Help

Six small steps take you from an empty app to an assistant that knows something about you. Each one is worth doing on its own.

1. Open Capture → Connect and connect one thing.
2. Open Home → Dashboard and click “The Mind” card to see everything captured so far, raw.
3. Click the card again to close it, then type a question in your own words into “Ask your mind.”
4. Open Capture and click “Scan now” on your source.
5. Open Review and approve one of the suggestions, then watch it appear in Memories.
6. Open Agent & Chat and click “Tell me about my memories.”

The left menu lists every page: Home, Life, Agent, Capture, Review, Vault, Settings and Help, plus Context and Analytics on Pro Plus. Most pages have tabs along the top.

The ← and → buttons above a page's title go back and forward through the pages and tabs you've visited, the way a browser does. Going back puts you where you were reading, not at the top.

If something is open over the page — a Moment, a linked memory, the Mind Map — Back closes that first.

Every page has a folded “Using …” strip at the top. It holds quick tips for that page, with a “Full guide →” link back here.

It's worth opening once on any page you're new to.

Nothing is captured until YOU connect a source.

The “Pause capture” button on Capture stops everything instantly.

Once memories are flowing, spend two minutes on Encryption at rest. It's the difference between a private archive and a readable file on disk.

While your archive is unencrypted, the Home tab shows a prompt for it. “Encrypt now →” takes you straight to the setting, under Settings → Settings → “Security — Encryption at rest.”

Home has three tabs: Home, Dashboard and Tour.

Home is a map of the whole app plus what happened today. Dashboard is your memory at a glance. Tour explains what OpenMindex can do, one capability at a time.

Tour opens on Welcome — what OpenMindex does for you and what each plan gets you. The plan cards are folded; click “What each plan gets you - view subscriptions” to open them.

The tabs along the top walk through one capability each: your timeline, following one thread through your life, your AI tools, search, the Agent, what it takes in, how you approve what's kept, and how it stays yours. “Help →” at the end of the strip opens this guide.

Nothing there can be dismissed. Pick Home or Dashboard when you'd rather land on your own data, and Tour is still there whenever you want it.

Tour → Get Started is the checklist: five things to set up, in order, each one ticking itself off as you do it.

The five steps are “Connect what you already use,” “Let it read them,” “Keep what matters,” “Give an AI tool access” and “See it pay off.” Each has a button that takes you to the right page.

Pick what you're setting OpenMindex up for — Personal, Work, Builder, Curator or Power User — and the same five steps are described for that. It changes the suggestions and the examples, never the steps.

Once all five are done it folds away to a single line. “Show the list” brings the steps back.

While setup is unfinished, the Home tab carries one line pointing at the Get Started checklist. “Finish setup →” opens the checklist.

“Later” sends the line away for good. The Get Started tab stays where it is, so nothing is lost by dismissing it.

The Home tab starts with Today — the daily recap and what needs your review.

Under that are eight cards, one for each part of the app: Life, Agent, Capture, Review, Vault, Context, Analytics and Settings. It's a map of the whole app on one screen.

Click a card's name to open that page, or one of the chips under it to jump straight to a tab.

“Using OpenMindex” sits underneath — a folded strip with six things worth knowing.

The daily brief card recaps the last 24 hours — captures, sources, prompts — and links to anything waiting for you.

It greets you by time of day, so it reads as your brief in the morning and today's recap later on.

The “Needs review” card shows the latest 5 suggestions for things worth keeping. Approve one with ✓ or reject it with ✕ without leaving Home.

The number beside the title is everything waiting, dream proposals included. “View all in review →” opens the full list.

A “Needs attention” card appears only when something is off — paused capture, a folder permission problem, a source folder that no longer exists, a store error — and tells you the fix.

It also counts down a free trial, and warns you when a subscription needs renewing or you need to sign in again, so paid features never stop without notice.

Each line has its own link to the place that fixes it. “Refresh ↻” re-checks after you change something.

Two more cards appear under the daily brief only when they have something to say.

“Needs attention — sharing” tells you when sharing is waiting on you — for example, when someone wants you to take over something they shared. “Open Share” takes you to Life → Share.

An invite you created that nobody has used yet shows as “Waiting to be used,” with “View invite” and “Revoke.”

The Dashboard tab is your memory at a glance, in three parts.

My Memory has the five tier cards, the “Ask your mind” box and “Add a memory.” Explore has the Mind Map and your stats. Activity has the live feed and today's top sources.

Click any tier card (Me · Memoirs · Memories · Mementos · The Mind) to browse what's inside.

Click a memory to edit it. Hover over it to delete — it asks twice, so one stray click never removes anything.

Click the card again, or “Show all tiers,” to close.

Use the “Ask your mind” box to find anything by meaning — “that restaurant Sarah mentioned” works even if the memory says “dinner spot.”

Your question opens as a new conversation on the Agent & Chat page, which looks through your memories first and then answers.

Searching finds memories. For plain lookup — browsing, filtering and searching by keyword — use Vault → Mind.

If you want reasoning ABOUT your memories — summaries, patterns, advice — that's the Agent & Chat page.

Click the compact Mind Map to open the full-screen constellation of your memories. Esc closes it.

Beside it, “Analytics at a glance” is a grid of nine numbers for today. Click one to open Analytics on the matching tab, AI or Activity.

“Live activity” lists the newest memories first. Click a row to read it in full, or “View all →” to open Vault → Mind.

“Top sources today” shows where the last 24 hours of captures came from.

Use “Add a memory” to store something directly — anything you'd want your AI to remember about you.

1. Pick the tier: Memories for facts, Me for who-you-are rules your AI should always know.
2. Type the memory.
3. Click “Store,” or press Ctrl (⌘ on a Mac)+Enter.

Memories is also where your life goes: births, moves, trips, the summer you learned to fish.

Anything you'd want remembered years from now is a Memory, not a Memento — Mementos expire on their own.

The memory editor formats as you write, like a word processor. The bar above it does bold, italic, headings, bullet lists, checklists and quotes.

Underneath, every memory is kept as plain Markdown, so your AI tools read the formatting natively and it travels with the memory.

Link one memory to another with [[double brackets]] around its first line. Typing [[ or pressing the [[ ]] button offers your memories to pick from, and the link shows as a chip.

Click a link to open that memory; links inside it can be followed too. A link that points at nothing yet shows faded.

Links draw as white lines on the Mind Map.

When two memories share a first line, the list shows each one with the line under it. Pick the one you mean and the link stays on it.

A typed link to a shared title stays faded rather than guess.

Open a tier card anywhere it appears — Home → Dashboard or Capture — and any memory in it can be edited in full: its words, when it happened, its tags, where it happened, and its photos.

It's the same editor life events get, so a memory behaves the same wherever you meet it.

Dates show when something actually happened, not when OpenMindex read it.

Hover any date to see both — “Happened … · captured …” — which is why a catch-up scan doesn't dump a week of work into one late-night hour.

Life shows the memories you've marked as life events, plotted by when they happened — not when the app read them.

The bars along the top are your years. A taller bar simply means that year holds more events. Click a year to zoom into it, and “← All years” to come back.

Life has six tabs: Timeline, Map, Import, Search, Transfer and Share. On the Timeline tab, three views sit side by side: “Life” (what you chose to keep), “Family” (other people's timelines) and “Everything” (what the app captured on its own).

“Everything” flips the timeline to the opposite view: not what you chose to keep, but everything the app captured on its own — files, chats, calendar, browsing.

It draws one bar per month. None of it is on your timeline and none of it was written by you.

Click a month's bar to open just that month in Vault → Mind, where it can be read and deleted.

A memory dated to a year shows as “1987”, never as a made-up January 1st. The timeline only claims what you actually stated.

The same goes for a month, a decade or a span of years: what you entered is what is shown.

Life events that have no date are listed first, at the top of your timeline under “No date yet”, so none get missed. Plenty of Memoirs never had a date.

Edit one, give it a date — even just a year — and it moves into its place.

Set your birth date under Settings → Profile (“Born”) and years gain ages — “1987 · age 7”.

“On this day” greets you with anniversaries of your own events.

It only uses events dated to an exact day, so the app never invents an occasion from a vague year.

The gap prompt points at the largest empty stretch of years on your timeline and asks what was happening then.

That's where to add the next memory: one event from those years and the story fills in.

Threads are your tags. Click family, places, or any tag you've created to follow one person or theme through your whole life.

Click “clear” to go back to everything.

★ Favorites is always the first thread. Click ☆ on any card or Moment to add it, and ★ again to take it off.

Click the Favorites thread to see only the events you starred.

While your computer sleeps, Dreaming reads each completed year that has enough life events (at least three) and drafts a short year-in-review for your Memoirs.

The draft appears in Review → Dreaming. It never lands without your approval.

You add a life event right on the Timeline tab. A single sentence counts.

1. Answer the question the page asks, or click “Next question →” for a different one.
2. Date it as vaguely as the truth requires: “Exact day”, “A month”, “A year” or “A decade”.
3. Tag it, and add pictures with “+ Photos & files” if you have them.
4. Click “Save to Life”.

The questions are a prompt, never a queue. Answer any of them and skip the rest.

“Skip the questions” hides them, so adding an event is just writing one down. “Use the questions” brings them back.

Click “…to” beside the date to give it an end, and a date becomes an era — a trip, a job, 1997 – 2001.

An era is listed once, under the year it began, showing its full range. It doesn't repeat under every year it covers, and the ribbon of years still shades the whole stretch.

Click into any year the era covers and it appears there too, because that's what was going on then.

Anything already in your memory can join your timeline. Hover any row — in Vault → Mind, or in Me, Memoirs, Memories or Mementos — and click “Add to Life”.

The editor opens so you can date it and fill it in.

Your calendar helps without making you work. Open a year on the ribbon and each month offers at most three calendar events that look like life rather than logistics — a trip, a wedding, a graduation.

“Add to Life” puts one on your timeline. The rest just stay in the Vault.

Hover an event on the timeline and click “Edit” to change it in place: its words, its date, its tags, its place and its photos.

“Delete” removes the event from your timeline and from your memory. It asks you to confirm first.

While editing any memory, “+ Add person” tags who's in it — your grandmother, a friend.

That's all it does. The memory stays exactly where it is; tagging someone never puts it on their timeline.

On a Life moment with people in it, you can also answer “Whose moment is this?”.

“Mine” is the default. Pick someone else and the moment moves off your timeline onto theirs, because remembering something and having been there are two different things.

“Family”, next to “Life” and “Everything”, is where other people's moments live: each person's own timeline, reaching back as far as you know — including long before you were born.

Pick a name to follow one person through their whole life.

On the “Life” view, “Show family alongside” adds the people you've tagged to your timeline view. Each row is badged with whose it is.

It's a way of looking, not a merge. Their events never join your timeline, your ribbon or your Life Book.

A family tree comes in as a GEDCOM file. Ancestry, MyHeritage and FamilySearch all offer that export.

The file is read entirely on your computer. Nothing is uploaded and no account is connected.

1. Export a GEDCOM from the genealogy site you use.
2. Drop the file in your exports folder (“Open the folder” on Life's Import tab takes you there).
3. Open Life → Import and answer “Which of these is you?”.
4. Review what it offers: your own events for your timeline, and everybody else's under their own name.
5. Press Import.

Genealogy dates are full of honest vagueness — about 1922, between 1998 and 2001, from 1974 to 1986.

