Enterprise-Grade AI Agent Orchestration

Build Intelligent Agent Workflows in Java

The only self-improving, multi-agent orchestration framework for Java. 7 process types, YAML DSL, built-in governance, budget tracking, and 24 tools. Powered by Spring AI & Spring Boot.

Maven CentralApache 2.0Java 21
1,013+Tests Passing
9Modules
24Built-in Tools
Java 21Spring Boot 3.4

Add to Your Project in 30 Seconds

Available on Maven Central. Import the BOM and let it manage versions for all SwarmAI modules.

<dependencyManagement>
    <dependencies>
        <dependency>
            <groupId>ai.intelliswarm</groupId>
            <artifactId>swarmai-bom</artifactId>
            <version>1.0.0</version>
            <type>pom</type>
            <scope>import</scope>
        </dependency>
    </dependencies>
</dependencyManagement>

<dependencies>
    <dependency>
        <groupId>ai.intelliswarm</groupId>
        <artifactId>swarmai-core</artifactId>
    </dependency>
    <dependency>
        <groupId>ai.intelliswarm</groupId>
        <artifactId>swarmai-tools</artifactId>
    </dependency>
    <dependency>
        <groupId>ai.intelliswarm</groupId>
        <artifactId>swarmai-dsl</artifactId>
    </dependency>
</dependencies>

info Enterprise + Studio modules are licensed under BSL 1.1 and distributed via GitHub Packages. View all modules on Maven Central

7 Process Types

From simple sequential pipelines to self-improving swarms with dynamic skill generation

linear_scale

Sequential

Tasks run in dependency order, each receiving prior outputs as context.

fork_right

Parallel

Independent tasks run concurrently in layers with automatic synchronization.

account_tree

Hierarchical

Manager agent creates plans, delegates to workers, and synthesizes results.

loop

Iterative

Execute, review, refine loop until the reviewer approves or max iterations reached.

auto_fix_high

Self-Improving

Generates new skills at runtime when capability gaps are detected. RL-powered decisions.

hub

Swarm

Distributed fan-out with parallel self-improving agents per target.

layers

Composite

Chain any processes into a pipeline: Parallel → Hierarchical → Iterative.

Enterprise Ready

Production-grade features that differentiate SwarmAI from personal AI assistants

admin_panel_settings

Governance Gates

Human-in-the-loop approval checkpoints pause workflows for review before sensitive operations.

account_balance_wallet

Budget Tracking

Real-time token and cost tracking with HARD_STOP or WARN enforcement per workflow.

groups

Multi-Tenancy

Tenant-isolated memory, knowledge, quotas, and budgets for enterprise deployments.

security

RBAC & Permissions

Role-based access control with tool permission levels: READ_ONLY, WORKSPACE_WRITE, DANGEROUS.

monitor_heart

Observability

Correlation IDs, structured logging, decision tracing, Prometheus metrics, and health checks.

shield

Circuit Breaker

resilience4j circuit breaker and retry on all LLM calls. Graceful degradation under failure.

Define Workflows in YAML or Java

Zero-code YAML DSL or type-safe Java builders — your choice

swarm:
  process: SELF_IMPROVING
  budget:
    maxTokens: 500000
    maxCostUsd: 5.00
    onExceeded: WARN

  agents:
    researcher:
      role: "Research Analyst"
      goal: "Find accurate data"
      tools: [web-search, sec-filings]
      permissionMode: READ_ONLY

    writer:
      role: "Report Writer"
      goal: "Write executive reports"

  tasks:
    research:
      description: "Research {{topic}}"
      agent: researcher
    report:
      description: "Write report from findings"
      agent: writer
      dependsOn: [research]

Start Building Today

Open source core • Enterprise features with BSL license • Spring Boot native