XMem

Real-time memory recall for AI conversations.

XMem watches what you type in ChatGPT, Claude, Gemini, Perplexity, and DeepSeek — then surfaces relevant memories from your personal memory store before you hit send.

---

See it in action

Demo

demo.mp4

Setup

extension-settings-1773462173011.mp4

---

How it works

You type in an AI chat
        |
        v
XMem debounces your input (600ms)
        |
        v
Semantic search hits your XMem server
        |
        v
Relevant memories appear as ghost text + floating chip
        |
        v
Tab to accept  ·  Click chip to browse  ·  Memories inject into your prompt

---

Features

Live suggestions

As you type in any supported AI chat, XMem performs semantic search against your memory store and renders ghost text inline — press Tab to accept, Escape to dismiss. A floating chip displays the count of matching memories. Searches fire only after 8+ characters and only when the cursor is at the end of input.

Sidebar panel   Ctrl+Shift+M

Tab	What it does
Memories	Browse results with domain tags (profile, temporal, summary) and relevance scores
Ask Memory	Ask natural-language questions — get LLM-generated answers backed by your stored memories
Settings	Connection status, configured domains, version info

Auto-save

Your outgoing messages are captured and ingested automatically after the AI finishes responding. You can also right-click any selected text on any page and choose "Save to XMem Memory".

Slash commands

Command	Mode
/Xingest	Queue-based ingestion — saves conversations to memory without blocking the UI
/Xsearch	Auto-inject synthesized memory context when you send a message
/Xide	Auto-inject codebase context on send (IDE mode)
/Xrepo	Browse and query your codebase structure via a file tree panel

IDE / Code mode

Browse a codebase directory tree from the left panel. Query it semantically ("What does the payment service do?") and get streaming responses with file and symbol references extracted automatically.

---

Supported sites

Platform	Domain
ChatGPT	chatgpt.com chat.openai.com
Claude	claude.ai
Gemini	gemini.google.com
Perplexity	perplexity.ai
DeepSeek	chat.deepseek.com
Other	Any page with standard <textarea> or contenteditable fields

---

Getting started

Prerequisites

• Chrome browser
• A running XMem server (default: http://localhost:8000)
• Node.js 20+

Build

npm install
npm run build

For development with hot reload:

npm run dev

Install in Chrome

1. Open chrome://extensions/
2. Enable Developer mode (top-right toggle)
3. Click Load unpacked
4. Select the dist/ folder

Configure

1. Click the XMem icon in the Chrome toolbar
2. Enter your API URL (e.g. http://localhost:8000)
3. Enter your API Key (if required)
4. Set your User ID
5. Click Save Settings
6. Click Test to verify the connection — a green dot confirms you're live

---

Configuration reference

Setting	Storage key	Default	Description
API URL	xmem_api_url	http://localhost:8000	XMem backend endpoint
API Key	xmem_api_key	—	Authentication token
User ID	xmem_user_id	chrome-extension-user	Your user identifier
Memory Active	xmem_enabled	true	Master on/off switch
Live Suggestions	xmem_live_suggest	true	Ghost text inline suggestions
IDE Org ID	xmem_ide_org_id	—	Organization for code context
IDE Repo	xmem_ide_repo	—	Repository for code context
Effort Level	xmem_effort_level	low	Ingestion detail level (low / high)

---

Architecture

xmem-extension/
  src/
    api.ts            API client + request queue (wraps xmem-ai SDK)
    background.ts     Service worker — context menus, message routing
    content.ts        Core logic — editor detection, live search, sidebar, auto-save
    popup.html        Settings UI (dark mode, 360px popup)
    popup.ts          Settings persistence via chrome.storage.sync
  icons/              Extension icons (16, 48, 128 px)
  manifest.json       Chrome Manifest V3
  vite.config.ts      Build config (@crxjs/vite-plugin)

API endpoints used

Method	Endpoint	Purpose
POST	/v1/memory/ingest	Save messages and responses
POST	/v1/memory/search	Semantic search across memory domains
POST	/v1/memory/retrieve	LLM-synthesized answers from memory
GET	/health	Server connectivity check
GET	/ping	Health status with version
POST	/v1/code/query_stream	Stream code query results
GET	/v1/code/directory_tree	Codebase directory structure
GET	/v1/code/list_repos	Available repositories

Key internals

• Request queue — All API calls are serialized to prevent INVALID_CONCURRENT_GRAPH_UPDATE errors from the LangGraph backend
• Editor detection — A MutationObserver watches the DOM for editable fields using per-site CSS selectors
• React compatibility — Uses native property setters to update React-controlled textareas (ChatGPT)
• Theme-aware — Detects light/dark backgrounds to style ghost text appropriately
• Ingestion queue — Separate queue for batch saves so the UI never blocks

---

Built by Xortex