Agent Orchestrator

Multi-Agent Workflows in Markdown

Define multi-step AI agent workflows in a simple markdown DSL. The compiler converts them into deterministic execution plans, then the orchestration engine runs each step with the right model, tools, and context. Parallel steps, approval gates, fan-out loops, and real-time progress tracking.

Get Started Free
Usage-based per token + per execution

What it does

Agent Orchestrator lets you define complex multi-agent workflows using a simple markdown format. You write steps in natural language, assign AI models and tools to each step, and the orchestrator handles compilation, validation, execution, and monitoring. No infrastructure to manage — workflows run on a managed orchestration engine with per-step agent invocations.

Each workflow step is an independent AI agent call. Steps can run in sequence or parallel, reference outputs from earlier steps, loop over collections, and pause for human approval. The orchestrator compiles your markdown into a deterministic JSON execution plan, validates it, then executes it step-by-step with full observability.

See It In Action

How It Works

1
Define
Markdown DSL
2
Compile
AI-powered
3
Validate
Schema + DAG check
4
Execute
State machine
5
Monitor
Real-time progress
Architecture
AI Compiler State Machine Agent Runtime MCP Tools

Markdown Workflow DSL

Write workflows the way you think about them — in natural language with simple markdown formatting. YAML frontmatter configures agents, models, and inputs. Each ## Step heading becomes a workflow step with its own model, prompt, and dependencies. No JSON, no YAML graphs, no proprietary DSL.

Parallel steps Step dependencies Fan-out loops Approval gates
Read the DSL docs

Workflow Copilot

New

Describe what you want in plain English and the AI copilot generates the workflow markdown for you. It knows your connected MCP servers and automatically adds the right Tools: directives. Chat back and forth to refine steps, add approval gates, change models, or restructure the workflow — each response updates the editor in real time.

Natural language input Tool-aware generation Iterative refinement Haiku or Sonnet models
Learn more in the console guide

Console Dashboard

Manage workflows from the console with a full-featured editor featuring syntax highlighting, model autocomplete, and live validation. The built-in AI copilot helps you write workflows conversationally. Browse example templates, run workflows with custom inputs, and monitor executions with real-time step-by-step progress tracking.

AI workflow copilot CodeMirror editor Template gallery Execution monitoring
Open the console

Native MCP Tool Access

New

Every MCP server connected to your gateway is available to your workflow agents. Built-in services like RAG Engine and Media Storage work out of the box — plus any server you've added: GitHub, Neon, Supabase, Zapier, and more. Just add Tools: rag-engine to an agent definition and it can search your knowledge base during execution.

RAG Engine search Media upload & storage GitHub, Slack, Jira Any MCP server
Read the tool integration guide

Key Features

Markdown-First

Write workflows in markdown with YAML frontmatter. No proprietary DSL or visual builder required.

Multi-Model Support

Assign different models per step — Haiku for fast tasks, Sonnet for reasoning, Opus for complex analysis. Use logical model names that resolve automatically.

Parallel Execution

Steps without dependencies run in parallel automatically. Fan-out over collections with for_each loops.

Approval Gates

Pause execution for human review before critical steps. Approve or reject from the console dashboard.

Real-Time Monitoring

Track execution progress step-by-step. See token usage, model selection, and output for each step as it completes.

Native MCP Tool Access

Agents connect natively to MCP Gateway Pro. Add Tools: rag-engine, github to any agent and it can search your knowledge base, create issues, send messages, and more.

Example Workflow

research-report.md
# Market Research Report

## Inputs
- topic: The subject to research (string, required)

## Agents

### Researcher
Expert market analyst who gathers data from internal
knowledge base and identifies key trends.
Model: Sonnet
Tools: rag-engine

### Writer
Technical report writer who produces clear, structured reports.
Model: Sonnet

## Steps

### Research
Search the knowledge base for information about {inputs.topic}.
Identify key players, trends, and opportunities.

### Write Report
Using the research: {steps.research.output}

Write a comprehensive market research report with
executive summary, key findings, and recommendations.

REST API

The orchestrator has a full REST API for programmatic workflow management and execution.

POST /workflows Create & compile workflow
POST /workflows/{id}/run Execute with inputs
GET /executions/{id} Status + progress
GET /executions/{id}/steps Per-step results
DELETE /executions/{id} Stop execution

Pricing

Standard tokens $1/1M
Haiku, Nova Lite/Pro
Advanced tokens $10/1M
Sonnet, Nova Premier
Premium tokens $50/1M
Opus
Workflow execution $0.01
Minimum No minimum

Tokens billed per million. Execution fee per workflow run.

Specs

Models Haiku, Sonnet, Opus
Max steps 50 per workflow
Max turns 50 per step
MCP Tools Any gateway server
Execution engine Managed state machine
Uptime SLA 99.9%

Dashboard

Workflows — create & manage
Copilot — AI workflow assistant
Editor — syntax highlighting & autocomplete
Gallery — example templates
Executions — run & monitor
Step detail — output & tokens

Ready to get started?

Deploy Agent Orchestrator in minutes. Usage-based pricing, no upfront commitment.