All of it survives as stated rather than being flattened into invented exact days.

FamilySearch also exports a .gdz, which is a tree with its photos inside it.

Those photos are offered in review beside the person or event they belong to, and only the ones you keep are saved. A photo never puts itself on your timeline.

A photo tied to no event is kept under that person's name.

You can bring a whole life in from a spreadsheet saved as CSV.

It needs a “text” column and, ideally, a “date” — plus optional end_date, person, place, tags and lat/lng.

1. Put the CSV in your exports folder.
2. Open Life → Import, where the file is offered for review.
3. Choose Mine, Family or Skip for each row.
4. Press Import.

Every row in an import is yours to claim. “Mine” puts it on your timeline. “Family” keeps it under that person's name, on their timeline in the Family view — never on yours. “Skip” drops it and never asks again.

A row you've already answered isn't offered a second time.

Importing the same events twice is guarded. An event is recognized by what it says and when it happened, not by which file it arrived in.

So dropping a file in again, downloading it again under another name, or exporting the same tree from two sites won't duplicate anything.

Anything already in your memory is marked “Already imported” and set to Skip. Making a second copy takes a deliberate click.

If an import goes wrong, “Undo this import” on the summary removes exactly what it just added and nothing else.

The Import tab keeps offering “Undo the last import” until you take it or delete those rows yourself.

A file that already carries coordinates pins them automatically. You exported it yourself, so nothing was looked up.

Most family trees only carry place names, so review offers a button to look those up — “Look up … places with OpenStreetMap”. The summary after an import still offers it for rows that named a place but had no point.

Each point it finds is an ordinary pin on that memory, which you can move or remove.

Tag a life event “places” plus the place itself (lisbon, the-lake-house) and a Places row appears on your timeline.

Click a place to follow everywhere it shows up in your life.

You click the point yourself. Nothing is looked up, and photo GPS is never read.

1. Hover the event and click “Edit”, then “Set place…” — or open the event's Moment and use its place card.
2. Click the map where it happened.
3. Name the place if you like — “the lake house”, “Lisbon”.
4. Click “Use this point”.

The “Drop on the map” checkbox, where you add an event, opens the map right after saving, so pinning where it happened is part of the same breath.

An event can hold more than one place, so a trip can be pinned everywhere it went.

Each pin can carry the day, and the clock time if you know it. That puts a day in the order you walked it: breakfast at eight, the hike at ten, dinner at six.

The map has its own tab, next to Timeline. It shows every place you've pinned.

Beside the map is a list of every place you have been. Sort it by “Newest first”, “Oldest first” or “By name”, and click one to open the moment that happened there — or pick from its list, when you have been more than once.

Pins close together group into a numbered circle that opens as you zoom in. A dashed line traces your places in the order they happened.

“Pins” / “Heat” swaps the individual places for a wash showing where your life happened most.

A Moment with a place shows a small map beside the picture. “Expand map”, under it, lays that map over the whole Moment, so you can pan and zoom around just that moment's places without leaving it.

“Close”, or Esc, puts you back in the Moment where you were.

Type a place name in the map's search box and press Enter to jump there.

Search only moves the map. It never drops a pin, so a place is still only ever where you clicked.

The map's tiles, and anything you type into its search box, go to openstreetmap.org. That is why the map shows a “Load the map” button, once, before loading anything.

Your memories and pins are never sent.

Click any event and it becomes a Moment: the whole page turns into that one memory.

Photos show as a slideshow with a filmstrip to jump around, video and voice notes play right there, and the date, map, people, tags and story sit alongside.

Every fact is editable where it stands.

“▶ Play” plays everything in the Moment in order. ←/→ step through it one item at a time, F is fullscreen and Esc goes back.

An audio file can be marked as the soundtrack (“♫ Use as soundtrack”). It loops under the photo slideshow while Play runs and goes quiet whenever a video or voice note plays.

A Moment is also where you add things. Drop files on the stage, or press + at the end of the filmstrip for the file picker.

The microphone button records a voice note, and the camera button records from a webcam or any connected camera. A recording gets a name as you keep it.

The note box at the bottom writes onto the memory — or, on a shared event, to the people on it.

The place card is always there on a Moment. It offers “Set place…” when the event has none, “Move” or “Remove” on each place it has, and “Add another place…” — an event can be pinned in more than one spot.

Each pin can carry the day, and the clock time if you know it, so a whole trip plots in order.

The row under the filmstrip acts on the item you have selected.

“Rename” changes any item's name. “Edit photo” opens a full editor — crop, rotate, brightness and color, filters, text, shapes and drawing. “Trim” on a video or audio file cuts it.

Neither editing nor trimming ever destroys anything. The edited copy is what you see, and the untouched original stays behind “Restore original”.

Each note you add to a Moment becomes its own paragraph of the story.

Drag a paragraph, or use ↑ ↓ beside it, to put the story in the order it happened.

“Add to timeline” beside a paragraph makes it its own event on the timeline.

Attach images, video, audio, PDFs or notes to any life event.

There are three ways in: “+ Photos & files” where you add an event, “+ Photo” when you hover an event, or inside Edit, where the event's photos sit alongside its date, place and tags.

Files are copied into OpenMindex's encrypted store, so moving or deleting the original later changes nothing.

Click a thumbnail for the full view. From there you can “Open” it in its own app, “Save a copy…”, or remove it.

The same photo on two events is stored once, and only leaves disk with its last use.

OpenMindex never reads what's inside a photo or document. The Agent can point you at a file, but it can't describe its contents.

“Life Book” on the Timeline tab lays your whole timeline out as a book in one click: white pages, one chapter per year, your age beside each one.

Tick what goes in it: your photos, the tags, the ages, the undated chapter, and a map at the back with every place listed under it in the order you went through them.

Reading your own book is free on any plan. Ultra is what prints it.

On Ultra, click “Print or save as PDF”. The print dialog's “Save as PDF” makes it a file you can keep or gift.

The Search tab is one box that finds any event among hundreds.

Every word you type has to match somewhere on the event — its story, a tag or person, a place, a photo or file name, or the date. So “lake 1987” narrows fast.

Results are newest first, name the photos that matched, and open as a Moment.

The Transfer tab sends photos, videos or files straight to a person or a circle — no moment, and nothing on anyone's timeline.

They get viewing-quality copies (photos up to 1,600 pixels on the long edge, video at 720p); your originals stay with you. It travels over the same encrypted connection as sharing, and it counts toward the same allowance.

1. Open Life → Transfer and choose who it's for under “To”.
2. Pick photos from any of your moments, or click “+ Add files” to add files from this device.
3. Say what it is, if you like.
4. Click Send.

What people send you lands under “Received” — never on your timeline unless you add it.

Open any photo or video full-size, or press Play on a voice note or other recording.

“Save all” saves everything to a folder you choose; on a phone it opens the share sheet — Save to Photos and the rest.

“Add to my Life” asks when it happened, so it lands on your timeline in the right place.

Until a transfer is delivered you can “Withdraw” it, which is a true unsend. After that it stays with them until they remove it.

“Remove” clears one from your Sent list. Anyone who already has it keeps their copy; anyone still waiting won't get it.

Sharing is computer-to-computer and end-to-end encrypted. OpenMindex hosts nothing and can't read what moves.

Contacts are people, not devices. They never join your device sync, never appear in mesh search, and never receive anything except events you explicitly share.

You pair with someone once, on the Share tab. After that, nothing moves until you explicitly share an event.

Sharing goes from your device to theirs, so the other person needs OpenMindex before an invite means anything to them.

On the Share tab, under the Invite card, the card “They don't have OpenMindex yet?” has a ready-made message and the link to the download page. “Copy link” copies the message for you to paste into a text or an email. Where your device has a share sheet — a phone, for one — “Share…” hands it straight to Messages, Mail or whichever app you pick.

Pair the people who were there. An invite is a one-time code, confirmed on both ends — no accounts, no directory, just your two machines meeting.

1. Open Life → Share and type who the invite is for (“Mom”).
2. Click “Invite someone”.
3. Click “Copy” to put both the Code and the Token on your clipboard.
4. Send them to the person — text it, email it, read it out.

When someone sends you a code and a token, you enter them on your own Share tab.

1. Open Life → Share and click “I have a code…”.
2. Paste everything they sent into either box — it fills them both.
3. Type your own name for them, if you like.
4. Click “Connect”.

You call her Mom, she calls you Lou. The name you give a person is only ever seen by you, and the name you type for someone always wins.

Rename anyone by clicking their name in People. The small text beside it is their computer's name, to tell one person's devices apart.

Under the invite, “Your name, as they'll see it” is how you introduce yourself. It comes from your profile, and editing it there saves it for next time. Left blank, they see your computer's name instead.

People and Circles are A–Z lists, each headed by how many you have, with a search box once they grow.

Choosing who to share with — or who belongs in a circle — uses one picker: search, tick a circle to pick everyone in it, tick people one by one, and see how many you've picked.

Circles are your private address book. Name a list (Family, Lake house crew) and nobody else ever sees it — only which events you shared with them.

A circle is a list of people you've already paired, so it only becomes available once you have at least one contact.

The Share tab keeps itself current. “Sync now”, on the Share and Transfer tabs, does it on demand: it names each person and says what moved.

Beside it is when sharing last synced.

Sharing has its own on/off switch at the bottom of the Invite card.

Off means no invites and nothing sent or received, while your people and everything already shared are kept.

To pause one person instead of all sharing, click “Block” beside them.

Nothing goes to them or comes from them — not even through someone else's moment — until you click “Unblock”. They aren't told.

They stay in your people and keep what they already have.

Anything that was still on its way to them is canceled: a moment or transfer they hadn't received is withdrawn, and photos or videos that hadn't finished arriving never do — not after “Unblock” either. What you share after unblocking goes as usual.

“Remove” takes a person off both sides after you confirm, and drops the connection both ways.

It goes further than Block: it also stops your moments with them and cuts them off completely. Unlike “Stop”, nothing more reaches them — not even photos that were still on their way.

Bringing them back takes a new invite, then “Share again” on each moment.

Everyone blocked, including people you removed, is listed under “Blocked”.

Unblocking someone you removed only makes a new invite possible. It does not hand back anything from before, and photos that were on their way when you blocked or removed them stay canceled.

If the list of people you removed or blocked is ever damaged and can't be read, Sharing stays off and the Invite card says so.

Restoring from a backup brings the list back. “Start a new list” sets the damaged one aside — anyone who was on it is no longer blocked.

Sharing sends one life event to the people who were there. The dialog shows the exact text that travels and who you're picking.

1. Hover the event on your Timeline and click “Share” (or “Share…” on its Moment).
2. Tick the people it's for; ticking a circle is a one-click way to tick its members.
3. Click “Share with …”.

The free plan can start 3 shares. After that, sharing out is part of Pro.

Receiving and contributing are free forever, on every plan.

Each person on a shared event shows a delivery status: invited → received → contributed.

“Invited” means it hasn't reached them yet. “Received” means their machine has it. “Contributed” means they've added something of their own.

Delivery takes care of itself. A share travels straight from your machine to theirs, so it goes when both are running OpenMindex and can reach each other. Until then it waits on your computer, showing “invited”.

There is no button to press on either end: the app tries again on its own, every few minutes while you're actively sharing and hourly when things are quiet.

Events arrive with photo thumbnails so they look complete immediately. The photos themselves follow when any participant holding them is online.

