# Universal Memory Tool - Windows Alpha Setup

Version: v1.1.0-alpha
Build status: unsigned alpha
Last updated: 2026-05-26

## What is in this zip

- UniversalMemoryAppInstaller.exe
- UniversalMemoryAppInstaller-1.bin
- SETUP.md
- PACKAGE-SHA256SUMS.txt
- public-release-notes.txt
- public_verify_key.txt
- README_WRAP.md
- SKILL.md
- release-manifest.sha256.txt
- runtime-integrity.sha256

## Before you install

1. Extract the whole zip into one folder.
2. Keep `UniversalMemoryAppInstaller.exe` and `UniversalMemoryAppInstaller-1.bin` in the same folder.
3. This alpha build is not Authenticode-signed. Windows may show Unknown Publisher or SmartScreen warnings.
4. Optional: verify the installer hash:

```powershell
Get-FileHash .\UniversalMemoryAppInstaller.exe -Algorithm SHA256
Get-FileHash .\UniversalMemoryAppInstaller-1.bin -Algorithm SHA256
```

Expected hashes are listed in `PACKAGE-SHA256SUMS.txt`.

## Install

1. Run `UniversalMemoryAppInstaller.exe`.
2. If Windows warns about an unsigned publisher, choose the manual allow/run path only if you trust this alpha package.
3. Install into the default folder unless you have a reason to change it.
4. Start Universal Memory Tool from the Start Menu or desktop shortcut.

## First launch

1. Open the app.
2. Open Dashboard and use the VPS/license connection button.
3. For a new trial, connect to VPS first so the app receives the local bootstrap/license state.
4. If you paid already, use the claim-key flow if shown, then connect to VPS again so paid state is synced.
5. Memory data stays local. The VPS is for license/auth/payment checks only.

## Codex Desktop setup (recommended baseline)

1. In Universal Memory Tool, open `MCP Connect`.
2. Click `Copy Codex Form Fields`.
3. Open Codex settings -> MCP Server -> add custom server.
4. Use name: `universal-memory`.
5. Paste the command and environment fields copied from the app.
6. Save, restart Codex, and start a new chat.
7. Use GPT-5.5 in Codex if available, or another Codex model if not.
8. In Codex, ask it to call `policy` and `skill_md` first before using the memory tools.

Recommended first Codex prompt:

```text
Use the Universal Memory MCP. First call policy and skill_md. Then search/latest/timeline before broad search, hydrate raw only when proof is needed, and write summaries only after raw evidence is stored.
```

## AnythingLLM / local assistant setup

Use Universal Memory Tool as a local memory backend when your AnythingLLM or local assistant setup can call local tools or MCP-style commands.

Recommended local command:

```text
C:\Program Files\UniversalMemoryApp\universal_memory_mcp_remote_server.exe
```

Use the environment fields shown in the app's MCP Connect page. Keep the memory root on the same machine for private-local mode.

## ChatGPT web / app connector note

Universal Memory Tool is private-local by default. ChatGPT web app/connector modes are separate from Codex local STDIO MCP and can involve remote connector infrastructure.

Use ChatGPT app/connector mode only if your workspace supports it and you understand that a relay/remote setup changes the privacy model. For private local memory, Codex Desktop with GPT-5.5 or local assistants are the safer baseline.

## Correct LLM workflow

The tool is storage and retrieval infrastructure. The assistant should still reason carefully.

1. Call `policy` and `skill_md` first.
2. Ingest full raw chats/files as compressed evidence first.
3. Write ordered stage summaries after raw evidence is searchable.
4. Write one final summary with decisions, mistakes, improvements, and next actions.
5. Use latest/timeline records for current state before broad semantic search.
6. Hydrate raw evidence only when exact proof, code, commands, or conflict resolution is needed.

## Support

Email: deepvalueagent@gmail.com