agentbreeder 2.0.0

Define Once. Deploy Anywhere. Govern Automatically.

AgentBreeder™ — v2.0

Stop wrangling agents. Start shipping them.

One YAML file. Any framework. Any language. Any cloud. Governance built in.

Quick Start · Install · Docs · Contributing

---

Your company has 47 AI agents. Nobody knows what they cost, who approved them, or which ones are still running. Three teams built the same summarizer. The security team hasn't audited any of them.

AgentBreeder fixes this.

Write one agent.yaml. Run agentbreeder deploy. Your agent is live — with RBAC, cost tracking, audit trail, and org-wide discoverability. Automatic. Not optional.

---

What's new in v2.0

v2 turns AgentBreeder from a CLI + engine into a full platform substrate. Six tracks ship together; the deploy pipeline contract is unchanged.

Track	Ships	Docs
F — 9-provider catalog	Generic OpenAI-compatible provider + presets for Nvidia · OpenRouter · Moonshot/Kimi · Groq · Together · Fireworks · DeepInfra · Cerebras · Hyperbolic. New agentbreeder provider list/add/test/publish.	providers
G — Model lifecycle	Auto-discover models from each provider's /models endpoint; daily diff; status badges (active/deprecated/retired); agentbreeder model sync.	providers
H — Gateways as first-class	LiteLLM + OpenRouter promoted into the catalog; <gateway>/<provider>/<model> syntax; workspace-level gateway config.	gateways
I — Polyglot SDKs	Stable HTTP runtime contract v1; language: field in agent.yaml; thin SDK targets for Go, Kotlin, Rust, .NET.	runtime contract · polyglot agents
J — Sidecar	Single Go binary auto-injected next to every agent; handles tracing, cost attribution, guardrails, A2A, MCP, bearer-token auth.	sidecar
K — Workspace secrets	OS keychain default; agentbreeder secret set/list/rotate/sync; auto-mirror to AWS Secrets Manager / GCP Secret Manager at deploy.	secrets

Backward-compatible: every v1 agent.yaml continues to work unchanged. v1 providers (openai/anthropic/google/ollama) keep their hand-written classes — the catalog is purely additive.

---

The Problem

AI coding tools make it easy to build agents. Nobody has made it easy to ship them responsibly.

What happens today	What happens with AgentBreeder
Every framework has its own deploy story	One YAML, any framework, any cloud
No RBAC — anyone deploys anything	RBAC validated before the first container builds
No cost tracking — $40k surprise cloud bills	Cost attributed per team, per agent, per model
No audit trail — "who deployed that?"	Every deploy logged with who, what, when, where
No discoverability — duplicate agents everywhere	Org-wide registry — search before you build
Governance is bolted on after the fact	Governance is a structural side effect of deploying

Governance is not configuration. It is a side effect of the pipeline. There is no way to skip it.

---

How It Works

# agent.yaml — this is the entire config
name: customer-support-agent
version: 1.0.0
team: customer-success
owner: [e-mail]

framework: langgraph          # or: openai_agents, claude_sdk, crewai, google_adk, custom

model:
  primary: claude-sonnet-4-6
  fallback: gpt-4o

tools:
  - ref: tools/zendesk-mcp    # pull from org registry
  - ref: tools/order-lookup

deploy:
  cloud: gcp                  # or: aws, azure, local, kubernetes
  scaling:
    min: 1
    max: 10

pip3 install agentbreeder
agentbreeder deploy ./agent.yaml

Eight atomic steps run in sequence: parse → RBAC check → (approval gate if required) → resolve deps → build container → provision infra → deploy → health check → register. If any step fails, the entire deploy rolls back.

---

Three Ways to Build

All three tiers compile to the same internal format. Same deploy pipeline. Same governance. No lock-in.

Tier	Who	How	Eject to
No Code	PMs, analysts, citizen builders	Visual drag-and-drop canvas — pick model, tools, prompts from the registry	Low Code
Low Code	ML engineers, DevOps	Write agent.yaml in any IDE	Full Code (agentbreeder eject)
Full Code	Senior engineers, researchers	Python/TS SDK with full programmatic control	—

from agenthub import Agent

agent = (
    Agent("support-agent", version="1.0.0", team="eng")
    .with_model(primary="claude-sonnet-4-6", fallback="gpt-4o")
    .with_tools(["tools/zendesk-mcp", "tools/order-lookup"])
    .with_deploy(cloud="gcp", min_scale=1, max_scale=10)
)
agent.deploy()

---

What's Supported

Agent languages — Python · TypeScript/Node.js · Go · Kotlin/Java · Rust · .NET (via runtime contract v1, Track I)

Python frameworks — LangGraph · OpenAI Agents · Claude SDK · CrewAI · Google ADK · Custom

TypeScript frameworks — Vercel AI SDK · Mastra · LangChain.js · OpenAI Agents TS · DeepAgent · Custom

