Cards: pulling a memory out as one shareable page

Memories in AiAkiv are yours alone. Showing one to someone meant copying and pasting it every time. Cards bridge that. Tell your AI "make a card on this topic" and it produces one public page organizing your memories, and hands back one address. Paste that address into any messenger, Slack, or blog and it unfolds as a card.

In one line. A card is a one-page public page your AI writes from your memories. It goes on the web the moment it's created, and anyone with the address sees it without signing in. Management (copying the address, images, deleting, search visibility) is in Console → Data → Cards.

This page is read by both people and AIs. §1–§7 are for people, §8 is for the AI. An AI may read §8 first, but must read §5 (things you must know) with it.

1. What a card is

One card holds these.

The recipient does not need an AiAkiv account. No sign-in. It's just a web page. Alongside the card page, two images (landscape and square) are created, with the card's address engraved in them.

The address is short:

https://card.aiakiv.com/kx7mabcd23

Ten characters, deliberately. In places like Instagram where links aren't clickable, someone has to read it off the screen and type it. Easily confused characters (i l o 0 1) are excluded entirely.

The roles of the card and the body

With only a card, clicking through shows the same summary you already saw in the preview. There's no reason to click. Add a body and the card becomes the hook, the page the substance. So the reading order is summary → body → conclusion. Without a body it behaves exactly like a card alone.

It is not a blog. No post list, no comments, no subscriptions, no author profile. Making one at a time and sharing one at a time is the whole scope of this feature.

2. Creating: you tell your AI

Cards are made only through an AI. There's no create button in the console. Organizing content is the AI's job, and the console is where you manage what was made.

Say this to an AI you've connected AiAkiv to (Claude, ChatGPT, Codex, Gemini…):

"Find 〈topic〉 in my memory and make it a card. Write a body too."

The AI does three things.

1. Searches your memories and gathers content on that topic
2. Writes the title, summary, conclusion (and body) itself, the AI writes, and the server never touches the text
3. Hands it to the card service to be baked → an address comes back

You decide what goes in. The AI decides how it's organized. If you don't like it, ask again (but read §5's "asking again makes a new card" first).

Asking well

• Keep the topic narrow. "Why AiAkiv is a graph" makes a better card than "an overview of AiAkiv." The summary is only 3–5 lines, so a broad topic goes blurry.
• Say whether you want a body. "Write a body too" / "card only". A body is what gives someone a reason to click.
• Set the search terms. Tags are the search terms (§4). Saying "tag it memory-graph and team-sharing" means you can find it later by those words on card.aiakiv.com.
• You can build one from public memory. Even with no memories of your own, you can make a card from public memory you can read.
• To see a draft first, say "just show me the content first." The AI shows a draft and creates it after you confirm. Since it's public the moment it's made (§5), that one step earns its keep on sensitive topics.

What comes back

• A notice that it's public: the AI says this first ("this card is already live on the web …"). It isn't boilerplate; it's the state right now
• The address: https://card.aiakiv.com/……, ten characters
• Two image addresses: landscape (…/og.png) and square (…/square.png)
• (If applicable) a truncation notice: if the summary was too long and got cut in the image, the AI tells you which lines. Then you decide whether to shorten and remake it, or leave it

3. Posting: where it goes and how it looks

Where to get the images

In Console → Data → Cards, each card has Images › Square / Landscape links. They open in a new tab; save from there. Opening the image addresses the AI gave you when creating is the same thing.

• Square: for Instagram. A line at the bottom carries the address and search terms legibly
• Landscape: for boards and blogs. The same picture used in link previews

The address is engraved in the image, so even where links don't work, a viewer has a path back to the card. Modern phones let you tap and copy text inside a photo, which makes that path wider than you'd think.

Worth knowing: an unfolded card is one link, whole

The card a messenger or Slack unfolds is that service reading our information and redrawing it their own way. You can't tap individual text inside it. anywhere you tap goes to the card page. Links you put in the body become clickable after entering the page. That isn't a defect, it's the intended flow: preview as hook, click as conversion, body as destination.

Preview support differs per service and changes with their redesigns. The first time you post somewhere, check once.

4. Managing: in the console

Console (app.aiakiv.com) → Data → Cards.

The same notice always sits at the top of the list. "the cards here are already public." There's also a collapsible how to make a card guide (a copyable prompt, the format limits).

Search visibility: the easily misread setting

A card is public even with search turned off. Off means "not findable", not "not visible". Anyone with the address sees it any time.

Turning search on additionally makes it findable by anyone at card.aiakiv.com. Search matches on title and tags. A card made to share in one Slack channel turning up in a stranger's search isn't the "public" its maker expected, so the default is off and you must turn it on explicitly. You can toggle it, so you can decide later.

Toggling happens only in the console. Asking an AI to "make it searchable" won't work. It isn't something to decide at creation time but a setting you change your mind about later, so it's kept there.

Public nickname: the name behind "all cards by this person"

Search results show your nickname as the author of your cards. Clicking it shows only your cards, and you can share that screen's address directly.