“Withdraw” is a true unsend, offered while the status still says invited.

After delivery, “Stop” ends future updates. What their machine already holds can't be unsent, and the app says so.

“Share again” puts back someone you stopped, or removed and later re-added as a contact. They get all of it, including what was added while they were off it.

To bring more people in later, use “Add people…” on the event on the Share tab, or simply share the same memory again.

They join the same moment and see all of it, not a separate copy.

Everyone on a moment knows everyone else on it. The person who shared it sends a signed list of who is in, with the names they use, so you all see the same “Mom”.

Messages and released photos travel between any two people on the list, so the conversation keeps moving while the sharer is away.

Share from your phone to your own computer (they sync as one), and the computer carries the moment and its files to the circle while the phone is asleep or away.

The people you share with accept your computer as you, because your phone signed it in.

Deleting ripples honestly. “Delete” on your own life event removes it from your timeline and memory, and — if it was shared — every participant gets a note suggesting they remove their copy.

Their copies are theirs; the app never reaches into their machines.

Editing a shared event sends the corrected text the same way.

On a moment you shared, what people write goes straight into the thread, like a group chat. Their words appear under the event, signed by their key.

Photos and videos they add land in your inbox first. The circle sees “Mom added a photo” but not the file, until you decide where it goes.

“Share with circle” gives it to everyone, in the thread. “Add to moment” makes it part of the moment itself, on everyone's timeline. “Remove from moment” sends it back to the thread.

“Make curator” beside a person on your event lets them move photos and videos just like you. Their own photos go straight to the moment.

Curators still can't add or remove people.

“Remove as curator” takes it back without undoing what they already moved.

“Hand over…” beside a person offers them the moment. Once they accept, it's theirs: they decide who is on it and what goes in the moment.

You stay on it as a participant, and your originals never leave your computer.

For the long run, “Carries it on” names who may take a moment over if something happens to you.

They can only take it after a full month with no word from you. Nothing changes before then.

You can manage curators, hand-overs and who carries a moment on from two places.

One is your event on the Share tab. The other is the “On this event” card beside any shared moment, which shows everyone on it with their role — owner, curator, carries it on.

React with a heart under any message or photo on a shared moment.

It's one per person, counted under the line, and it travels signed like everything else.

When someone adds to a moment you're on, a notification names who and which moment (if real-time alerts are on).

Home lists it under “Needs attention — sharing” until you look, and “Thread (n)” shows how many are new to you.

Opening the thread or the moment clears the count.

On events shared with you, add your piece with “Everyone here” or “Just to <them>”.

“Everyone here” goes to everybody on the event. “Just to <them>” reaches only the person who shared it and is never passed to the rest of the group.

“Show on my timeline” puts their event on your timeline view without making it yours.

Keeping asks whose it is. “Keep” remembers it as theirs — searchable, in your AI's context, on their timeline in the Family view.

“I was there” says it's your event too, and adds it to your own timeline as well.

Either way, keeping is the only path shared content ever takes into your own Memories.

Your timeline is your edit. Every comment or photo someone adds to a shared event has its own “Hide” for your view.

The event thread keeps everything and nobody is told. “Show” brings it back.

The Agent is your personal assistant inside OpenMindex. Ask it about your memories (“what did I decide about pricing?”) or about the app itself (“how do I connect Notion?”).

Ask in plain English, or tap one of the suggestion chips above the message box, such as “Tell me about my memories”.

Name a source and the Agent reads that source directly, newest first: “my last 10 emails”, “texts from Mom yesterday”, “latest meeting notes”, “photos from last weekend”, “what's on my todo list”.

You can ask for two at once — “the last 10 emails and my calendar for tomorrow” — and each part keeps its own time.

The Agent can count: “who emailed me the most this month” gives exact counts by person and by day.

It can catch you up: “catch me up on today” pulls a little from every source plus today's calendar.

It can follow a person: “when did I last talk to Erika” reads her across mail, messages, meetings and the calendar, newest first.

“My important emails”, “unread emails from Sam” and “flagged mail this week” use Gmail's and Outlook's own markings.

Press Ctrl+K (⌘K on a Mac) anywhere in the app for a quick ask. The answer lands on the Agent & Chat page.

Answers cite their sources. Click a [1] marker to see exactly which memories were used.

Use ↑ promote on any of them to send it to Review, where you approve it before it becomes a long-term memory.

“Take me to X →” opens that page of the app.

“▶ Run …” runs a skill. Your click is the consent; the Agent never runs anything on its own.

“save as memory” stores the answer at a tier you pick.

With your devices paired, a question also searches them. Under the answer, “Searched: your computer + …” says which devices answered, and any that didn’t are named.

What they found is listed under “from your other devices”; click a [D1] marker to see it.

The badge on each answer shows which engine answered it, and the line under it shows where the time went.

The first question after the model has been idle takes longer while it wakes up — the chat says so. Later questions come back much faster.

Quick questions — who, when, where, what did I decide — answer right away.

Questions that need reasoning — comparisons, decisions, plans, drafts — think first. You'll see “Thinking… 8s” while it works (open it to watch), and “Thought for 13s” above the answer afterwards.

The ⚡ bolt in the message box sends one question to the deep-thinker model, which always thinks first. It is slower and smarter.

For deeper work (skills, tools, multi-step research), click “⚡ Power up” to start the full agent engine when your build includes it.

Once it's running, it handles anything that needs its tools — opening a file or link, the web, a skill. Every other question still answers straight from the local model, which is many times faster.

After a dream run finishes, the 🌙 overnight brief leads the Agent & Chat page for the next day: when it ran, what it did, and how many proposals are waiting for you.

“Review →” takes you to them. Dismiss it and it stays away until the next run.

Conversations are saved in the left column — rename, delete, or start a new one any time.

Settings → Agent & Chat has a switch for each of these: act on source memories, save answers as memories, quick ask (Ctrl+K (⌘K on a Mac)), the ask box on Home, suggested follow-ups, the think-harder toggle, the overnight brief, and whether the agent may suggest skill runs.

Settings → Agent & Chat → Personality changes how the agent talks: its name, its tone (Balanced, Warm, Direct, Playful or Professional), how long its answers run (Brief, Normal or Detailed), and your own instructions like “call me LP” or “no emoji”.

The Agent & Chat page has two tabs. “Agent” is your local model working over your own memories, on your computer. “Chat” is a frontier model at a model company, on your own API key.

They share nothing. The Agent never uses the Chat models, and Chat cannot read your memories — it is given only the ones you choose.

Chat has four modes along the top — Solo, Experts, AI 2 AI and OmniPanel — and its own Settings beside them. With no provider key it starts on OpenMindex, the models on your computer; if your computer runs none, Settings is all it shows until a provider has a key and your approval.

Open Agent & Chat → Chat → Settings. It has seven sections: Providers and keys; Personas, experts, prompts; Memories; Masking; Search and budget; Usage; and Activity and history.

When this Help says “Chat's settings”, it means this tab, not the app's Settings page.

You need a key for at least one provider; one is enough to start. Under Providers and keys, each provider has “How to get a key”: the steps, and buttons that open the provider's key page (“Get an API key”), its list of models and prices (“Models and prices”), and its billing page (“Credit and usage”) in your browser.

1. Open Agent & Chat → Chat → Settings → Providers and keys.
2. Open “How to get a key” on a provider and follow its steps to create a key.
3. Paste the key into that provider's box and press Save.
4. Press Test to check that the key works and to list the models it can use.
5. Tick the box under the key to approve sending to that provider.

“OpenMindex” is first in the provider list. Its models, under “Local”, are the ones already running on your computer — the same engine the Agent uses. No key, no bill, and nothing is sent anywhere.

Because nothing leaves, a local chat has no masking and no web search, and it does not appear under Cloud activity.

Only models that are already loaded are listed, so a chat never pushes the Agent's model out of memory. Load or change models on the Models page.

An OpenRouter key gives you models from many companies — OpenAI, Anthropic, Google, xAI, Meta, Mistral and more — on one key and one balance.

OpenRouter passes each message on to a company that hosts the model you pick. Chat asks it to use only hosts that do not keep or train on your prompts.

Not sure which model to pick? Each provider's list starts with five recommended models, under “Recommended”. Every other model your key can use is under “All models”.

Long lists have a filter box. “Another model…” lets you type the name of a model that isn't listed.

“Model guide”, next to the model list, says what each recommended model is good for, how fast it is (Fast, Medium or Slower) and what it costs.

Prices are dollars per million tokens going in and coming out. A million tokens is about 750,000 words.

Solo is one chat with one model. Your chats are listed down the left; “+ New chat” starts another.

Under the message box are that chat's own switches: the model, Masking, Web, Memories, a persona, effort, Attach and saved prompts.

A persona gives the model a character and a way of talking for that chat. 31 come built in, and you can add your own under Chat's Settings → Personas, experts, prompts.

Effort asks the model to answer fast or to be thorough, where the model supports it. Left on default, the model decides.

Saved prompts are pieces of text you use often. Save them under Settings → Personas, experts, prompts, then drop one into your message from the “Saved prompts” list under the message box.

Under a reply: Copy, and “Save to memory” — which files the reply in your Mind, where you can find it, add it to Life, or promote it like any other memory. Each turns green once it has worked.

Under the latest reply there are two more. Regenerate asks the same model again. “Ask another model” sends the same question to a model you pick, so you can compare.

“Edit and resend” under one of your own messages puts it back in the message box to change.

Sending it replaces that message and everything after it. Nothing is deleted until you send; “Cancel edit” backs out.

Under each reply you see the tokens in and out, how many searches it ran, and about what it cost.

When something was masked, “What was sent” under your message shows the text the provider received, and “What the model wrote” under a reply shows it before your details were put back.

Web search is run by the provider on your key, and billed by them per search. Each chat has its own Web switch.

Sources are folded under the reply as “Sources” with a count. Open it to see where the answer came from: each source's title, with the site's name beside it.

Clicking a source opens a Google search for its title, not the page itself. A link inside a reply does the same. The app never opens, or shows, the address a model gave it — if a page is harmful, Google's filtering stands between you and it.

Attach text, Markdown, CSV, JSON, PDF or Word files, up to ten to a message. The text is read on this device, masked and sent as text — the file itself is not uploaded.

Images cannot be masked, so the first time you send a picture in a chat, Chat asks: anything visible in it goes to the provider as it is. “Send anyway” means that chat won't ask again; “Don't ask again” stops asking in every chat, and Chat's Settings → Masking turns the warning back on. A model on your computer never asks, because nothing leaves.

A picture stays with the message it was attached to. Regenerate and “Ask another model” send it again (Chat asks first), and a later message does not carry it — attach it again to ask more about it.

Type /image and a description to make a picture. “Save image…” under it saves it as a file.

“Search chats” above the list searches what was said in your chats. Each chat's row lets you Pin it to the top, Rename it or Delete it.

“Export…” at the top of an open chat saves it as a Markdown file wherever you choose.

Chat's history and saved keys are encrypted with a key kept in your computer's keystore. If that key and the history no longer match, Chat says the history would not open and offers “Start Chat over”.

Starting over moves the history and saved keys into a folder of their own next to them. Nothing is deleted. Chat then starts empty, and you paste your keys again.

Press Ctrl+J (⌘J on a Mac) anywhere for Quick Chat: one question to the model you last used, answered in place.

