Get Skeleton

For your history

Your history, finally useful.

Years of email, photos and messages are sitting in accounts and exports you never open. Bring them into your Skeleton, read on your own computer, and the AI and agents you use can answer from your real past, and show you where each answer came from.

Get SkeletonSee How Import Works

“When was our first trip to the cabin, and who came?”

Searching 2011 to 2026…
’11
’12
’13
’14
’15
’16
’17
’18
’19
’20
’21
’22
’23
’24
’25
’26

    01 · Why Skeleton

    Your AI is only pretending to know you, while years of your history sit unused.

    Ask an AI about your own past and it answers warmly and confidently, from a guess. Meanwhile fifteen years of bookings, photos and conversations are already written down in your mail, your camera roll and your messages. Skeleton brings that record together, so your AI can use the real thing.

    Your AI, on its own

    Starts from nothing, every time

    What did we do for our tenth anniversary?

    Already in your history

    Mail, photos and messages, by year

    20112026

    18,204

    emails

    12,440

    photos

    41,208

    messages

    June 2019 · Porto

    A booking for Casa das Laranjeiras, 214 photos by the river, and 36 messages with Ruth about where to eat.

    02 · Bring your history in

    Bring years of your history in, in one import.

    Google Takeout brings your Gmail, Calendar, Photos, Maps, YouTube and Chrome history. Amazon brings every order, and Spotify brings every song you have played. Each record keeps its date and where it came from.

    Google Takeout

    Importing on this Mac · 0 records

    • GmailWaiting
    • CalendarWaiting
    • Photos & VideosWaiting
    • MapsWaiting
    • YouTubeWaiting
    • ChromeWaiting

    Only the extracted records are uploaded. The raw export never leaves your computer.

    Amazon

    Every order, back to your first one.

    1,318 orders

    Spotify

    Your extended streaming history.

    62,904 plays

    Coming soon

    Exports from Apple, Meta, Microsoft, Netflix, Claude, ChatGPT, X, LinkedIn, Discord and Strava.

    Read on your computer

    Pick the export Google, Amazon or Spotify sent you in the Skeleton desktop app. It is parsed on your machine, and only the extracted records are uploaded.

    Or straight from Google Drive

    If you sent your Takeout to Drive, paste the link. Skeleton reads it from Drive on its servers, so nothing downloads to your computer.

    Only when you start it

    An import runs when you start it, into the space you choose. Nothing checks your accounts on a timer.

    Today’s messages, photos and notes, too.

    Messages, Photos, Notes and WhatsApp are read by the desktop app on your own Mac. Connect your accounts too, and your AI reads from them when a question needs it.

    and 86 more

    03 · Find it

    Find the photo, the message or the email from years ago.

    Describe it the way you remember it. Your AI looks through your photos, messages and imported mail together, and brings back the thing itself with everything from the same week around it.

    Person · MumNear · Wilbury AllotmentsMay to Aug 2019

    From the same summer

      Photos are searched on your Mac, by the people you have named in Apple Photos, by place and date, and by the text in the picture.

      04 · Ask it

      Ask about years ago, and get the answer with its sources.

      Your AI searches the archives you imported alongside everything else in your Skeleton, joins the pieces up, and cites each record it used, so you can open the actual email before you rely on it.

      What was the B&B in Porto we loved, and which room did we have?

      Your AI, with Skeleton

      Casa das Laranjeiras, on Rua de Cedofeita. You had room 41, the one with the balcony over the garden2, and you told Ruth you would go back for your fifteenth3.

      1. 1Booking confirmation: Casa das LaranjeirasGmail, from Takeout · 3 Jun 2019
      2. 2Balcony, room 4Photos · 21 Jun 2019
      3. 3Chat with RuthWhatsApp · 26 Jun 2019

      Which song did we walk out to at our wedding?

      Your AI, with Skeleton

      “Harvest Moon” by Neil Young. You sent it to the DJ a week before1, and it has been one of your most played songs every August since2.

      1. 1Re: Running order for SaturdayGmail, from Takeout · 2 Aug 2017
      2. 2Streaming historySpotify export · 2017 to 2026

      05 · Sorted and joined up

      Your places, orders and people, sorted from what you imported.

      Turn on a list and it fills from the records already in your Skeleton, with no model reading them to do it. The same person across Messages and WhatsApp becomes one record, and when a match is only a guess, Skeleton asks you first.

      From Maps in your Takeout

      • Casa das LaranjeirasPorto
        Saved Jun 2019
      • Glen Affric cabinHighlands
        Saved Jun 2014
      • Wilbury AllotmentsHove
        Saved Apr 2019
      • Bar ItaliaSoho, London
        Saved Nov 2012
      RO

      Ruth Okafor

      Messages since 2012 · 9,412 messages

      • Messages+44 7700 900418
      • WhatsApp+44 7700 900418

      Joined automatically: the same number in both apps.

      Proposal

      Is “Ruthie” in the WhatsApp group “Cabin 2014” the same person as Ruth Okafor?

      Only the name matches, so Skeleton asks instead of merging.

      Keep SeparateMerge

      06 · Your own agents

      Agents that do the digging through your past.

