Omniscio

Bring your AI with you

Your old AI (ChatGPT, Claude, Gemini, Copilot, Perplexity, Grok, DeepSeek or another) has learned things about you and holds your projects; Bring your AI with you copies those facts into About you and turns each project you are still working on into a hub. Add the data export your AI gives you, or copy a prompt into it and paste its reply back. Then you tick what you want to keep.

On this page

What it is

A guided import that brings the facts another AI has learned about you into About you, and your ongoing projects into hubs. It reads only what you add or paste, removes anything secret, lets your own AI turn the rest into short facts, hub cards, skills and schedules, and creates just the ones you tick.

Where to find it

Three places open the same window. Desktop app only.

  • The Welcome chat. Right after your first answer, if you said you use ChatGPT, Claude or another AI, your guide asks whether you want to bring over what it knows about you. Say yes and it opens the window for you.
  • The header "…" menu, under Help & support → Bring your AI with you. You can pin it to the toolbar from there.
  • Settings → Setup Wizards → Bring your AI with you → Bring it over.

The Command Palette lists it too. If the window is already open, opening it again keeps your place.

How it behaves

  1. The explanation. The first screen says what will happen and what never leaves your computer, and asks Which AIs did you use?: ChatGPT, Claude, Gemini, Copilot, Perplexity, Grok and DeepSeek. Tick each one you used; you only see the steps for the AIs you tick. Press Start.

    • Tick none and you go straight to the memory prompt (step 3).
    • Tick only DeepSeek and you go straight to the project prompts (step 5): DeepSeek has no export to add, and the project prompt is how its projects come over.
  2. Your chat history (optional). The screen has a section for each AI you ticked whose export Omniscio reads: Claude, ChatGPT, Gemini, Copilot, Perplexity and Grok. Add any of them, or none.

    Claude. If you used Claude, press Open Claude's export page and press Export data there. Claude emails you a download link, usually within minutes to a few hours, and the link works for 24 hours. Download the file (named like data-…-batch-0000.zip): while this screen is open Omniscio offers it from your Downloads folder, or press Choose a file… to pick it. Omniscio then reads it on this computer and shows what it found:

    • the projects in it, the newest 10 ticked; untick any you do not want,
    • Chats to read: all chats, the last 12, 6 or 3 months, or none,
    • an estimate of how much of your AI it will use, which shrinks as you untick projects or read fewer months,
    • how many parts of the export could not be read and will be skipped.

    ChatGPT. If you used ChatGPT, press Open ChatGPT's export page and press Export data there. ChatGPT emails you a download link, usually within minutes to a few hours (it can take up to 7 days), and the link works for 24 hours. Download the zip: while this screen is open Omniscio offers it from your Downloads folder in the ChatGPT section, or press that section's Choose a file…. It shows how many chats it found, Chats to read and the estimate, the same as for Claude.

    Gemini. If you used Gemini, press Open Google Takeout. There, press Deselect all, tick Gemini (for your Gems) and My Activity, and under My Activity choose Gemini Apps only, then create the export. Google emails you a download link, usually within hours (it can take a few days). A big export comes in several zips named like takeout-…-001.zip, -002.zip: download every part and add each one. Only the Gemini parts of a Takeout are read, never your search history, Drive or anything else in it. It shows how many Gems and chats it found, Chats to read and the estimate.

    Copilot. If you used Microsoft Copilot, press Open your Microsoft privacy page, find Copilot activity history and press Export all activity. A CSV file downloads straight away; Omniscio offers it in the Copilot section when its name starts with Copilot, or press that section's Choose a file….

    Perplexity. If you used Perplexity, press Open your Perplexity account page and press Export my data there. Perplexity emails you a download link, usually within a few days. Download the zip: Omniscio offers it in the Perplexity section when its name starts with perplexity, or press that section's Choose a file…. It shows how many chats it found, Chats to read and the estimate.

    Grok. If you used Grok, press Open your xAI account data page and press Download account data there. xAI emails you a download link, usually within minutes. Download the zip: Omniscio offers it in the Grok section when its name starts with grok or xai, or press that section's Choose a file…. Only your Grok chats are read from it.

    Not using any of them? Press Next. A big export can take a few minutes to read.

  3. Copy the prompt. Press Copy and paste the prompt into your old AI. It asks the AI to list everything it remembers about you, one memory per line, with the date if it knows it.

  4. Paste the reply. Paste the AI's whole answer into the box and press Next. Leave the box empty and press Next if you only want to bring projects.

