Opyt vs Supermemory

Supermemory carries one memory between your AI clients. The question is what put the documents in it.

This is the closest product on the site, so start with the overlap. Supermemory Personal is free, MIT-licensed and an MCP server, the same shape as Opyt. Supermemory Local self-hosts with one command and runs fully offline against Ollama. Retrieval is hybrid vector and keyword search. And it reaches clients Opyt does not: npx supermemory connect puts one memory in ChatGPT, Claude, Cursor, Windsurf, Perplexity, Gemini and Copilot, where Opyt is a local stdio server you install per client.

The two differ in one place, and it is not retrieval. It is what fills the store. The published connector list is Google Drive, Gmail, Notion, OneDrive, GitHub, Granola and a Web Crawler (supermemory.ai/docs/connectors, checked 2026-08-30). Every one of those is an account you already own or a folder you already keep, and none of them reads a social account. Everything else arrives because you sent it: a URL, a file, a note, a conversation. In fairness to a reader who remembers otherwise: the project began in 2024 as a bookmark app whose Chrome extension imported X bookmarks, and that path is not in today’s connector documentation.

The one rail that could reach the open web is the Web Crawler, and two facts bound it. Their docs describe it as crawling “with robots.txt compliance”, and x.com’s robots.txt answers Disallow: / to every agent except Googlebot, Bingbot and facebookexternalhit (fetched 2026-08-30). It is also a Scale-tier connector at $399/mo, so it is not in the personal product at all. Opyt takes the other route and reads the session your browser is already logged into, which is the only way to a bookmark list.

At a glance

Where Supermemory is better

• It is a finished consumer product. npx supermemory connect takes about thirty seconds, and the same memory then appears in ChatGPT, Claude, Cursor, Windsurf, Perplexity, Gemini and Copilot. Opyt needs Python 3.10+, a clone, a virtualenv and an API key, and reaches only the MCP clients you install it into.
• Conversation memory and user profiles: it “builds deep user profiles from behavior — intent, preferences, and context”. Opyt does not do this at all and should not claim to.
• Six connectors Opyt has none of: Google Drive, Gmail, Notion, OneDrive, GitHub and Granola, over OAuth with webhooks and a four-hour scheduled sync. Your own work documents are a rail Opyt does not have and is not building.
• Published results on public benchmarks: first on LongMemEval, LoCoMo and ConvoMem, 95% Recall@15 at 99.4% context reduction, retrieval under 300 ms. Opyt has published no public-benchmark number.
• Supermemory Local installs with one command and runs fully offline against Ollama, MIT-licensed. Opyt is MIT and local too, but the install is longer and its embedder is a hosted call.

Where Opyt is better

• It arrives with an acquisition rail. The first import was 691 bookmarks, 99 liked authors, 492 follows and 19 subscriptions, every one of them private to the account and none of them reachable by a connector on Supermemory’s list.
• X is out of reach on both of its paths. A plain fetch of an x.com post returns HTTP 402 (checked 2026-08-29), and the crawler that would otherwise try obeys a robots.txt that disallows every agent but three. On a personal plan the crawler is not available at all; it is a Scale connector at $399/mo.
• The unit of ingestion is a person. One add_oracle confirmation pulls 3,412 posts, 38 repos and 96 essays in a single action. Sending a URL stores one document, and a reading habit is thousands of documents wide.
• Provenance on every hit. Each atom carries a live source URL and the author you confirmed, and 1,551 of 2,804 saves came from people already confirmed (2026-08-23). A memory layer’s own description of what it holds is a profile of you.
• It keeps pulling with no prompt from you. frontier re-asks standing queries and stages what is new, deliberately re-reading a six-hour overlap behind its cursor because dedup absorbs it. A four-hour connector sync re-reads accounts you already connected; nothing on that list finds a source you have not.

Which one

Nothing here is exclusive. Both are MCP servers, and Supermemory for what you told your assistant alongside Opyt for what the people you trust published is a coherent setup. The line between them is the connector list: everything on Supermemory’s is an account you already own, and Opyt exists for the sources no connector will ever hold.

install Opyt free, MIT, five minutes