@agentbreeder/sdk

@agentbreeder/sdk

TypeScript SDK for AgentBreeder — Define Once. Deploy Anywhere. Govern Automatically.

What is this?

@agentbreeder/sdk is the Full Code tier of AgentBreeder. You define agents programmatically in TypeScript using a fluent builder API; the SDK serializes your config to agent.yaml and hands it off to the AgentBreeder deploy pipeline — with RBAC, cost tracking, audit trail, and org-wide discoverability automatic.

The same deploy pipeline powers the No Code (UI drag-and-drop) and Low Code (handwritten YAML) tiers. All three compile to the same format. There is no SDK-only fast path that skips governance.

Install

npm install @agentbreeder/sdk

Requires Node.js 18+. Ships as both ESM and CJS.

Quick Start

import { Agent, Tool, deploy } from "@agentbreeder/sdk";

const agent = new Agent("customer-support", {
  version: "1.0.0",
  team: "customer-success",
  owner: "[e-mail]",
  framework: "langgraph",
})
  .withModel("claude-sonnet-4", { fallback: "gpt-4o", temperature: 0.7 })
  .withPrompt("You are a helpful customer support agent.")
  .withTool(Tool.fromRef("tools/zendesk-mcp"))
  .withTool(Tool.fromRef("tools/order-lookup"))
  .withGuardrail("pii_detection")
  .withDeploy("aws", { region: "us-east-1" })
  .tag("support", "zendesk", "production");

// Validate before committing
const errors = agent.validate();
if (errors.length > 0) {
  console.error("Validation errors:", errors);
  process.exit(1);
}

// Write agent.yaml
import { writeFileSync } from "fs";
writeFileSync("agent.yaml", agent.toYaml());

// Or deploy directly
const result = await deploy(agent.toConfig(), "aws");
console.log("Live at:", result.endpoint);

Multi-Agent Orchestration

The SDK provides four orchestration patterns via purpose-built classes. All patterns serialize to orchestration.yaml and share the same deploy pipeline.

Pipeline — sequential chain

import { Pipeline } from "@agentbreeder/sdk";

const pipeline = new Pipeline("triage-and-resolve", {
  version: "1.0.0",
  team: "customer-success",
})
  .step("triage", "agents/triage-agent")
  .step("resolver", "agents/resolver-agent")
  .step("summarizer", "agents/summary-agent")
  .withDeploy("aws");

writeFileSync("orchestration.yaml", pipeline.toYaml());

FanOut — parallel workers with merge

import { FanOut } from "@agentbreeder/sdk";

const fanout = new FanOut("multi-region-analysis", {
  version: "1.0.0",
  team: "data",
})
  .worker("us-analyst", "agents/regional-analyst")
  .worker("eu-analyst", "agents/regional-analyst")
  .worker("apac-analyst", "agents/regional-analyst")
  .merge("agents/report-merger")
  .withMergeStrategy("aggregate")
  .withDeploy("gcp");

Supervisor — hierarchical delegation

import { Supervisor } from "@agentbreeder/sdk";

const supervisor = new Supervisor("research-team", {
  version: "1.0.0",
  team: "research",
})
  .withSupervisorAgent("planner", "agents/research-planner")
  .worker("web-researcher", "agents/web-search-agent")
  .worker("doc-analyst", "agents/document-analyst")
  .worker("writer", "agents/report-writer")
  .withMaxIterations(5)
  .withDeploy("aws");

Orchestration with routing

import { Orchestration } from "@agentbreeder/sdk";

const router = new Orchestration("support-router", "router", {
  version: "1.0.0",
  team: "customer-success",
})
  .addAgent("billing", "agents/billing-agent")
  .addAgent("technical", "agents/technical-agent")
  .addAgent("returns", "agents/returns-agent")
  .withRoute("billing", "intent == 'billing'", "billing")
  .withRoute("technical", "intent == 'technical'", "technical")
  .withSharedState("redis", "redis://cache:6379")
  .withDeploy("aws");

