The Problem
Developers building AI applications need persistent memory across sessions, but wiring up storage, retrieval, and context injection themselves is brittle and time-consuming. Supermemory solves this by offering a hosted memory API that ingests content from URLs, files, and connectors, then serves it back to AI tools via search, chat, and MCP integrations.
What This Does
This repo is a fork of supermemoryai/supermemory (28.9k stars) and contains the browser extension and documentation site. The extension (apps/browser-extension/) is a WXT-based Chrome extension with content scripts for site-specific extraction (entrypoints/content/chatgpt.ts, claude.ts, twitter.ts) and a React popup (entrypoints/popup/App.tsx). Utilities in utils/ handle API calls (api.ts), local storage (storage.ts), and query hooks (query-hooks.ts).
The apps/docs/ directory holds the full product documentation (140+ MDX files) covering the Memory API, connectors (Notion, Google Drive, S3), SDKs (TypeScript, Python, AI SDK), and migration guides from Mem0 and Zep. The root README describes the broader product: a web app, chat interface, MCP server, and Raycast extension — though those components aren't in this fork's file tree.
How To Use It
Setup: The extension has a package.json and tsconfig.json but no lockfile, so dependency versions aren't pinned. Install with npm or pnpm:
cd apps/browser-extension npm install
Configuration: Copy apps/browser-extension/.env.example to .env and populate API keys. The extension calls the Supermemory backend via utils/api.ts; the .env.example likely holds the API base URL and auth token.
Running it: This is a WXT project (wxt.config.ts). Use the WXT CLI to build or dev:
npx wxt build # production build npx wxt dev # watch mode
The entry points are entrypoints/background.ts and the popup React app. The docs site (apps/docs/) is a Mintlify project (per docs.json) and would run with mintlify dev if you have the CLI installed.
Real-World Use
A developer building a customer-support chatbot can use the Memory API to ingest support tickets and knowledge-base articles, then query them at runtime. The docs show an AI SDK integration (apps/docs/cookbook/ai-sdk-integration.mdx) that wires memory tools directly into an agent loop — the agent searches memories before answering, grounding responses in the ingested content.
Code Health & Issues
Med - No test files - repository-wide. The extension has zero tests despite non-trivial logic in content scripts and utilities. This is risky for a browser extension that must handle multiple DOM structures. Low - No lockfile - apps/browser-extension/package.json declares dependencies but no package-lock.json or pnpm-lock.yaml exists. Builds aren't reproducible. Low - CI is thin - .github/workflows/ has three workflows (Claude code review, AI SDK publish) but no test or lint job for the extension. Med - Site-specific content scripts - chatgpt.ts, claude.ts, twitter.ts hard-code selectors for third-party sites. These break when those sites change their DOM; the maintenance burden is real and untested.
The Bottom Line
This fork is a focused slice of the Supermemory ecosystem: a browser extension plus docs. The extension is functional but lacks tests and a lockfile, which makes it a maintenance risk. The documentation is the real asset here — it's thorough and covers the API surface well. Teams evaluating Supermemory's memory API should read the docs; teams wanting to contribute to the extension should expect to add test coverage and pin dependencies first.