memory.lol

Overview

This project is a tiny web service that provides historical information about social media accounts.

It can currently be used to look up 542 million historical screen names for 443 million Twitter accounts. Most of this data has been scraped from either the Twitter Stream Grab or the Wayback Machine (both published by the Internet Archive).

Coverage should be fairly good (for non-protected accounts) going back to 2011, which is when the Twitter Stream Grab was launched.

Please note that this software is not "open source", but the source is available for use and modification by individuals, non-profit organizations, and worker-owned businesses (see the license section below for details).

Safety

All information provided by this service has been gathered from public archives, and in most cases it can easily be found through other means (such as a Twitter search for replies to an account). The goal of the service is to make it easier for researchers or journalists to identify directions for further investigation, and more generally to indicate to users that an account may be operating a scam, spreading disinformation, etc. If you have concerns about safety or privacy, you can contact me (via Twitter DM or email) and your request will be handled privately.

Current access restrictions

In most cases public access to the tool is currently limited to historical facts that have been observed in the past 60 days. There are two exceptions to this rule:

• Some accounts are excluded at the request of the account owner.
• Full histories are provided for a set of accounts compiled from several "bad actor" lists.

The full twelve years and half a billion screen names (minus requested exclusions) are available to a trusted group of researchers, journalists, and activists.

The service currently supports authenticating via a GitHub or Google account.

The service only uses GitHub (or Google) for authentication, doesn't require any non-public or write access to the user's accounts, will never request any kind of password, and only the user's public information is stored on the servers.

The service does not currently log requests in a way that would allow anyone with access to the server to link individual queries to specific authorized users, but I reserve the right to implement such logging in the future if there's any suggestion of abuse.

To log in visit https://api.memory.lol/v1/login/github, which will take you to a GitHub "Authorize memory.lol" page that will ask you to authorize "Limited access to your public data". Click "Authorize" and you'll be taken to a status URL that will show your current access level (which will be empty unless your account has been specifically approved for access). From there the full index will be available (if your account has been approved).

To log out go to https://api.memory.lol/v1/logout.

It's possible to use the full version of the service from the command-line via GitHub's device flow, but this currently isn't very convenient (see instructions below). I'll be providing a client that makes command-line use a little easier.

If you're interested in having your account approved for non-date-restricted access, please contact me.

Use cases

Accounts that engage in hate speech, scams, harassment, etc. on social media platforms sometimes try to obscure their identities by changing their screen names, and they often also have really bad opsec (for example using real names or other identifying information on accounts that they later intend to use anonymously).

Being able to look up historical social media profiles often makes it possible to identify the offline identities of these people (or at least to trace connections between their activities).

Here are a few examples off the top of my head (the first three are examples of the service in action, and the last two show how it can be used to confirm the work of others):

• @OSINT_Ukraine: gained a large following in February 2022; looking up old screen names shows that it had previously been an NFT scam account.
• @libsoftiktok: a viral hate account that targets LGBTQ+ people; looking up her screen name in this service is how I found her name (Chaya Raichik) a couple of months ago.
• @_lktk: an abusively transphobic troll named Iratxo Lorca who has been active in the Scala community for years; he was one of the first people I identified using this service.
• @Mormonger: a homophobic Mormon who was identified as a person named Cole Noorda last September; this service confirms that he had previously used the screen name @colenoorda for his account.
• @_14words_: an account that was identified as white supremacist Illinois cop Aaron P. Nichols earlier this year; this service connects this account to the screen name @spd584.

In many cases the information provided by the service won't be enough to identify a person, but may provide hints about where to look next (for example looking up deleted tweets for old screen names with ✨ cancel-culture ✨ is often a reasonable second step).

Detailed example

If you visit https://api.memory.lol/v1/tw/libsoftiktok in your browser, you'll see the following data:

{
  "accounts": [
    {
      "id": 1326229737551912960,
      "screen-names": {
        "chayaraichik": null,
        "cuomomustgo": null,
        "houseplantpotus": null,
        "shaya69830552": [
          "2020-11-10"
        ],
        "shaya_ray": [
          "2020-11-27",
          "2020-12-17"
        ],
        "libsoftiktok": [
          "2021-08-18",
          "2022-06-16"
        ]
      }
    }
  ]
}

Note that for some screen names we don't currently have information about when they were observed (e.g. the ones with null values above). If an screen name was observed on only one day in our data sets, there will be a single date. If there are two dates, they indicate the first and last day that the screen name was observed.

These date ranges will not generally represent the entire time that the screen name has been used (they just indicate when the account appears with that screen name in our data sets).