Cloud targets — AWS (ECS Fargate, App Runner, EKS) · GCP (Cloud Run, GKE) · Azure Container Apps · Kubernetes (EKS/GKE/AKS/self-hosted) · Local Docker · Claude Managed Agents

LLM providers — direct — Anthropic · OpenAI · Google · Ollama (local, free)

LLM providers — OpenAI-compatible catalog (v2) — Nvidia NIM · Moonshot/Kimi · Groq · Together · Fireworks · DeepInfra · Cerebras · Hyperbolic (plus your own user-local entries)

LLM gateways — LiteLLM (self-hosted proxy) · OpenRouter (200+ models) — see gateways

Secrets backends — OS keychain (default) · .env · AWS Secrets Manager · GCP Secret Manager · HashiCorp Vault — auto-mirrored to the cloud at deploy

RAG & memory — ChromaDB (vector search) · Neo4j (knowledge graph / GraphRAG) · MCP memory server

MCP & A2A — MCP server registry · MCP sidecar injection · Agent-to-Agent (A2A) JSON-RPC protocol · multi-level orchestration

Platform — RBAC · cost tracking · audit trail · org registry · MCP hub · multi-agent orchestration · RAG · evaluations · A2A protocol · AgentOps fleet dashboard · community marketplace · v2 platform sidecar

Full feature matrix and supported versions → docs/features

---

CLI Reference

Command	What it does
agentbreeder quickstart	Full local bootstrap — Docker, stack, seed data, 5 sample agents, dashboard
agentbreeder setup	Configure Ollama + cloud API keys (interactive wizard)
agentbreeder seed	Seed ChromaDB and Neo4j; ingest your own docs with --docs
agentbreeder ui	Start the dashboard + API via Docker (lighter alternative to quickstart)
agentbreeder up / down	Start / stop the full local platform stack
agentbreeder init	Scaffold a new agent project (interactive)
agentbreeder deploy	Deploy an agent (local, AWS, GCP, Azure, K8s)
agentbreeder validate	Validate agent.yaml without deploying
agentbreeder chat	Chat with a deployed agent; --local uses Ollama directly
agentbreeder logs	Stream logs from a deployed agent
agentbreeder status	Show deploy status of all agents
agentbreeder list	List registered agents, tools, models, prompts
agentbreeder search	Search the org registry across all entity types
agentbreeder describe	Show full detail for a registered agent
agentbreeder teardown	Remove a deployed agent and its cloud resources
agentbreeder eval	Run LLM-as-judge evaluations against an agent
agentbreeder eject	Eject from Low Code to Full Code (generates SDK scaffold)
agentbreeder submit	Open a PR for an agent change (git workflow)
agentbreeder review	Review a pending agent PR
agentbreeder publish	Merge an approved agent PR
agentbreeder schedule	Create cron-based scheduled agent runs
agentbreeder provider	List/add/test/publish LLM providers — including the v2 OpenAI-compatible catalog (Nvidia, Groq, Together, …)
agentbreeder scan	Auto-discover Ollama models and MCP servers on your network
agentbreeder secret	Workspace-bound secrets (keychain default) with auto-mirror to AWS / GCP / Vault at deploy
agentbreeder template	Browse and apply agent templates from the marketplace
agentbreeder orchestration	Manage multi-agent orchestrations
agentbreeder compliance	Generate SOC 2 / HIPAA / GDPR / ISO 27001 evidence reports
`agentbreeder registry prompt push	list
`agentbreeder registry tool push	list
`agentbreeder registry agent push	list
agentbreeder --version	Print the installed version

Full CLI reference → agentbreeder.io/docs/cli-reference

---

Install

Requires Python 3.11+:

pip3 install agentbreeder

brew install and npx support are coming soon.

After install, the same commands are available:

agentbreeder quickstart       # full local platform in one command
agentbreeder setup            # configure Ollama + API keys
agentbreeder seed             # seed ChromaDB and Neo4j knowledge bases
agentbreeder deploy           # deploy an agent (local, AWS, GCP, Azure)
agentbreeder chat my-agent    # chat with a deployed agent

agentbreeder: command not found? pip's script directory may not be on your PATH — fix it here.

---

Quick Start

Option A — Full local platform (recommended for first-timers)

pip3 install agentbreeder
agentbreeder quickstart

After it boots, every prompt, tool, and agent lives in the registry — accessible from CLI, the API, or the dashboard:

# Login + export the JWT (CLI commands need it)
TOKEN=$(curl -s -X POST http://localhost:8000/api/v1/auth/login \
  -H 'Content-Type: application/json' \
  -d '{"email":"[e-mail]","password":"…"}' \
  | jq -r '.data.access_token')
export AGENTBREEDER_API_TOKEN=$TOKEN

# Browse and execute
agentbreeder registry prompt list
agentbreeder registry prompt try gemini-assistant-system --input "Greet me"

agentbreeder registry tool list
agentbreeder registry tool run web-search --args '{"query":"What is RAG?"}'

agentbreeder registry agent list
agentbreeder registry agent invoke gemini-assistant \
  --input "What time is it?" \
  --endpoint http://localhost:8080 --token $AGENT_AUTH_TOKEN

The dashboard at http://localhost:3001 has the same affordances under /prompts, /tools, and /agents — including a Try it tab on every tool, a Test tab on every prompt that calls a real LLM, and an Invoke tab on every agent that chats with the deployed runtime.