Key Classes

Class / Function	Description
Agent	Fluent builder for a single agent. Produces AgentConfig / agent.yaml.
Model	Standalone model config builder with chained .fallback(), .temperature(), .maxTokens(), .gateway().
Tool	Wraps a registry ref (Tool.fromRef("tools/...")) or an inline tool definition with .withSchema().
Pipeline	Sequential orchestration — each step receives the previous step's output.
FanOut	Fan-out to parallel workers, merge results with a configurable MergeStrategy.
Supervisor	Hierarchical orchestration with a supervisor agent that delegates to workers.
Orchestration	Base class for custom routing, shared state, and any strategy not covered by the helpers.
KeywordRouter	Routes messages to agents by keyword match.
IntentRouter	Routes based on a pre-classified context["intent"] value.
RoundRobinRouter	Distributes messages across agents in round-robin order.
deploy(config, target?)	Async function that submits an AgentConfig to the AgentBreeder deploy pipeline.
agentToYaml(config)	Serialize a plain AgentConfig object to a YAML string.
orchestrationToYaml(config)	Serialize a plain OrchestrationConfig object to a YAML string.

Agent builder methods

Method	Description
.withModel(primary, opts?)	Set primary model, optional fallback, temperature, maxTokens.
.withPrompt(system)	Set the system prompt (inline string or registry ref).
.withTool(tool)	Add a Tool instance.
.withSubagent(ref, opts?)	Reference another agent as a callable subagent.
.withMcpServer(ref, transport?)	Attach an MCP server (stdio, sse, or streamable_http).
.withGuardrail(name)	Add a guardrail by name (pii_detection, content_filter, hallucination_check, or a custom endpoint).
.withDeploy(cloud, opts?)	Set the deployment target and optional region, scaling, resources, env vars, secrets.
.tag(...tags)	Add discovery tags.
.validate()	Returns an array of validation error strings (empty = valid).
.toYaml()	Returns the agent.yaml content as a string.
.toConfig()	Returns the raw AgentConfig object.

Model builder

Use Model for reusable model configurations shared across agents:

import { Model, Agent } from "@agentbreeder/sdk";

const model = new Model("claude-sonnet-4")
  .fallback("gpt-4o")
  .temperature(0.3)
  .maxTokens(8192)
  .gateway("litellm");

const agentA = new Agent("agent-a", { team: "eng" })
  .withModel(model.toConfig().primary, {
    fallback: model.toConfig().fallback,
    temperature: model.toConfig().temperature,
  });

Tier Mobility

AgentBreeder supports moving between builder tiers at any time — no lock-in.

Start in the SDK, view or share as YAML:

import { writeFileSync } from "fs";
writeFileSync("agent.yaml", agent.toYaml());

The exported YAML is fully human-readable and can be edited directly or committed to your repo as a Low Code artifact.

Start from YAML, eject to SDK:

agentbreeder eject agent.yaml --target typescript

This generates a TypeScript file that recreates the same agent using the SDK, which you can then extend programmatically.

The invariant: every tier compiles to the same agent.yaml schema. The deploy pipeline does not know or care which tier produced it.

TypeScript types

All types are exported from the top-level package:

import type {
  AgentConfig,
  CloudType,        // "aws" | "gcp" | "kubernetes" | "local"
  DeployConfig,
  FrameworkType,    // "langgraph" | "crewai" | "claude_sdk" | "openai_agents" | "google_adk" | "custom"
  McpServerRef,
  ModelConfig,
  OrchestrationConfig,
  PromptConfig,
  SubagentRef,
  ToolConfig,
  Visibility,       // "public" | "team" | "private"
} from "@agentbreeder/sdk";

Links

• Documentation: https://agent-breeder.com/docs
• GitHub: https://github.com/rajitsaha/agentbreeder
• npm: https://www.npmjs.com/package/@agentbreeder/sdk
• Python SDK: agentbreeder-sdk on PyPI
• Issues: https://github.com/rajitsaha/agentbreeder/issues

License

Apache 2.0 — see LICENSE.