  5. Bring your projects. Press Copy and paste the second prompt into each project you want to keep working on (a ChatGPT or Claude project, a Gemini Gem, and so on). Your old AI writes a short handover that starts with # Omniscio project answer, as a file to download named omniscio-project-answer-<project name>.md when it can make files.

    • If it gives you a file, download it. While this screen is open, Omniscio spots it in your Downloads folder within a couple of seconds and asks: Add it or Ignore.

    • If it writes the handover in its reply, paste it into the box and press Add project; several at once is fine.

    • If you use Claude desktop, you can skip the copying: see The Claude desktop add-on.

    Each project appears in a list with a tick, and you can untick any you want to skip. Up to 10 projects at a time, counting the ones ticked from your Claude export.

    Coding tools on this computer. Below the projects, if you use Claude Code, Codex, Gemini CLI or Cursor, Omniscio lists each one it finds with the exact files and folders it would read. Nothing in them is opened yet. Untick any tool you want to leave out and press Look in these folders. Omniscio then shows the project folders those tools know about, each with a tick, and how many new skills it found. Untick any folder you do not want as a hub. A folder that is already a hub is not offered again; the plan lists it as already here.

  6. Read it. Press Read it. A processing screen shows the swirling Omniscio logo, how many parts have been read (your memories, each project and each export's chats are one part each) and roughly how long is left; reading takes about a minute per project.

    • Minimize (or closing the dialog) keeps the read going while you use the rest of Omniscio. Open Bring your AI with you again at any time to see how it is doing.
    • When a minimized read finishes, an inbox card says Your plan is ready; its Open the plan button opens the dialog straight on the plan.
  7. Check the plan. Each project that still has work going on, or a job that repeats, is a hub card: a tick box, its name (which you can change), where it came from (a project answer, a Claude project, a job you kept coming back to in your Claude, ChatGPT, Gemini, Copilot, Perplexity or Grok chats, or an existing folder), and its description. Projects that look finished are listed as left out. Below them are Skills to add (one for each project that had instructions, finished ones included, and one for each Gem in your Gemini export, followed by a list of skills found in your coding tools), Schedules to add, switched off (one for each repeating job, with when it runs, where it runs and what each run does), and the facts about you. Everything starts ticked; untick anything you do not want.

  8. Save. Save ticked items creates the ticked hubs, skills and schedules and adds the ticked facts to About you. The last screen lists what was created, anything that was already there, and anything that could not be made and why.

Skills from your assistants

  • A project's instructions (a ChatGPT or Claude project's instructions, a custom GPT, a Gemini Gem) become a skill: a SKILL.md in your skills folder (~/.claude/skills/<name>/), holding the instructions word for word and a short description of when to use it. Every chat can use it.
  • The skill's folder is named after the project in lower case with hyphens. A skill with exactly the same content is skipped as already there; a different skill that already has the name is never replaced, and the result says so.

Schedules from your repeating jobs

  • Each job listed under a project's repeating jobs becomes a schedule in Cron Jobs: a session that runs the job's instruction at its time, named <project>: <job>.
  • Every brought-over schedule starts switched off. Nothing runs, and nothing is spent, until you open Cron Jobs and turn it on.
  • It runs in that project's hub when you create the hub (or it already exists), otherwise in your general Claude chats.
  • A schedule whose time is not a regular repeating time, or that would run more often than every 5 minutes, is not added; the result says which. A schedule whose name is already in Cron Jobs is skipped.

Where your new hubs go

  • Each ticked hub gets its own folder in Documents → Omniscio Hubs, named after the hub, and appears in the sidebar like any other hub.
  • The folder holds CLAUDE.md with the project's instructions, so every agent in that hub follows them, and .claude/docs/where-we-left-off.md, a note your AI wrote from the handover: where the project stands, the open threads, the files your old AI knew of and any repeating jobs. Agents in the hub read it when they start.
  • Files from your old project are not copied. Drag them into the hub's folder yourself. For a Claude project the card lists the files it had, so you know which ones to bring.
  • A hub whose folder is already a hub is skipped, so running it twice does not make duplicates.

