MyMem

Your social history, made usable by agents. → mymem.space

MyMem turns the social media exports you already own into one searchable memory. It gives your agent the context to understand you better, gives you a visual timeline across platforms, and gives you a copy of your socials when a platform blocks your account or its search fails you.

The platforms are good at storing your history. They are bad at helping you search your own posts. MyMem puts that history in one place and lets you ask for the answer you remember:

• “What have I said about facial recognition?”
• “Which restaurants have I eaten at in Mumbai?”
• “Where did I buy a Thanka painting in Kathmandu”

Results come back from your archive with dates and links, newest first.

Why it exists

Search your own history properly

Platform search is built for finding public content, not for finding the thing you said three years ago. MyMem searches across your X, LinkedIn, Instagram and Facebook exports in one place. It combines SQLite FTS5 keyword search with optional local embeddings, so it can find both the words you used and the meaning behind a query.

Keep a backup that platforms cannot take away

Your archive is a copy of your own social history. If a government blocks your account, a platform locks you out, or a service changes its search, your exported posts, replies, likes, saves, follows and connections remain available in MyMem.

Give your agent better context

An agent can query the deployed archive over MCP and answer from what you actually posted. It can find your past positions, saved items, people and timelines instead of starting from a blank context every time.

See the shape of your social life

The web archive gives you a visual timeline of your social media across platforms. Posts, replies, likes, saves and other public activity stay connected to their platform, date and link.

How it works

1. Request your data from X, LinkedIn, Instagram or Facebook.
2. Open mymem.space and upload the export from the archive page. Large exports upload in chunks, so the page can handle them without one giant request.
3. MyMem parses each platform's own export, filters the public archive, and makes it searchable.
4. Browse the timeline yourself or connect an agent to https://mymem.space/mcp.

The adapters read exports from each platform. They do not depend on live platform APIs or OAuth connections.

What reaches the public archive

Only public content is imported into the public archive. Direct messages and private profile information never reach it. Public likes, saves, follows and connections are kept because they are part of the public social record represented by the exports.

Reposts and retweets are never attributed to you as something you said. They are credited to the original author. Quote-tweets keep your words as your own commentary; the quoted post is reduced to a link.

The full archive stays on your machine. The server holds only the published copy that agents and the web viewer can read.

Connect an agent

MyMem is deployed and working at mymem.space. Agents use MCP over HTTP at:

https://mymem.space/mcp

The endpoint requires a bearer token. It is read-only and serves the published copy. A Hermes agent on a Raspberry Pi uses this endpoint today:

mcp_servers:
  mymem:
    url: "https://mymem.space/mcp"
    headers:
      Authorization: "Bearer YOUR_MyMem_TOKEN"

Replace YOUR_MyMem_TOKEN with the token supplied to you, then reload Hermes. Claude Code and other MCP clients can use the same HTTP endpoint. See docs/REMOTE.md for the endpoint contract and local-development path.

Privacy boundary

MyMem keeps the working archive and the published archive separate. The published copy excludes direct messages and raw export records. Only that filtered copy is served by the web viewer and MCP endpoint. Deployment details live in docs/DEPLOY.md.

Run the engine locally

The deployed web path is the normal way to use MyMem. Local stdio exists for development and for working directly with an archive on the machine that holds it.

git clone https://github.com/nixxintools/mymem
cd mymem
npm install
npm test
npm run build

The local engine can import an export, search it, build a local timeline, and run the MCP server over stdio:

npm run mymem -- import path/to/export.zip
npm run mymem -- search "facial recognition"
npm run mymem -- stats
npm run mcp

Use docs/VERIFY.md for the contributor verification walkthrough. Use docs/REMOTE.md for remote MCP details and docs/DEPLOY.md for operating the deployed service.

Open source

MyMem is open source under the MIT license. The engine, adapters and web application are in this repository: github.com/nixxintools/mymem.