AI Architect Walkthrough
This walkthrough guides you through creating, loading, and exporting an app blueprint using AI Architect, then using that blueprint to generate and run an atPlatform app in your IDE.
1. Get Your Starter Pack Atsigns
Before building or testing any atPlatform app, you need two Atsigns. These act as the identities your app will use during testing and development.
Visit my.atsign.com/starterpack.
Verify your email and claim your two free starter-pack Atsigns.
These Atsigns will be used later when you run your generated app to test authentication and secure communication.
2. Open AI Architect
AI Architect is the visual blueprinting tool used to design your app’s structure before generating the LLM prompt.
Go to aiarchitect.atsign.com.
This opens the workspace where you create or load a Blueprint. The Blueprint you create here becomes the input for your LLM-powered code generation.
A Blueprint is a visual map of your application. Each box represents a node. This could be a person, a process, an AI agent, a service, or any other entity involved in your system. The lines between nodes show how information flows from one part of the system to another.
Every node includes a Notes section, which acts as the node’s job description. This is where you define:
what the node is responsible for
what information it needs
how it behaves
A clear Blueprint gives the LLM the structure it needs to build your application in stages.
3. Load the Demo Blueprint
To help you get started quickly, you can load a prebuilt example: a secure messaging app Blueprint.
Download the demo file from this link: Secure Messaging Example.
Select Load From File and choose the file you just downloaded.
AI Architect will populate the canvas with the full secure messaging blueprint.
The demo blueprint is a complete secure‑messaging example which gives you a fully populated map containing:
14 nodes (users, processes, data objects)
9 connections
Messaging flows, group logic, encryption paths
Entities such as messages, threads, files, and voice notes
Processes such as identity management, policy enforcement, and message pipeline.
4. Export the Prompt
Once the blueprint is loaded:
Select Export Prompt.
Copy the generated prompt to your clipboard.
Export Guide generates the full LLM prompt based on your blueprint.
The prompt includes:
A high‑level application description
A breakdown of all nodes (people, processes, things)
A breakdown of all connections and their types
atPlatform roles for each component
Implementation notes for each node
Stream/notification patterns for each connection
Required dependencies and initialization code
Authentication setup
A step‑by‑step implementation guide
This becomes the instruction set your LLM will follow to generate the full application.
5. Open Your IDE, Plan and Code your App
You can use any LLM or IDE you like. We recommend using Visual Studio Code and will refer to it here.
Visual Studio Code gives you the flexibility to work with a variety of LLMs, not just one. Depending on the extensions you install, you can choose from models like OpenAI, Gemini, Claude, Copilot, and others.
To get started, you’ll need to create a new empty folder (without any spaces in the name) and set your LLM to Plan Mode. Starting in Plan Mode is important because it helps to ensure the LLM is going to build what you want it to build.
Paste the exported prompt directly into the chat window. It will plan the project and present the plan to you. When you are happy, Proceed with implementation and it will create files, and set up the app.
You will be asked to confirm certain actions (file creation, folder setup, dependency installation).
The LLM may build the application in stages, allowing you to test each step and provide additional instructions. It will continue refining and completing the app based on your original prompt as you guide it through each iteration.
The LLM will:
Parse the entire blueprint
Create the full folder structure
Generate Dart/Flutter code using at_client and related packages
Implement each node as a module or service
Implement each connection using streams or notifications
Set up authentication, onboarding, and identity management
6. Build and Run the App
Once the code generation is complete:
Follow the build instructions created by your LLM.
Run the app on two separate devices/simulators/emulators.
When prompted, activate or sign into the app using your two starter-pack Atsigns, one for each device. If you need to access them, log into your Atsign Dashboard.
Test sending messages between the two Atsigns.
This validates:
Atsign onboarding
Secure key management
Encrypted messaging
7. Create your own Blueprint
Once you’ve explored the example Blueprint, you’re ready to create your own. Start Simple. A Blueprint doesn’t need to be perfect on the first pass. Its purpose is to help you think clearly about how your application works and to get you to working code quickly and securely.
When designing your Blueprint, focus on three core questions:
What are the nodes?
What does each node do?
How does information move between them?
These three decisions form the foundation of your application’s architecture and guide the LLM as it builds and refines your app.
Support and Further Help
If you run into issues, have questions about any step, or want to go deeper into building with the atPlatform, the Atsign team can help. Contact [email protected].
Last updated