Hubs whose code is on GitHub

  • Marked in the plan. A hub card shows GitHub: owner/repo when its project lives in a GitHub repository: a coding-tool folder whose git settings point at GitHub, or a project handover whose GitHub repository section names one (the per-project prompt asks for it; write None when there is none).
  • Connect GitHub in the plan. When any hub links to GitHub and GitHub is not connected, the plan shows a Connect GitHub button. It uses the same GitHub sign-in as the rest of Omniscio (it installs the GitHub command-line tool first if it is missing). Connecting does not change the plan or your ticks.
  • A handover's repository is copied to this computer. When you save with GitHub connected, the repository is copied (cloned) into Documents → Omniscio Hubs, in a folder named after the repository, and the hub opens it. The project's instructions and the "where we left off" note are added only if the repository does not already have them, and they are kept out of git so an agent never commits them.
  • Saved before connecting? Those hubs wait. The result lists them under Waiting for GitHub, with their schedules, and shows Connect GitHub again. As soon as GitHub is connected they are created from the same plan, without reading anything again; Create them now tries again if they are still waiting. They are kept only while Omniscio is running and the dialog is open: if Omniscio restarts or you close the dialog first, press Read it again.
  • A copy of that repository already in Omniscio Hubs is added as the hub as it is, with nothing written into it; a folder of that name holding a different repository is never used.
  • A coding-tool folder never waits. Its code is already on this computer, so it is added at save whether or not GitHub is connected. Hubs without GitHub are not affected at all.
  • Only the owner and repository name are kept from a link; a password or token in it is dropped, and the copy always uses the plain https://github.com/owner/repo.git address.

What Omniscio reads from your coding tools

  • First, only what exists. When the projects screen opens, Omniscio checks which of each tool's known files and folders are on your computer and shows them. It opens none of them.
  • Then, only what you ticked. Look in these folders reads the ticked tools' lists of project folders (Claude Code's .claude.json, Codex's config.toml, Gemini CLI's trusted folders, Cursor's recent workspaces) and their skills folders.
  • In each ticked project folder, at Read it, Omniscio only checks which instruction files are there (CLAUDE.md, AGENTS.md, GEMINI.md, .cursorrules, .cursor/rules) and reads the GitHub repository name from its git settings. It keeps just the owner and repository name, never a password or token in the link. No other file in the folder is opened.
  • Your tool's own instructions (Codex's AGENTS.md, Gemini CLI's GEMINI.md in your home folder) are read for facts about you, with passwords and keys removed first, like a paste. Claude Code's own instructions and skills are not read: Omniscio already uses them.
  • A project folder becomes a hub that opens that folder where it is. Nothing is written into it and nothing is copied to Omniscio Hubs. Its card says where it lives, which tools it came from, its GitHub repository and the instruction files already in it.
  • Skills. Each skill a tool has that Omniscio does not is listed under Skills to add, with where it came from. Only ticked skills are copied into your skills folder, where every chat in Omniscio can use them. A skill with the same name already there is never replaced.
  • Up to 40 project folders and 50 skills at a time.

Running it again

  • You can run Bring your AI with you as often as you like, for example after more work in your old AI. Omniscio keeps a record of what it has already brought over (never the text itself), so the plan proposes only what is new: new projects, new skills, new schedules and facts from memories it has not read before.
  • Memories and projects that have not changed are not read again, so they cost nothing from your AI allowance. The plan lists what an earlier import already looked at, or says Nothing new since your last import.
  • A project that already has a hub but whose handover has changed appears under Updates to your hubs. A ticked update is added to the end of that hub's "where we left off" note under a dated heading; nothing already in the note is removed. Untick it to leave the note as it is.
  • Hubs, skills and schedules are never made twice. Something you unticked last time is not offered again unless its project's handover changes. A hub that could not be made last time is offered again; a skill or schedule that could not be made is offered again once that project's handover changes.
  • The same goes for a Claude export added again: unchanged projects and memories are listed as already looked at, a changed project with a hub appears under Updates to your hubs with the list of files to drag in, and a job from your chats that already has a hub is not offered again. A fact found in your chats that you unticked last time may be offered again.
  • If parts of an export could not be read, the plan says how many, and it does not claim nothing is new.

What Omniscio looks at in your Downloads folder

  • Only file names. It looks for files whose name starts with omniscio-project-answer and ends in .md, .markdown or .txt, that arrived after you reached the projects screen. No other file there is opened or even looked at.
  • A file is opened only when you press Add it. Ignore leaves it unread, and it is not offered again. Either way the file stays in your Downloads folder.
  • It looks only while the projects screen is open. If Omniscio restarts while you are waiting, open Bring your AI with you again: the projects screen offers anything that arrived in the meantime.
  • Closing the dialog does not stop the wait. If your export email arrives later, even after you have closed Omniscio, open Bring your AI with you again and the file is offered. The wait ends when you press Read it, and is forgotten after a day.
  • It looks in your computer's own Downloads folder. If your browser saves downloads somewhere else, open the file and paste its text into the box instead.

