Epic: Multi-client support and server-side multi-user deployment

[gaoteng-git]

Overview

This issue tracks the expansion of EverMemOS from a single-user local tool into a
platform that supports multiple AI coding clients and centralized server-side deployment.

Sub-goals

1. OpenCode integration

Support OpenCode as an additional AI coding client, alongside Claude Code.
EverMemOS hooks and skills are registered with OpenCode so that memory capture
and retrieval work transparently during OpenCode sessions.

2. OpenClaw integration

Support OpenClaw as an additional AI coding client.
EverMemOS is registered as an OpenClaw plugin, enabling the same memory
capabilities within the OpenClaw gateway environment.

3. Server-side multi-user deployment

Enable EverMemOS to run as a shared server serving multiple users, each with
isolated memory and their own 0G KV stream. Three deployment modes are supported:

• Mode A — Local single-user (existing behavior, unchanged)
• Mode B — Server multi-user: SERVER_MODE=true, API key authentication,
  per-user data isolation across all storage backends
• Mode C — Remote client: user machine connects to a Mode B server with no
  local service required; one-command setup registers the user and configures
  the local AI client automatically

Motivation

The three goals share the same underlying memory engine and installer.
Keeping them in a single repo and a single release ensures all clients and
deployment modes stay in sync without cross-repo version drift.