      Give an agent a name and a face, and share the parts of your Skeleton it needs. It goes through years of photos, mail and messages for you, writes down each thing it finished with the thing it made attached, and every save or send goes through your approvals.

      The agent Saoirse

      Saoirse

      Reading the removals emails from 2016

      • I read 340 emails and 1,900 messages about the move.

        14:02

      • I built a timeline, from the first viewing to the last box.

        14:31

        Timeline: the move to Hove

        Doc · Personal › Family

      • Share the timeline with Ruth?

        14:32
        DenyShare

      • Reading the removals emails from 2016

        now

      When the session ends, this list becomes its summary in your Skeleton.

      Made by Saoirse

      Personal › Family

      Timeline: the move to Hove

      A document Saoirse wrote from 340 emails, 1,900 messages and 31 photos. Every line names the record it came from.

      1. Mar 2016

        First viewing, 14 Wilbury Road

        Gmail

      2. Apr 2016

        Offer accepted at £412,000

        Gmail

      3. Jun 2016

        Dad: “the survey found the damp”

        Messages

      4. Aug 2016

        Exchange of contracts

        Gmail

      5. Sep 2016

        Moving day, 31 photos

        Photos

      6. Oct 2016

        Mum: “we finally found the kettle”

        WhatsApp

      It knows who it is

      Its persona, memory and contract are documents in your Skeleton that you can read and edit.

      It shows its work

      Each finished step names what it made: an album, a document, a timeline, with the records it used.

      It asks the way you told it to

      Choose whether saves and sends wait for you or go through on a rule you set. Every decision is kept.

      07 · Benchmarks

      Still right when your history runs to thousands of records.

      Years of history make a big account. We asked AI models the same questions about a made-up person's life at 30, 200, 500 and 1,000 records, three ways: with everything pasted into the prompt, with one search, and through Skeleton.

      Answered correctly, by size of account

      Pasting everything falls from 84% to 62% as the account grows. Skeleton stays at about 99%.

      Higher is better

      Accuracy by number of records for each approach
      RecordsSkeletonPaste everythingOne search
      3098.9%84.4%74.4%
      20097.8%83.3%72.2%
      50098.9%71.1%76.7%
      1,00098.9%62.2%72.2%

      Prompt size per question

      Thousands of tokens sent to the model. One search sends the least, and gets about a quarter of its answers wrong.

      Lower is better

      Input tokens per question by number of records for each approach
      RecordsSkeletonPaste everythingOne search
      3029.7k11.1k0.7k
      20030k68.7k0.7k
      50030.1k170k0.7k
      1,00029.9k338.5k0.7k

      Personal details that reached the model

      The same 21 tasks at 30 and 120 contacts. Skeleton sent 99% fewer than pasting everything, and got every task right.

      Lower is better

      Skeleton67

      42 of 42 tasks right

      Paste everything8,904

      41 of 42 tasks right

      One search101

      34 of 42 tasks right

      Planted attacks that reached the model

      Instructions hidden inside records, 168 per approach. No attack was detected succeeding against any approach; this counts the ones that reached the model.

      Lower is better

      Skeleton0 of 168
      Paste everything168 of 168
      One search41 of 168

      Internal benchmark on synthetic accounts, September 2026. Not yet independently verified. Not yet publication ready under our own statistical checks. Accuracy and prompt size on Claude Haiku 4.5 and GPT-5.6 Luna; personal details and planted attacks on GPT-5.6 Luna. Pasting was simulated by putting every record in the prompt; no chat app or project was tested. No attack was detected succeeding against any approach; the count is attacks that reached the model.

      See All Benchmarks

      08 · You stay in control

      You decide who reads your past, and you can check.

      Old mail and messages are some of the most personal things you have. The access log shows which AI or agent read which record, and when, so you can always check.

      Access log

      live

      • Gemini read Streaming history, Aug 2017

        Personal · Spotify archive

      • Saoirse read Chat with Dad, Jun 2016

        Family · Messages

      • ChatGPT read Order: 4-person tunnel tent

        Personal · Amazon archive, 2021

      • Wren read 23 photos · Glen Affric

        Personal · Photos, 2014

      • Claude read Re: Flat 2, check-out inspection

        Personal · Gmail archive, 2019

      Every read is recorded. Entries cannot be edited or removed.

      Nothing reads on a timer

      Imports start when you start them. Connected accounts are read when you, your AI or a routine you set up asks.

      Raw exports stay on your computer

      An export you pick in the desktop app is parsed there. Only the records it holds are uploaded.

      Saves go through your approvals

      Decide whether saves and sends wait for you or go through on a rule you set.

      Every change is kept

      Memories and documents keep every version, with who or what made each change.

      Each app sees only what you grant

      Give an AI your Personal space and nothing else. Change it or revoke it at any time.

      Delete it or take it with you

      Delete an imported archive, export your Skeleton as a zip, or keep it in storage you own.

      Google Takeout, Amazon and Spotify import today. Exports from Apple, Meta, Microsoft, Netflix, Claude, ChatGPT, X, LinkedIn, Discord and Strava are coming soon.

      Bring your history in, and let your AI use facts from your actual life.

      Start with your Google Takeout. Skeleton keeps every record with where it came from, shares only what you grant, and records every read.

      Get SkeletonSee Examples

      People, messages and records on this page are invented examples.