to team
✓
Audit log written

The Market in April 2026

Every major platform just shipped.
Every one locks you in.

Google, Anthropic, OpenAI, and Azure all announced enterprise agent platforms in 2025–2026. Each requires your agents to run on their infrastructure. AgentBreeder deploys to any cloud — and makes governance automatic regardless of where you run.

Google Gemini Agent PlatformApr 2026
Deploy targetGCP only
Open source✗
Your data onGoogle Cloud
Claude Managed AgentsApr 2026
Deploy targetAnthropic only
Open source✗
Your data onAnthropic infra
OpenAI Agent Platform2025
Deploy targetOpenAI API only
Open source✗
Your data onOpenAI infra
Azure AI FoundryMay 2025
Deploy targetAzure only
Open source✗
Your data onAzure
YOUAgentBreederOpen source
Deploy targetAWS · GCP · Azure · K8s · local
Open source✓
Your data onYour infra

* Google Gemini Enterprise Agent Platform announced April 22, 2026. Claude Managed Agents GA April 2026. Azure AI Foundry GA May 2025.Read the full analysis →

Architecture

Five layers. One pipeline.

AgentBreeder is a layered platform. Every agent moves through all five layers in order — from how it is built to how it is governed. Click any layer to explore.

Registry

Build once. Reference everywhere.

Prompts, tools, knowledge bases, and MCP servers live in a shared org registry. Define once — wire into any agent, any framework, any cloud.

Prompt Lifecycle

1
2
3
4
5

agent.yaml

name: support-agent
framework: langgraph
 
prompts:
  system: prompts/support-v2← registry ref
  # or inline:
  # system: 'You are a {{role}}'
GET /registry/prompts/{id}/versions/history/diff — compare any two versions
name: support-agent
framework: openai_agents
 
tools:
  - ref: tools/zendesk-mcp← registry ref
  - ref: tools/order-lookup← registry ref
  # or inline: name, type, schema
POST /tools/sandbox/execute — test before wiring to agent
name: research-agent
framework: claude_sdk
 
knowledge_bases:
  - ref: kb/product-docs← registry ref
  - ref: kb/return-policy← registry ref
# queried automatically at runtime
POST /rag/search — hybrid vector + text, configurable weights
name: support-agent
framework: langgraph
 
tools:
  - ref: mcp-servers/zendesk← registry ref
  - ref: mcp-servers/slack← registry ref
# sidecar injected at deploy time
POST /mcp-servers/{id}/discover — enumerate available tools

Prompts

Versioned · diffable · testable

Tools

Sandboxed · schema-validated

Knowledge Bases

Hybrid search · auto-chunked

MCP Servers

Auto-discovered · sidecar-deployed

Why AgentBreeder

Everything you need to ship agents

Stop reinventing deployment, governance, and observability for every agent. AgentBreeder handles it automatically.

🔌

Framework Agnostic

LangGraph, CrewAI, Claude SDK, Google ADK, OpenAI Agents. One pipeline for all frameworks — no lock-in.

☁️

Multi-Cloud

AWS ECS · GCP Cloud Run · Azure Container Apps · Kubernetes · local. One `agent.yaml`, any target — no cloud lock-in.

🔒

Auto Governance

RBAC, cost attribution, audit trail, and org registry registration happen automatically on every deploy.

🗂️

Shared Registry

Agents, prompts, tools, MCP servers, models — one org-wide catalog. Search and reuse across teams.

🎯

Three Builder Tiers

No Code → Low Code → Full Code. Start visual, eject to YAML, eject to SDK. No lock-in at any level.

🔗

Multi-Agent Orchestration

6 orchestration strategies — router, sequential, parallel, supervisor, hierarchical, fan-out — via YAML or SDK.

🏆

LLM-as-Judge Eval Hub

Multi-criteria scoring (accuracy, helpfulness, safety, groundedness) via Claude, GPT-4o, or Gemini. Public leaderboard, regression detection, CSV export.

🛒

Provider Catalog (v2.0)

10 provider presets out of the box — 8 direct OpenAI-compatible (Nvidia NIM, Kimi K2 / Moonshot, Groq, Together, Fireworks, DeepInfra, Cerebras, Hyperbolic) plus 2 gateways (OpenRouter, LiteLLM). `agentbreeder provider add` for any custom upstream.

🛡️

Sidecar Pattern (v2.0)

Single Go binary auto-injected next to every agent. Bearer auth, OTel tracing, cost attribution, PII guardrails, A2A JSON-RPC, MCP passthrough — zero per-language re-implementation.

🔐

Workspace Secrets (v2.0)

OS keychain · Vault · AWS · GCP — pick one per workspace. `agentbreeder deploy` auto-mirrors declared secrets to the target cloud and grants the runtime SA `secretAccessor`. No plaintext in the image.

🌐

Polyglot Runtime Contract (v2.0)

Versioned HTTP contract every agent satisfies. First-party Go SDK ships v2.0 (Kotlin, Rust, .NET on the way). Any language with an HTTP server is a Tier-3 citizen — generate from the OpenAPI.

🔀

Gateways as First-Class (v2.0)

LiteLLM and OpenRouter promoted to `type: gateway` providers. Three-segment refs route the request: `model: openrouter/moonshotai/kimi-k2`. Configure once per workspace; switch upstreams without touching code.

How it works

From YAML to production in 3 steps

No infrastructure expertise required. AgentBreeder handles the entire deploy pipeline.

01 ──

Define your agent

Write agent.yaml with your model, tools, prompts, and deploy config. Schema-validated, human-readable, version-controlled.

02 ──

Run one command

agentbreeder deploy validates, builds a container, provisions infra, and deploys to your cloud in under 5 minutes.

03 ──

Governance is automatic

Your agent is live, registered in the org registry, RBAC enforced, costs attributed, audit trail written. Nothing extra to configure.

Built by

The inventor

Rajit Saha
LinkedIn

Rajit Saha

Inventor & Author

Director of Data Platform · Udemy + Coursera

Spent 20 years making data platforms bigger and faster. Then decided smarter was more interesting. At Udemy + Coursera, shipped AI agents that actually do things in production. AgentBreeder is the tool I kept wishing existed while building them.

The pattern across 8 companies and 23 years: architect it, build the MVP personally, hand it to a great team to scale. Passive data warehouses had a good run. Active, agent-driven systems are what comes next — and that's what I'm building now.

23+ years
industry experience with distributed systems & data infrastructure
8 companies
Oracle · IBM · Yahoo · Teradata · VMware · LendingClub · Experian · Udemy + Coursera
Coming soon

AgentBreeder Cloud

The open-source CLI you love — now with a fully managed control plane. Ship your first production agent in under 5 minutes, with no infrastructure to manage.

Same agent.yaml format. Same CLI. Just add cloud: claude-managed to your deploy block.

⚡

Zero-config deploys

Push agent.yaml. We provision, scale, and secure the infrastructure. No AWS console, no Kubernetes YAML, no Terraform.

🔒

Governance out of the box

RBAC, cost attribution, audit trail, and PII guardrails enforced at the platform level — not bolted on per team.

📊

Fleet-wide observability

Every LLM call traced, every token counted, every tool execution logged. Full cost tracking across all agents and teams.

🌐

Multi-cloud by default

Deploy to AWS, GCP, or Azure with one flag — same agent.yaml, same governance, same CLI.

🏪

Private marketplace

Share agents, prompts, tools, and MCP servers across your org. One-click deploy from an internal catalog.

🤝

Agent-to-agent network

A2A protocol built in. Your agents can discover and call each other across teams, clouds, and frameworks.

Early access opens Q3 2026 · No credit card required