# MemFree: hybrid AI search over personal knowledge bases and the web, with a UI generator > MemFree is an open-source tool that combines a hybrid AI search engine (querying both a personal vector knowledge base and the web) with an AI page generator that produces React/Tailwind/Shadcn UI code. Self-hosting requires a Zeabur backend and Vercel frontend with Upstash Redis and an OpenAI API key. **memfreeme/memfree** — MemFree - Hybrid AI Search Engine & AI Page Generator - Repository: https://github.com/memfreeme/memfree - Website: https://www.memfree.me?ref=github.com - Stars: 1,514 · Forks: 213 - Language: TypeScript - License: MIT - Published: 2026-10-11 · Updated: 2026-10-11 · Language: en - Canonical page: https://hysenlabs.com/projects/memfreeme-memfree ## Two products in one: a hybrid search engine and a UI page generator sharing a single deployment MemFree describes itself as 2 distinct products running from one codebase. One is a hybrid AI search engine that queries both a personal vector knowledge base and the public web, combining results through an AI model to produce a direct answer. An AI page generator is the second, converting text descriptions or reference images into React, Tailwind, and Shadcn UI code with a real-time preview. That dual identity creates an adoption trade-off that a quick feature scan does not reveal. Teams that need knowledge base search with web augmentation and teams that need AI-assisted UI generation have different infrastructure requirements. Running both together means accepting the full cost of both, and no path is described in the README for running only one side without deploying the other. ## Hybrid search: vector store on Upstash, web sources through Google and Exa, AI model selection MemFree draws on 2 retrieval sources in each search query. One is a personal knowledge base built from indexed Chrome bookmarks, uploaded files, and web pages, stored as vectors in Upstash Redis. Web search through Google and Exa is the other, giving access to public content not in the personal index. Both feeds pass through an AI model (ChatGPT, Claude, or Gemini, per the README) to produce the final answer. Repository layout reflects this architecture: frontend/ holds the UI, vector/ handles vector storage and search, mdreader/ processes markdown and document inputs, and extention/ is the Chrome extension. A workflow diagram (memfree-hybrid-ai-search.webp in the frontend/public/ directory) shows the full pipeline visually. Tech stack details live on an external blog post at memfree.me/blog/hybrid-ai-search-tech-stack rather than in the README. ## Deploying on Zeabur and Vercel: the 6 required environment variables for the frontend Deployment follows a 2-step sequence: the backend goes on Zeabur first using a provided template, then the frontend (in the frontend/ subdirectory) deploys on Vercel. Netlify is listed as an alternative but the README truncates those instructions. Vercel deployment requires 6 environment variables: ```bash UPSTASH_REDIS_REST_URL UPSTASH_REDIS_REST_TOKEN OPENAI_API_KEY MEMFREE_HOST AUTH_SECRET API_TOKEN ``` Full descriptions for each variable live in the env-example file at github.com/memfreeme/memfree/blob/main/frontend/env-example, which the README links to rather than reproducing. A dedicated one-click deployment guide at memfree.me/docs/one-click-deploy-ai-search covers the full process in more detail. ## Feeding the knowledge base: Chrome bookmarks, 5 local file formats, and multi-image search Knowledge base content comes in through several paths. A Chrome extension (Web Store ID dndjodcanbhkomcgihbhcejogiimdmpk) handles one-click bookmark synchronization and indexing. Manual uploads accept text, PDF, Docx, PPTX, and Markdown files. Web page inputs add specific URLs to the index directly. Search input supports text, images, files, and web pages. Multi-image input is specifically documented: a single query can span multiple images for comparison, summarization, or analysis together. Results appear in 4 presentation modes: text, mind maps, images, and videos. Cross-device sync keeps the search history consistent, and the interface supports 7 languages: English, Chinese, German, French, Spanish, Japanese, and Arabic. ## The AI page generator: Claude 3.5 Sonnet produces React and Tailwind code from text or image prompts Claude 3.5 Sonnet is named in the README as the AI model for page generation, alongside React, Tailwind, and Shadcn UI as the output stack. A text prompt or reference image goes in, the generator produces runnable code, and the UI renders in real time alongside the editor. Code editing uses a VSCode-like interface with syntax highlighting