“Open in Chat” carries the exchange into the Chat tab to keep going.

Before a message leaves, personal details are swapped for placeholders like [NAME_1] or [EMAIL_1], and swapped back in the reply. The provider only ever sees the placeholder.

“Show what will be sent” under the message box shows you the exact text first, with how many details are replaced and which of your memories go with it.

To ask about a public figure by name, open “Show what will be sent” and choose “send as is” for that name. It applies to that one message; “mask them again” undoes it.

Rule packs decide what gets masked. General is on by default; Legal, Finance, Medical and Technical are there for those kinds of work.

Make your own with “+ New rule pack” under Settings → Masking: choose what it masks, add words to always mask, and words to replace with a name of your choosing. Each pack can be set to be on in new chats.

Choose the packs for one chat from the Masking button under the message box.

1. Select the words in your message that should always be masked.
2. Press “Always mask selection” under the message box.

The people you have tagged in your memories are masked by name automatically. Only the list of names is copied over for this.

With “Let the local model spot names and places” switched on in Chat's settings, the on-device model also looks for names, organizations and places that the patterns miss.

If the local model doesn't finish looking, nothing is sent. Chat tells you the names and places it would have found are not masked yet, and asks what to do.

“Try again” runs the check again. “Send anyway” sends that one message without it.

Masking can be switched off: for one chat from the Masking button under the message box, or for every chat under Settings → Masking.

With it off, what you type is sent as typed.

Gmail and Google Calendar content is never sent to a model provider. There is no setting that changes this, and switching masking off does not change it either.

Settings → Agent & Chat → Chat — cloud activity lists every call Chat made: when, to which provider, how many tokens, and what kinds of detail were masked.

The same list is in Chat's Settings → Activity and history.

Chat sees none of your memories to begin with. Nothing is ticked, and the Memories pill under the message box starts off. It takes both steps below before anything is sent.

1. Open Chat's Settings → Memories. Each tier — Me, Memoirs, Memories, Mementos — is a folded card showing how many of its memories are picked. Open one and tick the memories Chat may be given.
2. Open the Memories pill under the message box in a chat.
3. Switch on “Send my memories in this chat”.
4. Choose which of your picked memories go with that chat.

Once the pill is on, it lists the memories you picked for Chat and you choose which go with that chat: All, None, or tick them one by one. A search box finds one in a long list.

Each chat keeps its own choice, so switching chats changes nothing.

“Most to send” under Settings → Memories sets how much goes with each message, in tokens. Picked memories are sent newest first until that much is used.

Pin a picked memory to always send it. A picked memory that is over the size limit is not sent unless you pin it.

“My name and what I wrote about myself” sends your name and the line about you from Settings as well.

A picked memory that came from Gmail or Calendar, or whose origin the app cannot show, is held back, and Settings → Memories says how many.

To give Chat what one of them says, write it as a new memory of your own and pick that.

Experts puts one question to a panel of specialists at once. 16 come built in — a designer, a developer, a financial analyst and so on — and you can pick up to 16 for a panel. One model answers as each of them.

Each answer shows its first ten lines or so; “Show all” opens it. While an answer is being written, its box shows the newest lines and says “writing…”. Follow up one-on-one with any expert, as many times as you like, and fold the thread away with “Hide follow-ups”. Press Synthesize to pull the answers together, or “Start a roundtable” to have the panel debate.

1. Open Experts and press “+ Custom expert” where you pick the panel.
2. Give it a name, a color and an icon.
3. Write its instructions: what kind of expert it is and what advice to give.
4. Press Save.

AI 2 AI has two to five speakers, each on a model you choose, debate a topic in rounds. Each speaker starts with No persona; pick one from its list to give it a character.

Press Play or go a turn at a time with “Next turn”, and jump in whenever you like — the next speaker answers you. “Bring it together” writes a synthesis.

OmniPanel drafts a brief you can edit, has two to four different models answer it side by side, and has a judge give a verdict with where they agreed and disagreed.

1. Pick a model for each seat, two to four, and a model for the judge.
2. Type your question and paste any evidence the panel should weigh; attach text files if you have them.
3. Press “Draft the brief” to have the judge write it, or “Write the brief myself”.
4. Read the brief and edit it if it is off.
5. Press “Run the panel”. The page moves down to the panel, where the answers and then the verdict appear.

The multi-model modes make several calls per question, so they cost several times what a Solo message does.

Settings → Usage adds up what each model was asked and what it cost, for this month, last month or all time.

Four cards across the top give the totals: estimated cost, calls, tokens and average reply time. Below them, each provider has a subtotal and a card for each model: its cost and share of the total, calls, tokens in and out, searches, and average reply time. “By kind of chat” splits the same calls by Solo, Experts, Roundtable, AI 2 AI and OmniPanel.

Settings → Search and budget has a monthly budget. It stops Chat when the estimate reaches it and says so.

The app knows the prices of the recommended models. For any other model, enter its price under Settings → Search and budget → Model prices, in dollars per million tokens from your provider's price page.

Once a model has a price, its calls are counted in Usage and toward the budget.

Chat history is saved on this device, encrypted, in its own store, which is not in the cloud backup and does not sync.

A copy of each conversation also goes into the Mind, like the AI chats the app captures from other services, so you and the Agent can find it later.

Those copies follow the Mind's rules, which means they sync between your computers and are part of the cloud backup if you use it.

Delete one chat from its row in the list.

Delete everything with “Delete all Chat history…” in Chat's Settings → Activity and history, or in the app's Settings → Agent & Chat. That removes every chat, image and the activity log from this device. The copies of those chats in the Mind stay; remove them there if you want them gone.

“Now capturing” lists only what is actually live, grouped by how it gets in: “Apps & tools”, “Watched folders”, “Browser sites” and “Accounts”. If something isn't here, it isn't capturing.

Apps and tools are read on a schedule and whenever you scan by hand. Watched folders are captured the moment a file changes. Browser sites are read from pages you already have open, on sites you authorized. Accounts are signed-in services fetched directly, and their tokens stay on your computer.

“My Memory” sits at the top of the tab — your tiers, search and ask, and a box to add a memory yourself. A feed of the latest captures sits beside the lists.

“Scan now” on an app's row forces an immediate pass and shows exactly what came in: +N captured · N redacted · N dropped. “Scan all” does the whole “Apps & tools” group.

When a scan brings in nothing, the row says why — “nothing new”, a permission it needs, or that the app has to be opened once to set it up.

A green dot means the source is capturing. An amber dot (with a “can't read” label) means it is connected but unreadable right now — usually a permission.

The “Pause capture” button stops ALL capture — the file watcher, scheduled scans, everything — until you click “Resume capture”. It survives restarts.

The Connect tab leads with “Start here” — messaging, email and calendar, photos, social, and AI chats, the places most of a life lives. Everything else sits one click away under “All connections”, grouped by outcome.

The list is filled in by a quick, read-only look at your computer. Nothing is touched until you connect it. Connect something and it moves to “Now capturing”.

The colored dots tell you where each one stands: green is capturing, amber was found on your computer and is waiting for you to connect it, grey was not detected here.

“Importing Tips”, just under the Connections legend, holds nine tips worth a skim before your first big import — chunking giant exports, one folder per service, and why a huge import keeps working in the background after the copy finishes.

The “Don't see what you need?” callout at the top of Connect points to the two catch-alls.

The browser extension covers sites you use in a browser. The Foundry covers everything else — a ready-made recipe, a fetch you record once and reuse, or a watched folder.

Automations is the one place for everything that runs on its own, in three layers: export fetchers and upkeep agents, fetches you recorded yourself in the Foundry, and the app's built-in rhythms.

Results never show here. They land in the capture feed and in Review, same as everything else.

Export fetchers download a data export for you when it is ready — Google Takeout, ChatGPT, your X archive, Meta, LinkedIn, Spotify and Discord. Upkeep agents keep the setup healthy: spotting export archives left in your Downloads folder, noticing a source that has gone quiet, and checking that connected AI tools still point at OpenMindex.

Each one is consent-gated: you switch it on, and it starts by asking first. A clean streak of runs can earn it “auto” mode. Every run is logged — the row shows the last result, “Run now” starts one on demand, and “Explain” describes the last run in plain language.

A fetcher works in its own browser session. When a service shows its sign-in page instead of your data, the run stops and asks you to take over — it never guesses a password and never skips the step.

The X archive fetcher does exactly this whenever X shows its login page. An amber card appears at the top of Review with an “Open sign-in window” button; the Review section of this guide walks through it.

“Recorded by you” lists fetch flows you taught by clicking through a site once. They run on the same schedule and approval rules as the built-in ones, and each shows “approved” or “awaiting approval”.

To record a new one, click “Teach it to fetch”, or search for any service on Connect and choose the same option there.

These are the timers that always ran but never said so. There is nothing to configure.

Source scans run every 10 minutes. Gmail, Google Calendar, Outlook and Linear are checked every 3 hours instead — and Google only every 12 hours on a phone that is paired with a computer.

Memory suggestions are looked for every 5 minutes, and the context files handed to your AI tools are refreshed every 15.

A due backup is checked for at least once an hour. Dreaming runs when your computer is idle. Skill drafts are written once a day, and only if you switched daily discovery on under Context → Skills.

Secrets — API keys, card numbers, SSNs — are scrubbed automatically before anything is stored. You control every rule in Settings → Privacy.

This page is INPUT: the things OpenMindex reads from. The OUTPUT side — giving AI tools access to your memory — lives on the Context page.

There are six ways in, and the source decides which one it uses. You pick what you want remembered; OpenMindex picks the route.

Already on your computer: your mail, notes, photos, calendar, contacts, messages, browser history and dev tools are read straight off the disk, wherever your computer keeps them. No account, no upload.

Sign in: Google, Outlook / Microsoft 365 and Linear connect to your account, and you see the permissions before granting them.

Paste an API key: services like Todoist connect with a key from your account settings and then sync on their own. The Foundry's recipe catalog extends this to hundreds more.

Watch a folder: point it at a notes vault, a synced drive, or wherever meeting transcripts land, and anything new that appears gets read.

Drop in an export: services with no live connection — ChatGPT, WhatsApp, Instagram, Amazon, Google Takeout and more — will hand you your own data if you ask, and the file becomes yours to search.

Browser extension: for the places that offer no export at all.

1. Open Capture → Connect.
2. Find the source under “Start here — where your life lives”, or open “All connections” and click a group to unfold it.
3. Click “Connect” on its row.
4. Do what it asks for: pick a folder, sign in in your browser, paste a key, or choose an export file.

One Facebook data export brings most of a Facebook life in: your posts and check-ins, who was tagged and where, your photo captions by day, comments, reactions, friends, events, Messenger chats and Meta AI chats. Each lands in your Mind on the day it happened.

Photos themselves aren't copied in; their captions and dates are. The pictures stay in the export's folders for you to add to a Moment when you choose.

1. On Facebook, open Accounts Center → Your information and permissions → Download your information, and request an export.
2. Choose JSON as the format — HTML is the default, and an HTML export can't be read.
3. When Meta emails that it's ready, download it (or let OpenMindex fetch it), and drop the zip into the “OpenMindex Exports” folder.
4. Chats marked end-to-end encrypted aren't in that export. In Messenger, go to Privacy & safety → End-to-end encrypted chats → Message storage → Download secure storage data, and drop that zip in too.

