DevTeam Orchestrator
Production-grade AI agent orchestration platform. Deploy, manage, and scale AI agents across distributed infrastructure with DAG workflows, human-in-the-loop approvals, and multi-model routing.
DevTeam Orchestrator v1.0 is now available. Get started in 5 minutes.
What is DevTeam Orchestrator?
DevTeam Orchestrator is a platform for running AI agents at scale. It provides:
- DAG Workflows -- Define complex multi-step agent pipelines as directed acyclic graphs. Each node can use a different model, tool, or data source.
- Human-in-the-Loop (HITL) -- Route high-stakes decisions to human reviewers with configurable confidence thresholds, timeout escalation, and notification channels.
- Multi-Model Routing -- Automatically select the best model for each task based on cost, latency, and capability. Supports Claude, GPT, Gemini, Llama, and 100+ models via OpenRouter.
- Template System -- Pre-built industry-specific workflow templates for legal, finance, DevOps, and more. Deploy complex agent pipelines in a single command.
- Distributed Workers -- Run agent workers on any machine in your cluster. Workers auto-register, accept tasks from Temporal queues, and report health via heartbeat.
- Real-Time Dashboard -- Monitor agent status, task progress, queue depths, and resource usage through a live WebSocket-powered dashboard.
Architecture Overview
+------------------------+
| Dashboard (Svelte) |
| doc.devteam.marsala |
+----------+-------------+
|
| WebSocket + REST
v
+------------------+ +----------+-------------+ +------------------+
| SDK / CLI | ---> | API Server (Express) | ---> | Temporal |
| devteam-sdk | | devteam.marsala.dev | | Workflow Engine |
+------------------+ +----------+-------------+ +--------+---------+
| |
+----------+-------------+ +---------+----------+
| PostgreSQL | | Worker Nodes |
| Tasks, Plans, HITL | | GPU / CPU pools |
+------------------------+ +--------------------+
|
+---------+----------+
| Weaviate RAG |
| Vector Search |
+--------------------+Quick Example
import { DevTeamClient } from 'devteam-sdk';
const client = new DevTeamClient({
apiUrl: 'https://devteam.marsala.dev',
apiKey: process.env.DEVTEAM_API_KEY,
});
// Create a DAG workflow
const plan = await client.createPlan({
name: 'contract-review',
steps: [
{ id: 'extract', prompt: 'Extract key clauses from this contract', model: 'sonnet' },
{ id: 'analyze', prompt: 'Analyze risks in these clauses', model: 'opus', dependsOn: ['extract'] },
{ id: 'report', prompt: 'Generate executive summary', model: 'sonnet', dependsOn: ['analyze'] },
],
});
// Execute with HITL approval on high-risk steps
const result = await client.executePlan(plan.id, {
input: { contract: contractText },
hitl: { requiredFor: ['analyze'], confidenceThreshold: 0.85 },
});Key Concepts
Supported Models
| Provider | Models | Routing Key |
|---|---|---|
| Anthropic | Claude Opus 4.6, Sonnet 4, Haiku 3.5 | opus, sonnet, haiku |
| OpenAI | GPT-4o, GPT-4o-mini | gpt4o, gpt4o-mini |
| Gemini 2.0 Flash, Gemini 2.0 Pro | gemini-flash, gemini-pro | |
| Meta | Llama 3.3 70B, Llama 4 Maverick | llama-70b, llama-maverick |
| Local | Any Ollama model | ollama:<model-name> |
| OpenRouter | 400+ models | openrouter:<model-id> |
Next Steps
- Quick Start -- Get running in 5 minutes
- SDK Reference -- Full TypeScript API documentation
- CLI Reference -- Command-line interface guide
- Template Gallery -- Browse pre-built workflow templates
- Deployment Guide -- Production deployment patterns