• You get a name whether you set one or not. Creating your first card assigns something like anonymous-473812 automatically, because without a name cards don't group by author in search. Change it in the console if you don't like it
• The number is random, not sequential. A sequential one would tell anyone who saw anonymous-7 that the service has seven users
• You can't pick an anonymous-… name yourself. Claiming an unused number early would block whoever is later assigned it
• The name only goes into the search listing. The card page carries no author, so the name only travels with cards that have search visibility on
• Your email is never made public under any circumstance. The search index file can be downloaded by anyone without signing in, so the only value in it that points at a person is the nickname
• No language restriction. Korean, Japanese, Arabic all work. 2–20 characters; spaces are stripped
• First come, first served. A name already in use can't be taken
• Clear it and save and it comes down. But the next card you create gets an automatic name again
• ⚠ Changing the name breaks any "all cards" address you shared under the old one. Card addresses are unaffected

Category tags: drawers you create

Two kinds of tag attach to a card, which is easy to confuse. They're different.

Two steps to use them.

1. Make a drawer: at the top of the console's card screen, create a tag like design or retro (up to 50)
2. Attach it to a card: press Categorize next to a card and pick from what you made (up to 5 per card)

An attached tag shows at the bottom of the card page, and clicking it collects cards with the same tag. The card image does not change: a tag edit shouldn't alter a picture you already spread around.

Deleting a drawer doesn't remove it from cards it's already on. Detach it from that card's category editor directly. (Rebaking every card on each delete would mean a long wait for anyone with many cards. Instead: "new attachments only from what you made, existing ones detachable any time.")

Three ways to find: and the screen is the address

card.aiakiv.com has three fields. They combine.

Each time you narrow, the address bar follows. There's no "pin" button to press. Copy the address of the screen you're looking at, send it, and the other person sees the same screen. The copy this screen's address button on the page does the same in one click.

5. Things you must know

It's public from the moment it's made

A card is on the web the instant it's created, and anyone with the address sees it without signing in. Posting the address makes it findable, not public.

The address is hard to guess, so in practice only you know it, but that isn't "private". This distinction matters enough that the AI tells you it's public right there when it makes one, and the same line always sits at the top of the console's card screen.

So it's worth reading it once after creation to check that personal content, information not yet ready to be public, or other people's names and emails didn't get in. The AI pulls from your memories, so unintended things can come along. On sensitive topics, ask for a draft first (§2).

Delete and the page comes down immediately: but caches may linger

• Delete a card and the page and images come down immediately. Not reversible
• However, preview caches held by messengers and Slack can linger for a while. Those belong to that service and we can't clear them
• What someone already saved is out of reach

"Publishing can't be undone" means exactly this.

Asking again makes a new card

Say "fix the content" and the AI makes a new card. The previous one is still public and still counts against your quota. After fixing, delete the old one in the console. That's why the AI tells you to (the AI cannot delete).

There is no "edit card" feature right now. So the goal is getting it right the first time.

The summary can be truncated

The 3–5 summary lines share a fixed vertical space in the image. So writing all five long lines will be cut in the image even within the per-line character limit. When it's cut the AI tells you ("lines 3 and 5 were cut, shorten to N characters"). Then you shorten and remake (= a new card), or leave it. The page shows the full text; only the image truncates.

Deleting your account doesn't remove your cards

Closing your account doesn't automatically remove cards you already shared. That's to stop links other people posted from suddenly dying. Cards are also anonymous. They contain none of the creator's name, email, or original memories. If you want them gone, clean up in the console before closing your account; if you've already closed it and a card needs to come down, ask the operator ([email protected]).

Limits

The limits exist to prevent abuse, not for cost. They keep a public domain from becoming a scratchpad. The values are provisional and will be adjusted based on real use.

6. How it works overall

[Person]  "make a card on 〈topic〉"
   │
[AI]    ① searches my memories (with my permissions)
        ② writes the title, summary, conclusion, body  ← the AI writes
        ③ hands it to AiAkiv: "make this in the card app"
   │
[AiAkiv]④ verifies who's creating · issues a short address · records only "whose"
        ⑤ forwards it to the card service
   │
[Card service] ⑥ checks format only (lengths, counts), never judges content
              ⑦ bakes page HTML + 2 images from a fixed template
              ⑧ uploads to public storage
   │
[AI]    ⑨ delivers the "it's public" notice + the address to the person
   │
[Anyone] visits the address → storage returns the page as-is (no sign-in, no code)

Properties worth knowing about this flow:

• The AI writes; the server doesn't. The card service never calls an AI model. So the summary's quality is your AI's quality, and if you don't like it, asking your AI again is the right move. That's why there's no "AI summarize" button in the console.
• The server bakes. The AI only sends card data (title, summary…). HTML and images are always produced by the server from a fixed template. So arbitrary code can't ride into a card page, and if we change the design later, existing cards can be rebaked in the new shape.
• It's cut off from the original memories. A card holds no reference back to them. It's a self-contained snapshot. Delete the original and the card remains; cards are deleted separately. Conversely, deleting a card leaves memories untouched.
• There's no code on the public path. The card page is returned as-is by public storage. No sign-in, no database, no server code in between. So even if the public page were breached, what leaks is "cards that are already public."
• Who made it isn't published. Ownership is recorded only inside AiAkiv, and no account information exists in the card's address, page, or images. One thing is visible though: cards sharing the first 4 characters of the address belong to the same person (a structure that lets an operator clean up one person's cards at once).
• Link previews use the standard mechanism. The card page carries Open Graph meta (title, description, image) and oEmbed. Messengers, Slack, and Notion read those to unfold the card.
• Search is a static index. Only cards with search visibility on go into the index file, and the search page filters it in the browser. Toggling rebuilds the index.