An amber dot with a “found” tag means we found the app on your computer. A green dot means it's already flowing in.

Detection is a read-only scan — nothing is touched until you connect.

Anything already connected shows “✓ capturing” on its row. The same row lets you “Disconnect” it, “Reconnect” an account or key that stopped working, watch another folder with “+ folder”, or add a second account with “+ account”.

Whether it is actually working — what came in and when — is answered on the “Now capturing” tab. Status lives in that one place.

Gmail brings in your newest mail first, then fills in the last 90 days, up to 1,000 emails a day.

Older mail can come in through a Google Takeout export.

Accounts you sign in to — Google, Outlook / Microsoft 365 and Linear — are checked every three hours while OpenMindex is running.

Sources read from your computer, watched folders and API-key services are checked about every ten minutes.

Is something missing from every group? The Foundry sits at the bottom of the tab, and “Build your own in the Foundry” at the top jumps to it.

Search for the service, or open “Browse everything” for the whole list — over seven hundred services, A to Z, with a filter box.

Each line says what connecting will involve, and picking one takes the best way in.

Many services in the Foundry connect with an API key and then sync by themselves.

Before anything is saved you see exactly what would sync and which sites it may talk to, and you choose which parts to remember.

Where a service lives at your own address — a company GitLab, your own subdomain — it asks for that address and is then locked to it.

You can also paste a recipe someone shared. “Inspect recipe” shows which sites it talks to and what it would remember before you connect it.

The Foundry still never dead-ends. You can capture the service from your browser, teach the recorder to fetch an export, or watch a folder.

Teaching the recorder means your own browser opens, you walk through the export once, and the steps are saved as a recorded skill. It needs your approval before it runs, and you can schedule it.

The browser extension is also at the bottom of the tab, and “Install the browser extension” at the top jumps to it.

It captures the AI conversations you have in a browser — the part of your day that never touches your computer's disk, so nothing else can reach it.

The Browser Extension section of this guide has the full steps.

Closing the window doesn't stop OpenMindex. It keeps running so capture carries on, and its icon in the menu bar (the system tray on Windows) brings the window back.

Quit properly from that icon's menu when you actually want it off.

Restarting your computer does stop OpenMindex, so first run offers to start it at login. If you skipped that, turn it on under Settings → License & Devices → “This device” → “Start at login”.

Without it, nothing is captured between booting up and opening the app, and nothing warns you — a conversation that was never captured leaves no trace to notice.

Sources that read from somewhere durable — mail, watched folders, your dev tools, your computer's own apps — catch up on their own. Each remembers where it got to, and the first sweep after launch collects whatever piled up while OpenMindex was closed.

Conversations read from a browser page are the exception. Nothing else holds a copy of those, so they can only be captured while the app is running.

The extension isn't in a browser store yet, so you install it by hand from a folder. The section has the exact clicks for Chrome, Edge and Firefox.

1. Open Capture → Connect → “Browser extension”.
2. Click “Show me the folder”. A folder window opens with the “extension” folder highlighted.
3. Open your browser's extensions page and switch on its developer mode.
4. Drag that whole “extension” folder onto the extensions page, or pick it with “Load unpacked”.

Loading the folder installs the extension, but nothing flows yet — pairing is what connects it.

1. Pin the extension's icon to your browser's toolbar.
2. Click “Show pairing code” in the app. An 8-character code appears and lasts 10 minutes.
3. Click the extension's icon in the browser and type the code into the box it shows.
4. Click “Pair with OpenMindex”.

The quickest way is to authorize a site the moment you're on it.

You can also type the address into “Authorized sites” in the app — but that only puts it on the list. Your browser still has to grant permission to read it, and only a click on the extension's icon while you're on the site can do that.

Each site row says which of the two it's waiting on.

1. Open the site in your browser.
2. Click the extension's icon.
3. Press “Authorize this site”.

Every authorized site is listed in the app with its own “Revoke” button.

This list is the real one — the extension only keeps a copy of it. A change reaches each paired browser on its next check-in, within a minute.

The extension captures AI conversations only. It has no way to read a feed, a mailbox or an ordinary page.

If you add a site that doesn't look like an AI chat, the app says so before adding it. For other services, look under Connect, where most have an export you can drop in.

The app and the extension talk over this machine only. The connection never leaves it, and the app refuses anything that isn't the extension.

Your redaction rules and the Pause button apply here exactly as everywhere else: masking happens in the app, not in the browser.

If you later import a data export from a service you'd already been capturing live, the conversations you already have are recognized and skipped rather than stored twice.

The extension ships inside the app, so when the app updates, your browser keeps running the copy it already loaded.

When they differ, this section says so and shows the address of your browser's extensions page. Go there and click the reload arrow, then reload any tab you had open on a captured site.

Until you do, fixes to what gets captured have no effect.

Sites change their page layout without warning, and when that happens the extension stops recognizing conversations there. It captures nothing rather than capturing nonsense, which is the right thing to do but easy to miss.

Each site says when it last captured anything. It also calls it out if it's reading the site but recognizing nothing, or capturing your questions while missing the replies.

“Remove extension” in the app stops OpenMindex accepting anything from it at once, so capture is off straight away.

Browsers don't let an app remove their extensions, so the last step is yours: open your browser's extensions page, find “OpenMindex Capture” and click Remove.

This is your machine, so you decide what counts as sensitive. Every rule on this page is checked before a capture is stored.

Rules apply to the NEXT capture. Nothing already stored is rewritten, so tighten your rules early — and use the Vault to delete anything that slipped through before.

“What's redacted by default” lists the built-in rules with a switch each. Switch one off and that kind of text is stored as you wrote it — useful when it's your own number and you'd rather have it searchable.

Under “Your own redactions”, add a pattern and anything matching it is masked, while the rest of the capture is kept.

“Never capture” is stronger than a redaction: a match there discards the whole capture, not just the matching words. Use it for subjects that shouldn't be recorded at all.

“Blocked paths” goes further still. Files matching a path pattern like */secrets/* are never even opened.

“Try it” shows exactly what would be stored, using the rules as you have them right now — saved or not.

1. Paste a sample into “Try it”.
2. Click “Test rules” and read what would be stored.
3. Adjust your rules until the result is what you want.
4. Click “Save rules”.

“What your rules caught” tallies the captures that were kept with something masked, by kind.

“What your rules refused” lists what was dropped or never read, most recent first — the rule and the file path, never the content. “Clear” wipes that list.

Where a memory had something masked, the Vault shows an amber line on that row: “Removed before storing: …”. You see which kinds were stripped, never the secret itself.

If your saved rules ever can't be read, capture pauses rather than store anything without them, and Privacy shows a red notice. The lists then show the defaults.

A Pause setting that can't be read keeps capture paused too.

1. Check the lists on the Privacy tab.
2. Add back any rules of your own.
3. Click “Save rules” to resume capture.

On the Review tab, OpenMindex periodically reads fresh captures and proposes promotions (“this looks like a durable fact”). It looks every 5 minutes; “Scan now” looks straight away.

The badge on Review counts everything waiting — promotions and dream insights together.

“Approve” stores the suggestion at the tier it proposes. “Modify” lets you edit the wording first, then “Save & promote”. “Reject” declines it.

Some cards say “Add topic tags” instead of naming a tier. Those suggest labels for a memory you already have; approving one adds the tags and moves nothing.

Rejected suggestions are kept forever under “Rejected”, next to “To review”. Click “Restore” on one to send it back for review.

Amber cards at the top of Review are background fetches that need you. The most common one says a fetch needs you to sign in.

1. Click “Open sign-in window”.
2. Sign in to the service in the window that opens.
3. Click “Run again” — it closes the sign-in window for you and retries the fetch.

A card that says a link is missing or expired means you need to request a fresh export from that service. The fetch picks it up from the “ready” email when it arrives.

A card that says the service's page changed is different: retrying won't help, and it needs an update to the fetch itself.

“Your … export is ready” appears when a service's ready email is spotted. “Let OpenMindex fetch it” runs the download; “Open in browser” and “Dismiss” are there if you'd rather do it yourself. A fetcher that has earned auto mode fetches first and tells you afterwards.

Other cards offer to file export archives sitting in your Downloads folder into your import folder, tell you a source has gone quiet (with a button to open Connect), or warn that a connected AI tool's settings look stale (with a button to open Context).

On the Dreaming tab: while your computer is idle, OpenMindex “dreams” — tidies duplicates and stale task notes into the archive, links related memories, distills insights, and prepares your AI tools' context for the morning.

A dream starts on its own after about 15 idle minutes with the computer plugged in, at most roughly once a day. The moment you touch the computer it stops, and picks up where it left off next time. Click “Dream now” to run a pass on demand.

Insights wait under “Insights to review” for your “Add to Memoirs” or “Dismiss”. Nothing is added to your Memoirs without your say-so.

The “Dream journal” logs every action with its evidence — click “why?” on anything.

Archived items have a “Restore” button. Restoring brings the memory back and tells the engine it was too aggressive.

Each type of action earns autonomy separately. One that you approve 10 times in a row graduates to running automatically; one rejection or restore puts it back on probation.

Tidying is applied automatically from the start because it's fully reversible — tidied captures go to the cold archive and are never deleted.

The Conflicts tab is for synced devices. If the same Me or Memoirs entry is edited on two devices independently, both versions appear here and you choose — sync never picks silently when it is about who you are.

Click “Keep this” on the version you want. Or click “Merge by hand…”, write the version to keep and click “Save merged version” — it replaces both, and both originals stay in version history.

The Vault opens straight on its three tabs: “Mind”, “Archive” and “Backup & Version History”.

My Memory — the tiers, search and ask, and add a memory — is not here. It heads Capture and Home → Dashboard instead.

The Mind tab is the whole hot store — everything OpenMindex holds on this device, not just the recent slice. Filter by tier and by source, search by keyword, and copy any row.

Memories you wrote or approved show with their formatting. Raw captures show exactly as they were captured.

Hover a row and click “Delete”, then “Confirm delete”. The system never deletes anything; only you can.

Hover a row and click “Date” to state when something actually happened — an exact day, a month, a year, or a decade. Click “…to” to add an end and make it a span, like a trip or a job.

The editor starts from whatever date the row already carries.

“Add to Life”, on the same row, promotes a raw capture to Memories, tags it life, and opens the date editor — one hover, and an old email becomes a life event.

Tick the “No date” checkbox next to the sort toggle to see only rows without an event date. It is the to-do list for the “Date” action: give a row a date and it joins your timeline.

Hover a row and click “Tags” to label it — family, vacation, work, or anything you invent.

Tags are separate from tiers on purpose. The tier decides how much attention your AI gives a memory; the tag says what it's about; and one memory can carry several tags.

The list is newest-first by when OpenMindex read something (“Captured”). Switch to “Happened” to order by when it actually took place — useful for a timeline, but a synced calendar's future events then sit on top.

The dates in the list always follow whichever you picked. Hover one to see both.

Hover a row and click “Injection” for that memory's per-app switches. “on” inserts it into that tool's pushed file. “pin” does the same and keeps it at the top of its section, never clipped by the budget. “hide” means that tool can't even find it by searching.

“hide everywhere” keeps the memory out of every AI tool, pushed or searched.

Clicking a month bar in Life › Everything opens the Mind tab narrowed to that month. The tier and source filters are off while it's on; click “Show everything” to go back to the full list.