That single command:

• Detects and guides Docker/Podman install if needed
• Starts the full stack: API · Dashboard · PostgreSQL · Redis · ChromaDB (RAG) · Neo4j (GraphRAG) · MCP servers · LiteLLM gateway
• Seeds a ChromaDB knowledge base and a Neo4j knowledge graph with sample data
• Deploys 5 sample agents (RAG, GraphRAG, MCP search, A2A orchestrator, assistant)
• Opens the visual dashboard at http://localhost:3001

Takes ~3 minutes on first run (image pulls). Then:

agentbreeder chat assistant                   # chat with the assistant agent
agentbreeder chat rag-agent                   # ask questions about AgentBreeder docs
agentbreeder chat graph-agent                 # query the knowledge graph
agentbreeder chat a2a-orchestrator            # let the orchestrator route your question
agentbreeder chat my-agent --local            # chat via Ollama — no API server needed

Deploy to cloud from the same setup:

agentbreeder quickstart --cloud aws           # local + deploy to AWS ECS Fargate
agentbreeder quickstart --cloud gcp           # local + deploy to GCP Cloud Run
agentbreeder quickstart --cloud azure         # local + deploy to Azure Container Apps

Option B — Build your own agent

pip3 install agentbreeder

# v2: pick a provider from the catalog and stash the key in your workspace backend
agentbreeder provider list                # see all 9 OpenAI-compatible presets + legacy providers
agentbreeder secret set NVIDIA_API_KEY    # prompted securely; stored in OS keychain by default

agentbreeder init                         # scaffold a new agent project
agentbreeder validate                     # validate agent.yaml
agentbreeder deploy --target local        # deploy locally with Docker
agentbreeder deploy --target aws          # deploy to AWS ECS Fargate (secrets auto-mirrored)
agentbreeder deploy --target gcp          # deploy to GCP Cloud Run   (secrets auto-mirrored)

Full quickstart guide → agentbreeder.io/docs/quickstart · How AgentBreeder compares →

---

Viewing deployed agents

After deploying, start the UI stack to see your agents in the dashboard (requires Docker):

agentbreeder ui

Then open http://localhost:3001 and log in (default: [e-mail] / plant). Deployed agents appear automatically in the Agents tab.

Docker networking note: Agent containers reach the API at http://host.docker.internal:8000 (macOS/Windows with Docker Desktop) or http://172.17.0.1:8000 (Linux). Use localhost:8000 only from your host terminal.

---

Deploying to production

The reference microlearning-ebook-agent is deployed and serving at https://microlearning-ebook-agent-sizukgalta-uc.a.run.app. The deploy script at microlearning-ebook-agent/scripts/deploy_gcp.sh automates the full flow:

1. Enable GCP APIs (Cloud Run, Artifact Registry, Cloud Build, Secret Manager)
2. Create the image repository
3. Push secrets (GOOGLE_API_KEY, TAVILY_API_KEY, AGENT_AUTH_TOKEN) to Secret Manager
4. Build + push the container via Cloud Build (~3 min)
5. Deploy to Cloud Run with min=0 (scale-to-zero), max=5

cd microlearning-ebook-agent
bash scripts/deploy_gcp.sh
# → Deployed: https://<service>-<hash>-uc.a.run.app

Auth, config, and verification details — agentbreeder.io/docs/deployment

The same pattern (agentbreeder deploy) works for AWS ECS Fargate, App Runner, Azure Container Apps, and Kubernetes — set deploy.cloud: in agent.yaml.

---

Documentation

User docs (guides, references, examples) — agentbreeder.io/docs

Quickstart	Full local platform in one command
Examples	18 working examples — every framework, cloud, and pattern
agent.yaml reference	Every field, every option
CLI reference	All commands and flags
How-To guides	Install, deploy, orchestrate, evaluate
Model Gateway	LiteLLM proxy — routing, budgets, guardrails, caching
RAG & GraphRAG	ChromaDB vector search + Neo4j knowledge graphs
MCP servers	MCP server registry + sidecar injection
A2A protocol	Agent-to-Agent JSON-RPC communication
Comparisons	AgentBreeder vs Google, Anthropic, OpenAI, Azure, AWS
SDK reference	Python + TypeScript full-code SDK

For contributors — internal engineering references in this repo:

ARCHITECTURE.md	Platform architecture — deploy pipeline, abstractions, data model
docs/design/	Feature design docs — RBAC, LiteLLM gateway, polyglot agents
ROADMAP.md	Release plan and milestone status
CHANGELOG.md	Version history
CONTRIBUTING.md	How to contribute — setup, standards, PR process
SECURITY.md	Security policy and vulnerability reporting

---

Contributing · Issues · Discussions · Apache 2.0