Other features

The service is very minimal. One of these few things it does support is querying multiple screen names via a comma-separated list (for example: https://api.memory.lol/v1/tw/jr_majewski,MayraFlores2022). It also supports searching for a screen name prefix (currently limited to 100 results; for example: https://api.memory.lol/v1/tw/tradwife*).

It currently only supports JSON output, but if you want a spreadsheet, for example, you can convert the JSON to CSV using a tool like gojq:

$ curl -s https://api.memory.lol/v1/tw/jr_majewski,MayraFlores2022 |
> gojq -r '.[].accounts | .[] | .id as $id | ."screen-names" | keys | [$id] + . | @csv'
89469296,"LaRepublicana86","MayraFlores2022","MayraNohemiF"
726873022603362304,"JRMajewski","jr_majewski"
1533878962455293953,"jr_majewski"

Or if you want one screen name per row:

$ curl -s https://api.memory.lol/v1/tw/jr_majewski,MayraFlores2022 |
> gojq -r '.[].accounts | .[] | .id as $id | ."screen-names" | keys | .[] | [$id, .] | @csv'
89469296,"LaRepublicana86"
89469296,"MayraFlores2022"
89469296,"MayraNohemiF"
726873022603362304,"JRMajewski"
726873022603362304,"jr_majewski"
1533878962455293953,"jr_majewski"

Note that screen name queries are case-insensitive, but the results distinguish case (which can be useful for archives such as Archive Today, which only provide case-sensitive search).

Other endpoints

You can also look up an account's history by account ID (e.g. https://api.memory.lol/v1/tw/id/1326229737551912960 also shows the screen names for Raichik's account).

Authorized access via device flow

There are currently several steps if you want to access the full index from the command line. By default you will receive date-restricted results:

$ curl -s https://api.memory.lol/v1/tw/USForcesKorea | jq
{
  "accounts": [
    {
      "id": 4749974413,
      "id_str": "4749974413",
      "screen_names": {
        "USForcesKorea": [
          "2018-06-08",
          "2022-07-29"
        ]
      }
    }
  ]
}

To access the full index (assuming you have an approved account), you'll first need to get a device code and user code, using exactly this command:

$ curl -X POST -d 'client_id=b8ab5a8c1a2745d514b7' https://github.com/login/device/code
device_code=xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx&expires_in=898&interval=5&user_code=ABCD-0123&verification_uri=https%3A%2F%2Fgithub.com%2Flogin%2Fdevice

Next visit https://github.com/login/device in a browser and enter the user code you just received when prompted.

Lastly you need to get your bearer token (replacing device_code below with the one you were given, but again using the client_id shown here):

$ curl -X POST -d 'device_code=xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx&client_id=b8ab5a8c1a2745d514b7&grant_type=urn:ietf:params:oauth:grant-type:device_code' https://github.com/login/oauth/access_token
access_token=gho_XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX&scope=&token_type=bearer

You can then use this token to make authenticated queries:

$ curl -s -X POST -d 'token=gho_XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX' https://api.memory.lol/v1/tw/USForcesKorea | jq
{
  "accounts": [
    {
      "id": 26847645,
      "id_str": "26847645",
      "screen_names": {
        "USFKPAO": [
          "2011-10-19",
          "2016-06-19"
        ],
        "usforceskorea": [
          "2017-02-20",
          "2018-03-27"
        ]
      }
    },
    {
      "id": 4749974413,
      "id_str": "4749974413",
      "screen_names": {
        "usforceskorea": [
          "2016-02-26",
          "2017-02-07"
        ],
        "usforceskorea_": [
          "2017-02-19",
          "2018-04-24"
        ],
        "USForcesKorea": [
          "2018-06-08",
          "2022-07-29"
        ]
      }
    }
  ]
}

Eventually this process will be bundled up into a command-line client, but for now this approach will work with existing tools like curl.

Importing data

The application currently supports importing data in two file formats. The first requires one Twitter user object per line (in JSON format with an additional snapshot field representing the observation time as an epoch second). The second is a CSV format with at least three columns (Twitter user ID, screen name, and observation time as epoch second).

Future

Anything about the web service is subject to change at any time, including its availability.

There are non-public endpoints that I'm likely to open up at some point. These provide full historical user profiles, information about suspension or deactivation status, etc.

Terms of service compliance

This web service simply provides an interface to an index for content that is hosted in public archives, and the project aims to be compliant with the terms of service of all platforms that were accessed in generating this index.

This repository does not contain data from any social media platform.

License

This software is published under the Anti-Capitalist Software License (v. 1.4).