TMCRARequest access

Codex desktop memory / 01

Let Codex continue the project, not restart the context.

The TMCRA Memory app installs the Codex integration and lets you sign in and connect from one place. After you submit a prompt and before Codex answers, it recalls and injects relevant memory. When the answer finishes, it captures both your message and the final assistant response as separate actor records. Both remain recallable: user records carry requirements and facts, while Codex records carry prior progress and results. Codex output is never promoted to a user statement. New projects start automatically; existing projects can import retained Codex history instead of beginning from zero.

Download for Windows View setup steps Open memory console

02 / WINDOWS SETUP

Install the app, then connect Codex inside it.

The standard Windows path does not require SSH, server access, or copying an API Key. You still sign in yourself, approve the Codex connection, and decide whether to trust each Hook.

  1. 01

    Download and install the Windows app

    Download the setup file, double-click it, and follow the installation prompts. There is no ZIP to extract on the standard path.

    Download TMCRA Memory for Windows · SHA-256 checksum · Release manifest

    This preview installer is not yet code-signed. Windows SmartScreen may show a warning; verify the SHA-256 before running it.

  2. 02

    Sign in and connect Codex in the app

    Open TMCRA Memory, sign in to your account, and choose Connect Codex. Review the authorization shown in the app and approve it to finish the connection. The scoped Token is written directly to protected local configuration; you do not copy it by hand.

  3. 03

    Restart Codex and confirm three Hooks

    Restart Codex and confirm TMCRA Memory is enabled in the Plugins page. Enter /hooks in a Codex task, review all three TMCRA lifecycle Hooks, and trust them once. Codex keeps this decision in your hands; neither the installer nor the app can silently approve Hooks.

    /hooks
  4. 04

    Preview and import retained Codex history

    The History migration panel is visible before sign-in. Select one project for a read-only local preview; no history is uploaded during preview. After connecting your account, confirm the import separately. Reasoning, tool logs, developer instructions, passwords, keys, verification codes and credential-like messages are excluded.

  5. 05

    View quota and the memory graph

    Use the app or personal console to inspect memory Scope and Session structure, view the memory graph, check server-reported usage and quota, or revoke one Codex connection without affecting the others.

03 / ADVANCED MANUAL SETUP

For CLI-first workflows and non-Windows systems.

The ZIP package remains available for developers who want to inspect or control each step. It is the secondary path, not the default Windows experience.

  1. 01

    Download and extract the package

    Extract it to a stable local directory. Do not leave the integration inside a temporary download folder that the system may clean up.

    Download the manual Codex package · SHA-256 checksum · Release manifest

  2. 02

    Run the Windows PowerShell installer

    .\Install-TMCRA.ps1

    The script registers the Codex plugin, checks the local setup, and starts TMCRA account authorization.

  3. 03

    Run the macOS or Linux installer

    sh ./install.sh

    After installation, follow the authorization prompt, then restart Codex and review the same three Hook requests.

  4. 04

    Keep the authorization boundary intact

    Manual setup still does not require server login, SSH access, or copying an API Key. Sign in and approve the short code yourself; a Token limited by Scope is delivered to protected local configuration.

04 / MEMORY BOUNDARIES

Global facts cross projects. Project work does not.

TMCRA keeps one user-global memory layer for stable preferences and identity. Every project receives a separate scope, while each Codex task remains a session inside that project. This preserves cross-session continuity without combining ten projects into one graph.

memory.boundaries
user
├── global
└── project_scope
    ├── session_a
    ├── session_b
    └── session_c

05 / EXISTING PROJECTS

Import what Codex still retains.

Historical import is never automatic. Preview first, choose one project, then confirm the upload. Only user and assistant messages are imported; reasoning, developer instructions, tool logs and credential-like messages are excluded.

Preview only
node .\plugins\tmcra-memory\scripts\history_import.mjs preview
Import one project
node .\plugins\tmcra-memory\scripts\history_import.mjs import --project "D:\work\my-project" --confirm --wait
Preview a repository baseline
node .\plugins\tmcra-memory\scripts\project_bootstrap.mjs preview --project "D:\work\my-project"
Confirm the repository baseline
node .\plugins\tmcra-memory\scripts\project_bootstrap.mjs import --project "D:\work\my-project" --confirm

If the old transcript no longer exists locally, TMCRA cannot reconstruct it. The repository baseline uses selected project documents and recent Git history, labels them as current repository evidence, and still requires explicit confirmation before upload.

06 / GET STARTED

Install TMCRA Memory and connect Codex from the app.

After you sign in and approve the connection, restart Codex and confirm the three Hook requests. The app and personal console then give you one place to manage connections, usage, quota and memory boundaries.

Download for Windows