and auto-completion for manual adjustments after generation. File-to-page generation converts uploaded documents into structured web pages. Image-driven generation produces code that matches a reference image's visual layout. Publishing sends the page to a public URL with a single click. Small errors in the generated code are expected: the README acknowledges that occasional errors may occur and describes a Smart Error Correction feature that fixes issues with one click, though the categories of errors it handles are not documented. ## OPENAI_API_KEY is a required deployment variable while Claude and Gemini appear only in the feature list Among the 6 required Vercel environment variables, OPENAI_API_KEY appears explicitly while no Anthropic or Google API key is listed. In the feature list, ChatGPT, Claude, and Gemini are named as equal AI model options. Required deployment variables tell a different story: OpenAI is a hard dependency at installation time, while Claude and Gemini are not represented in those named env vars. Where Claude or Gemini API keys live (whether in env-example or configured elsewhere) is not described in the README. That gap between the multi-model feature list and the single-provider required key is the most practically useful detail for anyone planning a self-hosted deployment. Upstash Redis credentials are also a hard dependency, meaning 2 external services require accounts before any search query can run. ## Self-hosting versus the hosted product at memfree.me MemFree runs as a hosted product at memfree.me in addition to the self-hosted path. Using the hosted version requires no deployment. Self-hosting gives operational control and the ability to modify the code, at the cost of managing the Zeabur backend, Vercel frontend, Upstash Redis instance, and the necessary API keys. For teams aiming to replace multiple AI subscriptions, the README specifically names ChatGPT Plus, Claude Pro, and Gemini Advanced as monthly costs MemFree targets. Self-hosting trades those subscription costs for infrastructure and setup work. Whether that trade is favorable depends on usage patterns the README does not document. The last push to the repository was on 2026-07-06, and MemFree is distributed under the MIT licence. ## Conclusion Use MemFree when you want personal knowledge base search combined with web AI search in one self-hosted tool, and you are comfortable managing Upstash Redis and API key dependencies. Skip it if you need only one of its 2 functions, since running both adds infrastructure cost with no benefit for single-purpose use. Before deploying, read the env-example file at github.com/memfreeme/memfree/blob/main/frontend/env-example, because the Vercel deployment needs 6 environment variables and their full descriptions live there rather than in the README. ## FAQ ### What is MemFree? MemFree is an open-source hybrid AI search engine that queries both a personal knowledge base and the web, and an AI page generator that produces React, Tailwind, and Shadcn UI code. It is available as a hosted service at memfree.me and as a self-hosted deployment. ### How do I self-host MemFree? Deploy the backend on Zeabur using the provided template, then deploy the frontend (in the frontend/ directory) on Vercel. Vercel deployment requires 6 environment variables including OPENAI_API_KEY and Upstash Redis credentials. Full variable descriptions are in the env-example file, and a deployment guide is at memfree.me/docs/one-click-deploy-ai-search. ### What file formats can MemFree index? MemFree accepts text, PDF, Docx, PPTX, and Markdown files as local uploads. It also indexes web pages directly and syncs Chrome bookmarks through the MemFree Chrome extension. ### What AI models does MemFree support? The README lists ChatGPT, Claude, and Gemini as supported AI models for search. For page generation, Claude 3.5 Sonnet is the named model. Only OPENAI_API_KEY appears in the required Vercel deployment variables, so whether other providers need separate keys is explained in the env-example file. ### Is MemFree open source? Yes. MemFree is published at github.com/memfreeme/memfree with a MIT licence. The README includes translations in German, French, Spanish, Japanese, and Chinese alongside the English version. ## Sources - [Issues](https://github.com/memfreeme/memfree/issues) - [License: MIT](https://github.com/memfreeme/memfree/blob/main/LICENSE) - [memfreeme/memfree on GitHub](https://github.com/memfreeme/memfree) - [Project website](https://www.memfree.me?ref=github.com) - [README](https://github.com/memfreeme/memfree/blob/main/README.md) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/memfreeme-memfree