The Archive tab is the cold archive — everything that aged out of the hot store, still searchable forever. Filter by source or search by keyword.

“Restore to Mind” moves an item back. “Delete” removes it for good, and asks you to confirm first.

A backup is one encrypted file holding everything — all five tiers, including Sealed memories, and the cold archive. Once set up, it runs by itself on the schedule you chose (daily unless you change it), keeping the newest 7.

1. Open Vault → “Backup & Version History”.
2. Click “Choose folder…” and pick a folder — an external drive or a synced folder works great.
3. Pick a “Schedule”: every 6 hours, every 12 hours, daily, weekly, or “None — manual only”.
4. Click “Save changes”.
5. Click “Back up now” to take the first snapshot.

Backups are encrypted with a key held in your computer's secure keystore — a stolen backup file is unreadable. That key is why “Restore…” on the machine that made the backup needs nothing.

Your Recovery Key is the one you saved yourself, and it's the one that works anywhere. To restore on a different machine, OpenMindex asks for it.

A snapshot tagged “This computer only” had no Recovery Key available when it was written, so it opens on this machine and nowhere else.

Save a Recovery Key under Settings → Encryption at rest and everything written afterwards travels with you.

Restoring asks first, then stages: nothing changes until you next launch OpenMindex. Your current data is kept as a safety copy.

1. Click “Restore…” next to the snapshot you want.
2. Read the warning and click “Restore this backup”.
3. Quit and reopen OpenMindex to apply it.

1. Copy the .omxbak file across to the other machine.
2. Install OpenMindex there and point its backup folder at the folder the file is in.
3. Click “Restore…” and enter your Recovery Key when it asks.
4. Restart when it says the restore is staged.

“Delete…” removes a snapshot from the backup folder, and asks first. That one cannot be undone — it's gone even if it was your only copy.

Under “Version history”, pick a memory and see its last 10 versions. “Restore” any of them — nothing is lost, because the newer text becomes a version too.

“Pin” the versions you never want pruned; pinned ones are kept forever.

The “Storage” card on the backup tab totals your attachments — how many, how many unique files, and their size on your computer — and lists the heaviest ones.

“Remove file”, then “Confirm”, deletes a file. Removing a file never touches its memory.

Settings → Settings tab → “Take your data” → “Export everything…” writes a dated folder you can open without OpenMindex: the database as an ordinary SQLite file, a JSON file per table, a Markdown file per memory (grouped by tier, ready for Obsidian), your attached files, and a README explaining it all.

Free on every plan.

Take your data is the opposite of a backup: the folder is NOT encrypted, so anyone with it can read everything. You tick that you understand before anything is written, and with App Lock on you re-enter your PIN or passphrase.

On the Apps tab, pick a tool — Claude Code, Codex, Cursor… — and connect it. Every connect shows the exact file change first, and is backed up and reversible.

1. Open Context → Apps and pick the tool.
2. Click “Connect…”.
3. Read the exact change it will make to that tool's settings file.
4. Click “Approve & connect”.

MCP is the live route: the tool asks your archive when it needs something.

Push is the standing route: a small OPENMIND.md file with the memories you chose, kept refreshed, that rides along in every session. “Enable push…” shows both changes before it makes them — the file itself, and the short reference added to the tool's own instructions file so the pushed file is actually read.

Nothing is pushed automatically. Under “What gets inserted”, click “Show memories” and tick each memory you want in that tool's file — from Me, Memoirs, Memories or Mementos. Raw captures in the Mind are never pushed.

“pin” floats a memory to the top of its section and protects it from the token budget. “My name & profile” adds the “who you're talking to” block from your profile.

Scope pills under “What … may see” limit a tool to certain memory labels. With none selected the tool may see everything, and untagged memories are always visible.

The token budget slider caps the size of the pushed file. Pinned memories always fit; the rest fill what is left, newest first, and the meter tells you how many were clipped.

Each connected tool shows when it last actually used your memory. Tools update themselves, and an update can move or rename the settings file OpenMindex wrote into — so if a tool is open and still hasn't reached your memory, the row says so and offers “Reconnect”.

A tool you simply haven't opened is never flagged as broken.

“Exactly what gets injected” shows the pushed document word for word, with its size in tokens.

The “Injection ledger” is the receipt drawer for every write ever made to that tool.

Connecting a tool over MCP issues it a private identity token, written into that tool's settings. Only apps you've connected can query your memory.

“Disconnect” revokes the token on the spot. “Disable push” stops the pushed file.

A skill is a small instruction file that teaches an AI tool a behavior — like a recipe. Click “Install memory skills” once for the built-in ones, or “+ New skill” to write your own. “Draft with AI” works from plain English and drafts on your computer; nothing is written until you click “Save skill”.

Tick a checkbox to deploy a skill to an app. That copies it into the app's own skills folder — with an approval preview every time, and removable any time.

“Skill discovery” looks through your coding-session history for something you've done the same way several times, and writes it up as a draft skill using your local AI. Click “Look now”, or tick “Also look once a day on its own”.

Drafts land in “Skills to review”. Read one, then “Keep” it to add it to your library, or “Discard” it. Until kept, a draft can't be given to any app.

“Install Token Optimizer” adds a skill of habits that cut token spend without changing what gets done — local search first, reuse accepted work, load tools only when needed, match answer length to the ask.

Once it's active for a tool, that tool's row on the Apps tab shows a “Token Optimizer” block, and Analytics shows its impact.

The Lab is a scratch workbench. “Scratch Chat” is a memory-free conversation with an app's context and skills simulated — or with none, to see the model alone.

“Context A/B” runs one prompt with and without an app's context so you can see what the injection buys. “Context Cost” itemizes the token bill of an app's standing prompt, layer by layer; paste a draft prompt to price it on top.

Nothing in the Lab is captured or remembered. No memories are read or written, and the transcript dies when you leave.

Analytics has two tabs. “AI Insights” is about your AI tools: prompts sent, how many tools you use, tokens, usage history, peak windows, model mix, topics and tags, and a “Delegation snapshot” of which tool you lean on.

“Activity Insights” is about everything else: an activity timeline, your active hours, “Capture health” (how much each source brings in and how recently), topics and tags, and “Archive footprint” — what your whole archive holds in each tier and what it takes on disk.

The buttons at the top right set how far back every chart looks: 1, 3, 6 or 12 months. Your choice is remembered.

If your archive is younger than the window you pick, the page says so and shows everything you have — the same numbers as a shorter window. That is the selector working, not a fault.

“Worth a look” spots things you keep re-typing to your AI tools, and tells you how many times you sent each one and across how many tools.

Click “Suggest as a durable memory” to send one to Review. Approve it there and every connected tool just knows it — no more typing it again.

Under “Well-being”, turn on “Mindful nudges” if you want occasional reflective prompts and usage observations in “Worth a look” — for example, a gentle note when a large share of your AI sessions happen between 10pm and 5am.

“Quiet hours” holds those nudges during the hours you choose. Pick a start and an end hour; set either to “off” to switch quiet hours off.

If you installed the Token Optimizer skill (Context → Skills), “AI Insights” adds a “Token Optimizer impact” card: tokens per prompt before and after you activated it, taken from the tools' own transcripts.

It is measured per prompt, so simply using a tool more or less doesn't move the number.

The first time you sign this app in to your account, 7 free days of Pro begin. No card is needed.

It is one trial per account, and it covers everything in Pro.

Home shows how many days are left, and so does the License card in Settings → License & Devices. When it ends you are back on Free and nothing on your computer is lost.

Your license checks in with your account once a day and follows your plan. An upgrade widens what your computer may do and a downgrade narrows it, without pasting a new code.

If a check-in is refused — the device was removed from the account, or the account was revoked — the License card says “Your license couldn't be refreshed” and why. Sign in again from Settings → My account.

Offline, a paid license keeps working for two weeks past its date. A trial does not get those two weeks: it ends when it ends.

Pairing your own computers is Device sync, in Settings → License & Devices.

Adding someone else is a contact, made in Life → Share. A contact never receives your archive, only the specific moments you share.

Every connection tries to go straight to the other device first. When it can, nothing passes through our servers at all, and the device's row in Device sync says “direct”.

When two networks won't let that happen — mobile data, hotel or office Wi-Fi — it falls back to an encrypted relay we run, and the row says “via relay”.

The relay forwards sealed traffic it has no key for. It cannot read anything, either way.

If the top of Device sync says “Same network only — devices won't sync when they're apart”, no relay is set up on your computer, so your devices simply cannot find each other when they are apart.

The relay arrives when your computer signs in, and it works straight away. Sign in from Settings → My account. If it is already signed in, press “Refresh license now” on the License card.

On the same network, no. Your devices and contacts connect straight to each other.

When you're apart, yes, on both sides. The connection then goes through our relay, and the relay only lets in devices that are signed in. Any plan works, including Free.

The chip at the bottom of the sidebar always shows whether your computer is signed in. Click it to sign in or out.

A computer that isn't signed in also says so on the Share and Transfer tabs, and on Home once you have someone to share with, with the sign-in right there.

The “Use the relay” switch under Device sync turns off the internet path entirely, so your devices only meet on the same network.

It takes effect the next time the app launches, and the line at the top of Device sync says so until then.

“Connecting to …”, “Syncing with …” and “Getting files from … — 3 of 7” say what the session is doing right now.

Files are counted rather than shown as a percentage, because a file is the unit you'll recognize in your library.

“… isn't reachable. Trying again in 2 minutes.” means the other device is off, asleep, or on a network that won't answer.

Nothing is lost and you don't need to do anything. It retries on its own, waiting longer each time — from 30 seconds up to half an hour — so one sleeping laptop doesn't hold up everyone else.

“3 files waiting — Mom's laptop has used this month's sharing allowance” means the person sharing has sent more over the internet this month than their plan allows.

Nothing is wrong on your side and nothing is lost. The files arrive when they upgrade, next month, or when you're on the same network.

The sharing allowance is what you send over the internet each month: photos and videos you share, and the copies your devices fetch when they are apart. Direct connections never count.

The line appears under Device sync once most of the month's allowance is used — about 80%. Past it, the moment and its thumbnails still deliver; only the full-size files wait.

Upgrade to keep sending, or let them go when you're on the same network. It resets at the start of the month.

Sharing with people has its own switch on the Life → Share tab. Off means no invites and nothing sent or received; your people and what you've shared are kept.

The Share and Transfer tabs also show when sharing last worked — “Last synced 5 min ago” — with a “Sync now” button that names each person and says what moved.

Photos and videos you share are sent at viewing quality — 1600 pixels on the long edge for photos, 720p for video — never the original.

Your own copy is untouched. It's the version that travels that's smaller.

“Download photos and videos shared with you” under Device sync offers Always, Wi-Fi only, or Never (tap to load).

The moment, the map, the thread and the small previews always arrive whatever you choose. This setting is only about the full-size files.

Anything not downloaded shows “Tap to load”, and a tap fetches that one file from the person who shared it.

A shared photo or video arrives in the same connection as the moment it belongs to, while the person sharing is still online — the moment first, then the file.

If the connection drops part-way, what already arrived is kept, and the next connection picks up from there rather than starting over. A small thumbnail shows until the file is complete.

Sharing something doesn't hand over a copy of your memory. The other person sees the moment you shared.

Keeping it is their explicit choice — “Keep”, or “I was there” to put it on their own timeline too. A kept copy can never be re-shared onward.

