How to build an AI design team with AitherOS

How to build an AI design team with AitherOS\n\nWant to move beyond single LLM prompts and create a coordinated team of AI agents that can collaboratively produce design assets? With AitherOS, you can build a self-hosted AI design workforce that plans, discusses, executes, and reviews work together—all through a real-time dashboard. This tutorial walks you through setting up a Design Assets workforce from scratch, configuring agents with MCP tools for image generation, and establishing a kanban workflow for iterative design tasks.\n\n## Prerequisites\n\nBefore you begin, ensure you have:\n- AitherOS installed and running (follow the installation guide)\n- Access to the AitherOS dashboard at http://localhost:3000\n- API keys for image generation services (e.g., OpenAI DALL-E, Stability AI) configured in your environment\n\n## Step 1: Create a Design Assets Workforce\n\nStart by creating an isolated workspace for your design team.\n\n1. In the AitherOS dashboard, click \"Workforces\" in the sidebar.\n2. Click + New Workforce.\n3. Name it Design Assets Team and set the objective: \n \"Collaboratively generate, review, and refine visual design assets based on user requirements.\"\n4. Choose a workspace directory (e.g., ./workforces/design-assets).\n5. Click Create Workforce.\n\n\n\n## Step 2: Add Specialized Agents\n\nA design team benefits from diverse roles. Add these agents to your workforce:\n\n### 2.1 Designer Agent\n- Role: Generates initial design concepts from text prompts.\n- System Prompt: \n You are a creative designer AI. Your goal is to produce high-quality visual concepts based on user descriptions. Iterate on ideas, suggest variations, and explain your design choices.\n\n### 2.2 Reviewer Agent\n- Role: Evaluates designs against brand guidelines and user requirements.\n- System Prompt: \n You are a design critic. Assess each asset for clarity, aesthetic appeal, and alignment with the brief. Provide constructive feedback and specific improvement suggestions.\n\n### 2.3 Producer Agent\n- Role: Refines designs based on feedback and prepares final assets.\n- System Prompt: \n You are a production designer. Take feedback from the Reviewer and implement changes efficiently. Prepare assets in the required formats and resolutions.\n\nTo add agents:\n1. Navigate to your workforce's page.\n2. Click + Add Agent.\n3. Select Create New Agent and fill in the details above.\n4. Repeat for each agent type.\n\n\n\n## Step 3: Configure MCP Tools for Image Generation\n\nAgents need tools to act. We'll configure the Model Context Protocol (MCP) to connect image generation services.\n\n### 3.1 Set Up an MCP Server for Image Generation\nAitherOS includes built-in MCP servers for common tools. For image generation, we'll use the image-gen MCP server.\n\n1. Go to Workforce Settings \u2192 MCP Servers.\n2. Click + Add MCP Server.\n3. Choose Image Generation from the template list.\n4. Configure the provider:\n - For DALL-E: Set provider to openai and add your API key.\n - For Stable Diffusion: Set provider to stability and add your API key.\n5. Save the server.\n\n\n\n### 3.2 Assign Tools to Agents\nAssign the image generation tool to your Designer and Producer agents.\n\n1. Edit an agent (e.g., Designer Agent).\n2. Under Tools, enable the MCP server you just created.\n3. Specify the tools available (e.g., generate_image, edit_image).\n4. Save the agent configuration.\n\nRepeat for the Producer Agent. The Reviewer Agent may not need image generation tools but could benefit from file analysis tools.\n\n## Step 4: Configure the Kanban Workflow\n\nUse the kanban board to manage design tasks from concept to completion.\n\n1. In your workforce, click the Kanban tab.\n2. Create columns that reflect your design process:\n - Backlog\n - Concepting\n - In Review\n - Refining\n - Done\n3. Optional: Add WIP limits to prevent bottlenecks.\n\n\n\n## Step 5: Run Your First Design Task\n\nNow, put your team to work.\n\n1. Create a new task in the Backlog column:\n - Title: "Social media banner for product launch"\n - Description: \n \"Create a banner announcing our new product launch. Use brand colors (blue and orange), include space for a headline and call-to-action button. Target audience: tech professionals.\"\n - Attach any reference files or brand guidelines.\n2. Move the task to Concepting.\n3. Assign the task to the Designer Agent (or let the team self-assign via discussion).\n4. Watch as agents collaborate:\n - The Designer Agent generates initial concepts using the image generation tool.\n - The Reviewer Agent provides feedback.\n - The Producer Agent iterates based on feedback.\n5. Move the task through columns as work progresses.\n6. When satisfied, move to Done and download the final asset from the task's output.\n\n\n\n## Step 6: Iterate and Improve\n\nLeverage AitherOS's memory and discussion features:\n- Agents share context via long-term memory, remembering past designs and feedback.\n- Use the Discussion tab in each task to see agent debates and rationale.\n- Adjust agent prompts or MCP tool configurations based on outcomes.\n\n## Conclusion\n\nYou've now built a self-hosted AI design team that can autonomously handle creative workflows. By combining specialized agents, MCP-powered tools, and a visible kanban process, AitherOS enables true agent coordination—not just chaining. This approach produces higher-quality, more consistent results than individual LLM calls.\n\nReady to start? Download AitherOS and run your first AI workforce today.\n\n> External Reference: Learn more about the Model Context Protocol powering AitherOS's tool layer: Model Context Protocol Specification\n\n---\n*Internal Links*: Why multi-agent teams outperform single LLM calls | AitherOS Architecture Documentation