What Omniscio reads in your Claude export

  • Your projects (name, description, instructions and the NAMES of their files, not the files), what Claude remembers about you, and your chats: every chat's title and a short excerpt of the newest ones, up to a fixed amount. Account details in the export are never opened.
  • It reads the file one chat at a time, so even a very large export does not slow Omniscio down.
  • Claude's export does not say which chats belong to which project, so your AI looks at the chats together and suggests up to 5 jobs you kept coming back to, each as its own hub.
  • The export is kept in memory only while you work through the dialog, and forgotten once your plan is read or you close the dialog. Nothing from it is saved except what you tick. The zip stays in your Downloads folder; delete it yourself when you are done.

What Omniscio reads in your ChatGPT export

  • Your chats: every chat's title and a short excerpt of the newest ones, up to a fixed amount, and your custom instructions (what you told ChatGPT about yourself and how it should answer), which become facts for About you. user.json, with your email and phone number, is never opened.
  • A big account's export is split into several files (conversations-000.json, -001.json, …); Omniscio reads them all as one, and a chat that appears twice is read once.
  • When you edited a message or asked ChatGPT to answer again, only the version you kept is read, as ChatGPT shows it, never the other versions.
  • Only what you and ChatGPT wrote is read: ChatGPT's hidden reasoning, its tool and code output and any pictures are skipped.
  • ChatGPT's export does not name your projects or include their instructions, saved memories or custom GPTs, so your AI looks at the chats together and suggests up to 5 jobs you kept coming back to, each as its own hub. Bring a project's instructions with step 5's prompt instead.
  • It is read one chat at a time and kept in memory only, exactly like a Claude export.

What Omniscio reads in your Gemini and Copilot exports

  • Gemini: a Google Takeout can hold everything you keep with Google, so only the Gemini parts are opened: your Gemini Apps activity (each question you asked and Gemini's answer is read as one chat) and your Gems. Search, Drive, YouTube and every other part are never opened.
  • Each Gem (its name, description and instructions) is proposed as a skill marked "From your Gemini Gem", with passwords and keys removed; it costs no AI request, and only a ticked one is made. Up to 10 Gems are brought over, the newest first.
  • Copilot: the CSV is read row by row; each conversation becomes one chat (or each day, when the file has no conversation column).
  • Both are read on this computer and kept in memory only, exactly like a Claude export.

What Omniscio reads in your Perplexity and Grok exports

  • Perplexity: your threads, each read as one chat from its questions and answers.
  • Grok: only the file that holds your Grok chats. Your profile, billing details and uploaded files in the same zip are never opened.
  • Neither Perplexity nor xAI publishes how its export is laid out, so Omniscio reads them carefully: anything it does not recognise is skipped and counted with the parts that could not be read, and the rest is still read.
  • Both are read on this computer and kept in memory only, exactly like a Claude export.

The Claude desktop add-on

If the Claude desktop app is installed on this computer (Windows or Mac), the projects screen offers an add-on for it, so you can bring a Claude project over without copying anything.

  1. Press Add to Claude desktop. Omniscio asks first: it adds one Omniscio entry to Claude desktop's settings on this computer, and nothing is sent over the internet. Press Yes, add it (or Not now to leave Claude desktop unchanged).
  2. Restart Claude desktop so it picks the add-on up.
  3. Keep Omniscio's projects screen open. In any Claude project, type send this to Omniscio. The first time, Claude desktop asks whether it may use Omniscio's tool: allow it.
  4. Claude writes the project's handover and hands it to Omniscio, and the project appears in the list on the projects screen, ticked, within a couple of seconds. If the projects screen was not open, it appears the next time you open it (within a week).
  • Everything stays on your computer: Claude desktop hands the handover to a small Omniscio helper, which leaves it in Omniscio's own folder for the projects screen to pick up. Each handover is read once it is in the list and then deleted.
  • Up to 10 handovers can wait at a time; after that Claude says to open Omniscio first.
  • The list holds up to 10 projects. A handover that arrives once it is full keeps waiting, and it appears the next time you run the import.
  • Remove add-on takes it out of Claude desktop again: it is on the projects screen, and on the Bring your AI with you card in Settings → Setup Wizards while the add-on is on. Only Omniscio's entry is removed; everything else in Claude desktop's settings stays as it was. Restart Claude desktop afterwards.
  • If you move or reinstall Omniscio, the card says the add-on points at an older copy. Press Update it, then restart Claude desktop.