Withdrawing a share takes the files with it: the other device stops being served the photos, not just the moment.

Circles live in Life → Share, in the Circles card.

Only people you've already paired can go in one, so invite someone first if the card has nothing to pick from.

1. Click “+ New circle”.
2. Type a name for it.
3. Tick the people to include.
4. Click “Save circle”.

Hover a moment on your Timeline and click Share. Your circles are listed above your people, and ticking one ticks everyone in it.

You can then tick or untick anybody by hand before sending. Unticking the circle unticks all of its people at once.

Click “Edit” beside the circle, tick or untick people, change the name if you like, and “Save circle”.

Two circles can't share a name, whatever the capitals — the app says so and waits for a different name, rather than replacing the other circle.

Click the × beside it, then “Confirm”. “Keep” backs out.

Deleting a circle never unshares anything. What you already shared stays exactly as it was, and the people in it stay in your contacts.

A circle lives only on your computer. It isn't part of device sync, it isn't in your backup, and it isn't in an export.

So a circle you build here won't appear on your phone, and it's gone if you reinstall. The people themselves are safe; it's only the shortlist that's local.

Nobody else ever sees a circle or its name — not the people in it, not us.

What the people you share with see is the moment and who else is on it, exactly as if you had ticked each of them yourself.

Changing a circle never reaches back into anything already shared, in either direction.

Adding your new brother-in-law to “Family” doesn't add him to last summer's moment, and removing somebody doesn't take away what they already have.

A circle costs one share, however many people are in it. The free plan's 3 shares count moments, not people.

Receiving and contributing are free forever.

If you remove someone from your contacts, they quietly stay listed in any circle they were in, and the count on the Circles card still includes them.

Sharing with that circle only reaches the people who are still your contacts. Open “Edit” if you want the count to be right again.

1. Open Settings → License & Devices on the computer and click “Pair a new device…” under Device sync. It shows a code and a token.
2. Open Settings → Sync on the phone and tap “Pair with another device”.
3. Enter the code and the token, then tap “Join”.

On the same Wi-Fi the two talk directly and it's near-instant.

Away from home they meet through an encrypted relay that can't read anything passing through it.

Syncing runs while the app is open and in front of you. A phone in your pocket is not syncing.

On top of that the system allows a short background pass — at most once every fifteen minutes, and on an iPhone only when the system feels like granting it. So the phone catches up on its own sometimes, but never as reliably as opening the app.

Under Sync in the phone's Settings you choose which tiers this phone receives — Me always comes — and whole sources you'd rather leave off the phone, like work email.

Switching one off keeps what's already there.

Ask a question on the phone and it can search your computer too, live, quoting which device an answer came from.

Those quotes aren't copied onto the phone — the raw archive stays on the machine that captured it.

Moments and previews always arrive.

Full-size photos and videos from your own other devices wait for your media setting on the phone, and are held back entirely if the phone has less than 2 GB free.

Wi-Fi only waits for a network the phone itself calls unmetered — Wi-Fi at home or work.

Mobile data, a hotspot from another phone and iPhone's Low Data Mode all count as mobile data. So does a moment when the phone can't tell: it waits rather than spend your data.

Life on the phone has the same Share and Transfer tabs as a computer: share a moment with the people who were there, or send photos and videos straight to someone.

Photos and videos are shrunk to viewing quality on the phone before they go, exactly as on a computer. Your originals stay where they are.

The share sheet has OpenMindex in it, so a link, a photo, a PDF, a note or a voice memo goes straight to your archive.

It lands the next time you open the app.

On an iPhone, the Safari extension puts a one-tap button in the toolbar that saves the page's address and title.

It never reads the page itself — that's the desktop extension's job.

Settings → “Capture from this phone” lists Calendar, Contacts and Photos, each with its own Import button. Nothing is read until you tap one.

A refusal is remembered. The app then tells you to allow it in the phone's own Settings rather than asking again.

Import Photos opens the phone's photo picker: choose the photos to import, up to 50 at a time.

The app asks for no photo permission and never sees the rest of your library.

Photos come in 50 at a time. The first tap takes your newest 50.

Each tap after that brings in photos taken since the last import, oldest first, skipping anything already in — so nothing in between is missed and nothing comes in twice.

It stays a decision, not a silent sweep: older photos from before your first import are not pulled in this way.

Connect online accounts in Settings → “Live accounts”, the same way as on a computer: you sign in once in your browser, and new mail, events and work land each time you open the app.

Watched folders and the connector catalog stay on the computer.

1. Open Life → Import.
2. Tap “Choose a file” and pick a CSV, or a .ged, .gedcom or .gdz tree from the phone's files.
3. Review it row by row, as on a computer.

Recording straight onto a moment needs the phone's camera or microphone permission, which the app asks for the first time you try.

Search on a phone matches words rather than meaning.

A question phrased differently from the memory may not find it there, while the same question on your computer would.

1. Open Settings → Account and type your email.
2. Tap “Email me a code”.
3. Type the six-digit code from the email and tap “Sign in”.

On an iPhone, a plan can be bought in the app, and “Restore Purchases” brings back one you already own. Prices there are set by the App Store and run a little higher than the website.

On Android there's no purchasing at all — subscribe on the web or on a computer and simply sign in here.

One plan covers every device on your account, in both directions: buy on the web and the phone unlocks; buy on the iPhone and your computer unlocks too.

Cancelling an App Store subscription is done in Apple's own Settings, under your name → Subscriptions, at least a day before it renews.

Turn on notifications when the app offers, or later from the Notifications card in Settings. They're off until you say so.

They only ever carry counts and titles, never anything you wrote.

“Ask for Face ID or fingerprint to open” in the phone's Settings puts the phone's own biometric check in front of the app. If you set an App Lock PIN, the lock screen offers “Use PIN instead”.

On an iPhone that check also guards the key to the archive itself. On Android it guards the app, and the key stays usable underneath — so treat Android's lock as a door, not a safe.

Encrypting the archive on a phone walks you through the Recovery Key first, and it isn't optional: a sealed archive on a replacement phone opens with that key or not at all.

Android phones are also excluded from the system's own device backup deliberately, which makes the Recovery Key the only way across.

Some things stay on the computer: exporting your whole archive, watching folders, the browser extension, and updates (a phone updates through its app store).

The app never nags you to upgrade on a phone. A locked feature simply says it's locked.

The first question after you open the app is slower than the rest — the model is waking up.

On a phone with less memory the app runs in a limited mode: the smallest model, leaner answers, and it says so rather than pretending.

An answer that fails with “local model unavailable”, or a search that finds nothing new, means the model engine on your computer is off.

Click “Start local engine, then ask again” under the failed answer on the Agent & Chat page, or relaunch the app — it recovers automatically.

The first answer after a quiet spell is slower because the model has to wake up first. The Agent says so while you wait: “Thinking and waking up may take a little bit longer.”

If every answer is slow, check whether “Think harder” is switched on beside the question box. It sends questions to the deep-thinker model, which is smarter and slower.

Open Context and find that app's card. A stale reference shows a “repair” link beside it.

“Exactly what gets injected” proves what the tool receives.

Most of the time this is a permission problem on the source's folder.

Open Capture → Now capturing, press “Scan now” on that source and read the error it reports.

On Capture → Now capturing, a green dot means the source is capturing. An amber dot, with a “can't read” tag, means it is connected but can't be read right now.

Press “Scan now” on that row to see why. It is usually the same thing as above: the folder moved, or the app lost permission to read it.

On Capture → Connect, a connected account can pick up an amber tag: “needs re-auth” or “sign in to keep” means its access stopped working, and “not syncing” means it has been failing for days.

For the first two, sign in to that account again. A “not syncing” connection keeps retrying on its own; hover the tag to see the last error.

A scheduled fetch — a Takeout or ChatGPT export, say — sometimes finds itself signed out. A card on the Review page says “… fetch needs you to sign in”.

Signing in with your regular browser won't help: the fetch uses its own browser window, with its own sign-ins.

1. Click “Open sign-in window” on the card.
2. Sign in to the service in the window that opens.
3. Click “Run again”. It closes the sign-in window for you.

A card on the Review page reading “…'s page changed” means the fetch couldn't find its way — the service likely rearranged its website.

Retrying won't fix this one, which is why the card has no “Run again” button. The fetch itself needs an update.

A card reading “Last … fetch failed” is different: that one shows what went wrong, and “Run again” is worth a try. Either card disappears once a run succeeds.

Open Settings → Privacy and switch that built-in rule off, or use “Try it” to see which pattern fired.

The change applies to the next capture. Earlier memories keep their mask.

1. Click “Forgot it? Use your recovery code” on the lock screen.
2. Enter the 8-word code you saved when you set up the lock.

“Your memory archive is locked” at launch asks for your Recovery Key. Enter it and you're back in.

This is expected after a keystore reset, or when the app's data folder moves to another machine. Capture stays paused until you're back in, and nothing is lost.

Start in Settings → License & Devices → Device sync. Check the switch at the top is on, then read the line under each device.

“… isn't reachable” means the other device is off, asleep or on a network that won't answer; it retries by itself. “Same network only” means there is no relay, so devices only meet on the same Wi-Fi.

A phone only syncs while the app is open and in the foreground, so open it.

There are two kinds of invite, and each only works in its own place.

An invite for one of your own devices goes in Settings → License & Devices. An invite to share with a person goes in Life → Share. Paste one into the wrong box and the app says which it is and where to take it.

If the Share tab says the list of people you removed or blocked can't be read, Sharing stays off on purpose — the app won't guess who you had blocked.

Restoring from a backup brings the list back. “Start a new list” sets the damaged one aside instead, and anyone who was on it is no longer blocked.

“Your license couldn't be refreshed” on the License card means the daily check-in with your account was refused — usually because your computer was removed from the account.

Sign in again from Settings → My account. Pasting a fresh activation code into “Enter license…” works too.

The target folder must exist and be writable.

Pick it again with “Choose folder…” in Vault → Backup & Version History.

Open Vault → Backup & Version History for a single memory's versions, or restore last night's snapshot.

Both are non-destructive.

“Contact support” at the bottom of this page opens your account page at openmindex.ai, where you can open a support ticket.

We reply there and notify you by email.

The Settings tab holds your profile, Display size, App Lock, Encryption at rest, “My account”, capture rules, notifications and the memory lifecycle.

Below those are “Take your data”, “Delete by source”, Delete All and “Uninstall OpenMindex”.

Your profile feeds pushed context: your name and “About” line are included in the context files pushed to your AI tools, and the email is the default address for digests.

Text too small on a big screen? Display size (Settings tab, desktop only) offers 75%, 90%, 100%, 110%, 125% and 150%.

The whole app changes together — text, icons and spacing — so nothing lands out of place. The bigger sizes fit less on screen at once.

The zoom keys work too: hold Ctrl (⌘ on a Mac) and press + to grow, - to shrink, 0 for 100%.

Capture rules tell OpenMindex what matters, in your own words — “always remember pricing decisions”, “never promote diary entries”.

Rules steer what gets proposed. Promotion still needs your click in Review.

“Real-time alerts” sends a system notification when reviews or skill drafts need you.

“Digest” is a periodic summary, every day or every week. It is counts only — never memory content.

“Email digest” sends that summary through your own mail account. No OpenMindex cloud is involved.

