# Memorall — Unikorn submission profile

## Product Name

Memorall

## Positioning statement (tagline)

Memorall turns your browser into a workspace for an AI agent: with long-term memory, real tools, and you choose the model — running directly on your machine by default.

## Product Description (paste into the description box)

**What is the product**

Memorall is a *browser-native agent workspace* — an extension for Chrome/Edge (also having a web version and a desktop version that share the same codebase) that transforms the browser into a fully equipped workspace for an AI agent: long-term memory, a library of documents, operational tools (browser, files, code running sandbox) and the freedom to choose a model. All default data resides on the user's machine (local-first), no login or cloud upload is required.

**Problem Solved**

Most knowledge work takes place in browser tabs, but current AI tools are separate from that. You read documents, open dozens of tabs, save PDFs, take notes — yet still have to manually copy–paste each section into a chatbot, and each conversation starts from scratch because the assistant remembers nothing about your project. Additionally, there are privacy risks: internal documents, contracts, customer data being sent to third-party servers.

Memorall addresses both problems. What you read and actively save — web pages, highlighted text, screenshots, PDFs, Excel sheets, notes — are converted into structured memory: a knowledge graph by topic, combining keyword search and semantic search via vectors. This allows the agent to understand the project context, terminology, and your long-term research trajectories, rather than just answering on a disjointed prompt. And because everything runs locally, the data remains on your machine.

**Key Features**

1. **Assistant right in the page.** Open a chat panel on any website; attach highlighted text, current visible content, all text/HTML of the page, or screenshots as context; save content to a topic with just a few clicks.

2. **Memory and knowledge graph.** Convert text, Markdown, PDFs, Excel into nodes and relationships in a knowledge graph by topic; browse a visual graph, conduct semantic searches, edit and clean data right in the interface.

3. **Agent with real tools, not just text responses.** The agent can read/write from the document library and workspace files, control the browser (open pages, read DOM, wait for selectors, operate), and run Node.js code in a sandbox right in the browser — install packages, run servers, prototype applications.

4. **Create your own agent.** Each agent is a graph defined by you: system prompt, skills, tools, connect to MCP (HTTP/SSE), schedule periodic runs; enable/disable specific capabilities (knowledge retrieval, web browsing, sandbox, file system, delegation) for each agent.

5. **Freedom to choose the model.** Run models directly in the browser using WebLLM, Wllama, or Transformers.js (supporting WebGPU acceleration when the device allows), or connect to OpenAI, OpenRouter, Ollama, LM Studio when stronger models are needed. Choose embedding dimensions of 384d / 768d / 1536d.

*Required permissions and reasons:* `activeTab` + content script (to read page content when you request), `storage` (to save configuration and local data), `tabs`/`webNavigation` (to open pages and track navigation for the agent), `contextMenus` (quickly save highlighted content), `notifications` (notify when background jobs finish), `offscreen` (run models and heavy tasks in the background so the interface doesn't lag).

**Who Should Use It**

Memorall is aimed at those who work heavily in the browser and need accumulated memory, specifically: researchers, PhD candidates, and analysts who must read and synthesize dozens of papers/reports on the same topic; programmers and AI engineers wanting an agent with sandbox and file access for quick experimentation, or those wanting to build their own agents instead of using a closed chatbot; content writers, editors, and SEO specialists needing to gather material by topic and trace sources; lawyers, accountants, HR, and teams handling sensitive documents that cannot push data to the cloud; and privacy-focused users wanting an AI assistant that runs completely offline. Memorall is not suitable for those who only need to quickly ask a question — its value lies in the gradually accumulated memory over time.

**How to Get Started**

- **Chrome / Edge (main platform):** install from the Chrome Web Store — https://chromewebstore.google.com/detail/memorall/kcienfpnencfkniagcjlpaehofblfhab — pin the icon to the toolbar, open Memorall, choose a model (download a model to run locally, or paste your OpenAI/OpenRouter API key