GitHubJoin the waitlist

Install and start AtMem AI Architect

AtMem AI Architect is a local, browser-based AI stack design workspace. Install it from source. There is no pip package installation step and no hosted agent service on atmem.ai. The website provides information and training; the application and its model connections run on your machine.

1. Check prerequisites

Install Git, Node.js 22.12 or later, and Python 3.11 or later. Current local verification uses Node 26 and Python 3.14. A provider account or API key is optional: Local patterns works without either.

node --version
python3 --version
git --version

2. Get the source

git clone https://github.com/aetna000/atmem-ai-architect.git
cd atmem-ai-architect
npm ci --ignore-scripts

Review the repository's README, notices and current limits. Keep application credentials out of this checkout and out of project files.

3. Install the local services

On macOS or Linux:

python3 -m venv .venv
.venv/bin/python -m pip install -r server/requirements.txt
npm run setup:memory

The architect service uses .venv. AtMem uses the separate .memory-venv because its cryptographic dependencies differ from the model SDK environment. Setup downloads dependencies; it does not sign in to a provider or provision cloud resources. AtMem AI Architect currently pins its own tested AtMem runtime; do not replace its memory requirements merely because a newer AtMem package exists.

On Windows PowerShell, the corresponding source setup is:

python -m venv .venv
.\.venv\Scripts\python.exe -m pip install -r server/requirements.txt
python scripts/setup-memory.py

The launcher selects the Windows virtualenv executable. Native Windows execution has not been browser-qualified; these commands are not a claim of completed Windows acceptance. Report problems with OS and tool versions.

4. Start the application

npm run local

Keep this terminal open and visit http://127.0.0.1:5174. The local agent service listens on http://127.0.0.1:8766. Use the numeric loopback address. Stop the launcher with Ctrl+C when finished. This does not delete saved project files or AtMem AI Architect's memory store.

For a frontend-only local preview, npm run dev runs the zero-key interface. It does not start the Python service needed for connected model requests. A static website host cannot supply that local service.

5. Choose your architect

Open Connections. Start with Local patterns for a zero-key exploration. For a model-connected interview, choose an official account client or a supported model API connection. Install and sign in to only the desired native client, then select an accessible model or its default setting.

The existing README documents the Claude, Codex, Copilot and Gemini client sign-in commands. AtMem AI Architect does not copy their account tokens. Account login, subscription entitlement and individual model availability are separate checks. API-key billing is separate from subscription usage. Enter keys through the local Connections dialog or the documented service environment, never in a project, diagram, chat example or source file.

6. Run your first project

Continue with the hands-on training guide. Describe an idea, review the diagram and requirements, compare clouds, estimate from explicit assumptions, then inspect and download a delivery pack. The application does not apply cloud changes.

Troubleshooting

  • Missing local sidecar: check that .venv exists and the requirements installation succeeded. Run the commands from the repository directory.
  • Browser cannot connect: leave npm run local running and inspect its terminal for a port conflict or service error. Do not expose the service publicly to fix a local connection problem.
  • Provider/model unavailable: check that the chosen official client is installed and signed in, and that the model is available to that account. Local patterns remains an alternative; it is not a live model session.
  • Memory unavailable: rerun the documented memory setup and inspect the error. The architect can still use the current project; do not delete encryption keys or overwrite another AtMem store as a repair.
  • Terraform checks unavailable: Terraform is optional. A generated draft, provider schema check, real cloud plan and deployment are different stages.

AtMem AI Architect stores new app-owned memory under ~/.atmem-ai-architect/memory (existing installations retain their previous directory) (or the configured ATMEM_ARCHITECT_HOME), separate from ~/.atmem. Save an editable project file before upgrading or starting over. Preserve your files, memory store and keys. For backend hosting boundaries, read self-hosting.