The memory lifecycle knobs decide when raw captures get summarized and archived: how old a session is before it's summarized, how many days before raw captures move to the cold archive, and the most raw rows to keep hot.

“Run now” does a pass immediately.

“Take your data” exports everything to a folder in open formats: a SQLite database, JSON, a Markdown file per memory, and your attached files. It all opens without OpenMindex.

Not included: passwords, keys and sign-in tokens, deleted memories, and the search index.

1. Open Settings → Settings → “Take your data”.
2. Click “Export everything…”.
3. Tick “Include Sealed memories” only if you want those written out too.
4. Tick “I understand this copy is not encrypted”.
5. Click “Choose folder and export”.

Decided a source isn't worth keeping? “Delete by source” removes its whole past history — archived rows included — from your computer, leaving every other source untouched.

Back up first — there's no undo here.

1. Open Settings → Settings → “Delete by source”.
2. Tick the sources you want gone.
3. Click “Delete selected…”.
4. Choose what capture does afterwards: “Keep capturing” or “Stop capturing”.
5. Type the word delete to confirm.

Delete All, just below “Delete by source”, does the same for everything captured on your computer: the whole Mind, the cold archive, version history and derived data. App settings and connections are kept so the app still runs.

Both deletes work on your computer only. Your other devices keep their copies, and your Me, Memoirs, Memories and Mementos come back from them on the next sync. Raw captures don't.

To delete something everywhere, do it on each device.

By default capture keeps going, collecting only new activity. The past you deleted isn't collected again.

Or choose “Stop capturing”. A stopped source that you reconnect later goes all the way back and collects its whole history again, including what you deleted.

Connected accounts and API-key sources (like GitHub, Linear or Google) keep capturing either way — disconnect those on the Connect page. Reconnecting one of those later picks up where it left off; it doesn't collect the past again.

The Privacy tab holds every redaction rule — what's masked, what's dropped, what's never read. It also shows what your rules caught and what they refused.

See the Privacy section of this guide.

Your account — billing, magic-link sign-in, device seats — lives at openmindex.ai. There are no passwords: you sign in with a link sent to your email.

“Open my account page” on the Settings tab's “My account” card takes you there. So does “Open account page” on the License & Devices tab.

Signing in on the website signs in your browser, not this app — the app needs its own sign-in.

Once signed in, the app works without internet; it checks in with your account once a day.

1. Click “Not signed in” at the bottom of the sidebar — or open Settings → My account and click “Sign in”.
2. Type your email and click “Email me a code”.
3. Type the six-digit code from the email and click “Sign in with this code”.
4. Restart the app when it asks, so it can sync when you're not on the same network.

The free trial belongs to your account, not to a device. Sign in to your account and activate this device, and you get 7 days of everything in Pro, with no card.

Home shows how many days are left, and so does the License card: “Pro trial”, with the days left and the end date.

When it ends, everything on this device stays. Pick a plan on your account page to bring Pro back.

Everything core is free forever and runs without an account, on one computer.

Pro is the share plan: share moments of your own, and keep two devices in sync.

Pro Plus adds your coding AI: Claude, Codex and Cursor know you, with the Context and Analytics pages and your coding sessions in your Mind. It syncs up to four devices.

Ultra syncs up to ten and adds online backup.

Prices live on the website's pricing page, never in the app.

Device sync is on the License & Devices tab. Pairing is free; syncing needs a paid plan or the 7-day trial on your account.

1. Open Settings → License & Devices → “Device sync” on the first device and click “Pair a new device…”.
2. Open the same card on the other device and click “Join from another device…”.
3. Enter the pairing code and paste the pairing token shown on the first device.
4. Click “Pair”.

What syncs is your identity tiers — Me, Memoirs, Memories and Mementos — encrypted peer-to-peer, direct wherever possible.

Private and Shared doors only. Sealed never leaves your computer.

Under Device sync you choose which of those tiers your computer receives. Me always comes along.

Switching a tier off keeps what's already here and still lets deletions through. Switching it back on fetches what was skipped on the next sync.

Below that you can leave out whole sources — say, work Slack on your phone — in every tier except Me. The same keep-what's-here and fetch-on-undo rules apply.

Raw captures don't copy between devices. Each machine keeps its own archive and the Agent searches all of them live, telling you if a device couldn't answer.

If the same Me or Memoirs entry changes on two devices independently, it appears on Review → Conflicts and you choose. Sync never picks silently.

The Device sync switch is only for your own devices. Sharing with people has its own switch on Life → Share, and turning sync off never stops it.

Backups are separate and bigger: a backup always contains everything, all five tiers.

“Export archive…” writes your syncable memories to an encrypted file, protected by a passphrase of at least 8 characters. “Import archive…” merges one in on another device.

It's the fast way to seed a new machine before pairing. Sealed memories never export.

Cloud backup (License & Devices tab, Ultra plans) is an encrypted off-site copy for when the computer itself is the thing that fails.

It has its own Recovery Key ceremony — save that key too. The copy is encrypted before it leaves, so we can't read it or restore it for you.

“Start at login”, on the “This device” card of the License & Devices tab, opens OpenMindex quietly when you sign in, so capture keeps running from the menu bar (the system tray on Windows).

Right-sized local models share the work by role: everyday (always on, tiny), reviewer (reviews and the Agent) and deep-thinker (deep synthesis only), plus a search model for meaning-based lookup.

First run only downloads the reviewer model (and its thinking partner, where it has one) and the search model — enough to use the app. The bigger ones are here whenever you want them.

On a 16 GB computer the reviewer defaults to the compact Qwen 3 4B Instruct, with Qwen 3 4B Thinking beside it as its thinking partner for questions that need reasoning.

On 8 GB the everyday role also defaults to Qwen 3 1.7B, so the models fit.

Each role has a picker: assign any installed model, or reset to the default. Your choices are saved on your computer.

“Get more models” downloads hand-picked candidates. Models too big for your computer stay visible with the download disabled and a plain reason.

Each role card has a “Measure” button. It loads the model and runs a fixed test of about 30 seconds, then shows real speed and memory measured on your computer.

Results are kept per engine version, so you can compare before and after an engine update.

“Installed on your computer” lists every model, with a “Remove” button on the ones OpenMindex installed. Remove one to reclaim disk, and re-download it any time.

The work on your memory — reviews, search, the Agent — runs on local models only. Cloud models are deliberately unsupported there.

Bringing your own local model is live: install one in the local engine and assign it to a role.

The first question after you open the app is slower than the rest — the model is waking up.

On a phone with less memory the app runs in a limited mode: the smallest model, leaner answers, and it says so rather than pretending.

“Personality” sets how your Agent talks to you: its name, tone, answer length and your own instructions. It changes the Agent's voice, not its rules.

Below it are on/off switches for every Agent feature: acting on source memories, saving answers as memories, quick ask (Ctrl+K (⌘K on a Mac)), the ask box on Home, suggested follow-ups, think harder, the overnight brief, and suggested skill runs.

The What's New tab lists what changed in each release, with the version you're running marked.

It is read straight from the app — nothing is fetched to show it.

OpenMindex checks for a new version on its own and shows a banner when one is ready. You choose when to install.

The version you're running sits at the bottom of the left column, and turns into an “update ready” link when one is waiting.

To ask on demand, click that link, press “Check for updates” on the What's New tab, or choose “Check for Updates…” in the OpenMindex icon's menu in the menu bar (the system tray on Windows).

App Lock asks for a PIN or passphrase before OpenMindex opens on this device.

It is a privacy curtain for the screen, not encryption — capture, dreaming, and sync keep running while locked. To protect the archive file itself, see Encryption at rest below.

1. Open Settings → Settings → “Security — App Lock”.
2. Click “Set up App Lock”.
3. Choose “PIN” or “Passphrase” and enter it twice. Either needs at least 4 characters.
4. Click “Enable lock”.
5. Save the 8-word recovery code it shows you, then click “I've saved it”.

On the lock screen, click “Forgot it? Use your recovery code” and enter the eight words.

The recovery code unlocks the app and turns the lock off, so you can set a new one.

While the lock is on, pick an auto-lock timeout: after 30 minutes idle, 1 hour, 4 hours, or never while the app is running. It starts at 30 minutes.

Click “Lock now” to lock at once. The same command is in the OpenMindex icon's menu in the menu bar (the system tray on Windows).

“Change…” and “Turn off…” sit beside “Lock now”. Both ask for your current PIN or passphrase first.

Your connected AI tools keep working exactly as before; everything else sees noise.

1. Open Settings → Settings → “Security — Encryption at rest”.
2. Click “Encrypt my archive…”.
3. Save the Recovery Key it shows in your password manager or on paper.
4. Tick “I've saved this key somewhere safe.”
5. Click “Encrypt now”.

A byte-for-byte plaintext copy of the old archive stays behind as your rollback.

You don't have to manage it: the next time OpenMindex starts and the encrypted archive opens cleanly, that copy is deleted for you.

Use “Delete plaintext copy…” if you'd rather clear it right away.

This can happen after a keystore reset, or when the app's data folder is moved to another machine.

OpenMindex asks for your Recovery Key before anything opens. Capture pauses until then; nothing is lost.

Two keys open it: this device's key in the system keystore, and your Recovery Key.

Lose both and the archive itself is unrecoverable — that's what encryption means.

Want a real gate at launch? Once the archive is encrypted, the same card shows “Require passphrase at launch”. Click “Turn on…” and choose a passphrase of at least 8 characters.

From the next launch the key stays sealed until you type your passphrase. Your Recovery Key still works if you forget it.

It covers the hot store and the cold archive.

Device sync and the command-line tool open the archive keyed, like the app does.

Backups are a separate system with their own key, and they were always encrypted — turning this on doesn't change them.

So if you lose both ways into the archive, a snapshot still gets your memories back. A snapshot carries the cold archive as well as the hot store, so nothing you aged out is left behind.

To open a snapshot on a different machine you need your Recovery Key — so save one before you need it.

It protects the file, not a running app. While OpenMindex is open and unlocked, anything you've connected can read what you allowed it to.

OPENMIND.md files pushed into your AI tools are plain text by design — that's the point of them.

Encryption is also per-device: each machine has its own key.

Models OpenMindex downloaded have a “Remove” button. Removing frees the disk immediately, and you can re-download from the same page any time.

1. Open Settings → Models.
2. Find the model under “Installed on your computer”.
3. Click “Remove”, then confirm.

This removes the models OpenMindex downloaded, clears this device's OpenMindex keys from the system keystore (the Keychain on macOS, Credential Manager on Windows), deletes the app's entire data folder (memory archive, settings, the bundled engine), and quits.

It is irreversible unless you exported a backup first.

1. Open Settings → Settings and scroll to “Uninstall OpenMindex” at the bottom.
2. Click “Uninstall…”.
3. Type the word uninstall to confirm.
4. Click “Uninstall everything”.
5. Remove the app itself: on macOS drag OpenMindex.app from Applications to the Trash; on Windows remove OpenMindex from Settings → Apps.

Reinstalling later is a true fresh start: first-run setup returns and the models download again.

For the record, the data folder lives at ~/Library/Application Support/ai.openmindex.app on macOS and %LOCALAPPDATA%\ai.openmindex.app on Windows.

The shared model store is the .ollama folder in your home folder. Never delete the model store wholesale if you use Ollama yourself.

Your account and any license at openmindex.ai are untouched — uninstalling a device doesn't cancel anything. Manage billing from your account page.