What happens to your text

  • Passwords, keys, tokens, email addresses and phone numbers are removed before any AI reads the text, including a password or PIN written as a sentence ("my bank PIN is 4821"). The plan tells you how many memory lines had something removed. Project handovers are cleaned the same way, so nothing secret ends up in a hub's files either.
  • The date on each memory goes to your AI with its line, so it can tell an old fact from a new one.
  • The sorting is done by your own AI: your connected Claude account, or your own API key when no Claude account is connected. It never uses an allowance Omniscio pays for, and the text never goes to an Omniscio server. Each project is a separate request, and the project's instructions are not sent: they are copied into the hub as written.
  • Only what you tick is saved. The pasted text is not kept anywhere; closing the dialog before you press Read it forgets it, and a plan is forgotten as soon as you save it. While a read is running, what you brought over and the parts already read are held in memory only, never on disk, until the plan is ready or you go back.
  • A fact that is already on About you (ignoring capital letters and spacing) is skipped, so running it twice adds nothing new.

When something goes wrong

  • That file is not a Claude export — choose the zip from the link Claude emailed you, named like data-…-batch-0000.zip.
  • That file is not a ChatGPT export — choose the zip from the link ChatGPT emailed you. A Claude export picked in the ChatGPT section (or the other way round) says this too: use the other section.
  • That file has no Gemini data in it — the Takeout was made without Gemini or My Activity → Gemini Apps ticked. Make it again with those ticked.
  • That file is not a Copilot export — choose the CSV Microsoft downloaded from Export all activity.
  • That file is not a Perplexity export or not a Grok export — choose the zip from the link in the email Perplexity or xAI sent you.
  • That file could not be opened — another program is using it, or it was moved. Close that program, then press Choose a file… and pick it.
  • That file is too big to read — exports over 2 GB are not read. Ask for a new export.
  • Your export is no longer loaded — it was already read, the dialog was closed, or Omniscio restarted. Go back to step 1 and add it again.
  • Parts of your export could not be read — those chats or projects were skipped; the rest was read.
  • No memories were found — nothing was pasted to read. Paste the whole reply, or add a project.
  • That file is no longer in your Downloads folder — it was moved or deleted, or you already added or ignored it. Download it again, or paste its text.
  • That is not a project answer — the pasted text does not start with # Omniscio project answer. Paste the AI's whole reply. If your old AI translated the headings, ask it again to keep them in English.
  • Claude desktop's settings file could not be read — Omniscio changed nothing. Open Claude desktop once so it writes a fresh settings file, then press Add to Claude desktop again.
  • Claude desktop's settings could not be saved — close Claude desktop and try again.
  • Projects your AI could not read — the plan counts any project whose handover could not be judged. Paste that one again and read it on its own.
  • A hub was not created — the result says why: your plan's hub limit is reached, its folder could not be made in Documents, or something else went wrong. The other hubs are still created.
  • Nothing new was found in those coding tools — you looked in your coding tools but ticked no folder, and they hold no new skill or instructions. Tick a project folder, or go back and paste a reply.
  • Its folder is no longer there — a project folder was moved or deleted before you saved. The other items are still saved.
  • It could not be copied from GitHub — the repository could not be cloned: it may be private to an account that is not the one connected, renamed, or GitHub could not be reached. Check the repository name in the handover, then read it again.
  • Could not check whether GitHub is connected — press Check again; the plan is unchanged.
  • A skill was not added — copying it failed. Try again; nothing half-copied is left behind.
  • Coding-tool files that could not be read — a tool's project list was damaged or in a format Omniscio does not know. The other tools are still read.
  • A skill or schedule was not added — the result says why: a different skill already has that name, its timing is not a repeating time, it would run more often than every 5 minutes, or it had nothing to do. Everything else ticked is still created.
  • Your AI could not be reached — no Claude account or API key is connected, or it refused. Connect one in Settings → Accounts and press Try again.
  • Reading stopped part-way — the processing screen says why and offers Try again. What was already read is kept, so trying again sends only the parts not yet read to your AI. If the dialog was minimized, an inbox card Bring your AI with you stopped part-way opens it. Back forgets the stopped read and returns to your projects.
  • This read is no longer running — Omniscio restarted while it was reading, so nothing was kept. Go back and start it again.
  • Your export is no longer loaded — the read could not start, so the dialog goes back to your projects; add the export again in step 1.
  • Your AI's answer could not be read — try again; the answer did not come back in the expected shape.
  • This plan is no longer available — Omniscio restarted, or several newer plans were made since. Go back and press Read it again.
  • About you is full — About you holds 100 facts. Delete some there to make room.

About you is the list the facts are saved to, and every chat reads it.

Last verified 2026-11-13