Discover and design
Build the architecture together.
Capture requirements, sketch components and connections, and compare cloud options. Review proposed changes, keep a decision history and revisit earlier stages.
AtMem AI Architect · Ready to run locally
Design your AI stack.
Keep the decisions.
AtMem AI Architect is an AI architecture platform built around modern agent memory. Describe what you want to build. Work with an architect on an editable canvas, review the trade-offs, and export a delivery pack from the same project.
AtMem AI Architect runs locally in your browser with a local agent service. Connect a supported model provider to guide the design conversation.
Hosted platform coming soon. The local application is available now.
Get AtMem AI Architect on GitHub ↗Install and learn AtMem AI Architect.
Follow the local installation guide, work through a first architecture, or read the complete project README. These guides are imported from a reviewed source revision.
Review and estimate
Make assumptions visible.
Discuss security, reliability, availability, scale, performance and maintainability. Inspect usage assumptions, price sources, missing rates and cost drivers. Estimates are not bills or guarantees.
Export and revisit
Keep one editable project.
Review outputs in the application before downloading: specification PDF, diagram PDF/SVG/PNG, Terraform draft and an implementation Spec Kit with task dependencies, acceptance checks and coding-agent instructions. Save the project file, reopen it later and regenerate outputs after changes.
Modern agent memory beneath the workflow.
AtMem AI Architect uses its own AtMem store for explicit reusable preferences, accepted project decisions and selected requirements. Inspect current and earlier values, preview forgetting, or turn memory off for a project.
AtMem AI Architect keeps its own interface. Your existing AtMem home is separate.
Explore AtMem memory and governance →Choose how the architect runs.
Use official Claude, Codex, Copilot or Gemini account clients, or supported API-key connections. Provider access, model availability and billing depend on your chosen connection.
Account clients retain their authentication. A static website cannot run the local agent service.
Current scope
This is a local design and delivery-planning application. It does not provision cloud resources or provide a hosted multi-user agent service. Cloud coverage varies; missing prices and engineering gaps stay visible. Windows execution has not been browser-qualified.
The application remains a separate local installation. This page describes the project; it does not upload your diagrams or launch local account clients.
Based on the documented local milestone ↗. Running and hosting boundaries ↗.