7. Common questions

Q. I turned search visibility off: why is it still visible? Search visibility is about findability, not visibility. To make it invisible, delete it (§4).

Q. I asked my AI to delete a card and it says it can't. Correct. An AI can only create; it can't delete or change search visibility. Do it in the console. Taking down something public is a decision left to a person confirming it on screen.

Q. The summary got cut in the image. Several summary lines share one space in the image. The AI tells you which lines were cut when it creates. Shorten and remake (§5), or leave it, the page has the full text.

Q. Someone else's name ended up in a card. Delete it now. The page comes down immediately. Caches can linger for a while, so if you already shared it, know that.

Q. Can I get back from a card to my memory? No. A card is a snapshot cut off from the original. The recipient isn't an AiAkiv member and has no permission to read your memories.

Q. Can I make a card from team memory? If you can read it, yes. But a card is personally owned by its creator and its publication scope is the whole world. You're putting the team's memory outside, so check it fits your team's rules first.

8. For the AI

This section is read by an AI connected to AiAkiv. It covers what to do, in what order, when a person says "make me a card", and what you must always say.

Tools

Cards are not a dedicated tool but one shared app endpoint.

run_aiakiv_app_action(app="card", action="describe")            # ask for the format
run_aiakiv_app_action(app="card", action="create", data={...})  # create

• action is only describe and create. Deleting, editing, and changing search visibility cannot be done with this tool. Direct the user to the console (Data → Cards).
• The public nickname and category tags can't be done here either. Both are set by a person in the console (§4). To "add a tag" or "put my name on it", do not say you can: give the console location. data.tags is a different thing: those are the card's own tags, engraved into the image and set once at creation.
• data is a JSON object or a JSON string. If the body is long and multi-line, the argument can get mangled. Send a string in that case.
• If you're unsure of the format, call describe first. Don't guess.

Procedure

1. Search the memories: gather the topic with search_memory (Searching). Don't invent what isn't in memory. Put the actual number of memories used in source_count.
2. Filter out sensitive content: personal names, emails, internal addresses, decisions not yet public. A card is public the moment it's made. When in doubt, show the user a draft first.
3. Write to the format: the fields and limits below. In particular, keep the summary within the image safe line.
4. Call create.
5. Deliver the response's notice to the user before anything else. Then the address, then the summary. Do not shorten it.
6. If there's a fit, report exactly what was truncated. Whether to remake is the user's call. Say alongside it that remaking creates a new card and leaves the old one.

Fields and limits (same as the describe response)

Title, summary, conclusion, and tags are single-line fields. Newlines in them collapse to spaces. Only the body treats newlines as syntax.

Syntax allowed in the body: paragraphs, ## heading, - list, 1. list, > quote, `code`, ``` blocks, **bold**, *italic*, [text](https://…), ---. Anything else comes out as literal text. Only http(s) becomes a link. HTML put in is not executed as code.

The summary safe line: staying within the limit still truncates

describe's bullets.renders gives measured values. Currently:

• safe_chars_per_line: 30 characters (worst case for Korean; English and spaces fit more)
• max_lines_each: 3 lines

Several summary items share one space. Writing five items at 90 characters each respects the 120-character limit and still truncates everything in the image. Safe targets:

• 5 summary items → about 30 characters each (1 line): or
• 3 summary items → 60–90 characters each (2–3 lines)

These values were measured, so don't memorize them. Read them from describe. The goal is getting it right the first time. Remaking isn't free.

Reading the response

Success:

{
  "notice": "TELL THE USER BEFORE ANYTHING ELSE: this card is already live … ",
  "ok": true, "app": "card", "key": "kx7mabcd23",
  "url": "https://card.aiakiv.com/kx7mabcd23",
  "og_url": "…/og.png", "square_url": "…/square.png",
  "title": "…", "tags": ["…"], "revision": 1,
  "fit": { "fits": false, "bullets_truncated": [3, 5], "hint": "…" }   // only when truncated
}

• notice: deliver it first, verbatim. It states that the card is now public and how to remove it (the console). Even though it's in the tool description and in this document, saying it again from that specific response is the rule.
• url: show it literally, as a clickable link. Don't shorten or alter it.
• og_url / square_url: images for Instagram and boards. When the user asks for images, point at these addresses or the console's "Images".
• fit: if present, something was truncated. It holds bullets_truncated (ellipsized item numbers), bullets_dropped (items omitted entirely), title_truncated, conclusion_truncated, and a hint with how far to shorten. Deliver it as is, don't summarize it.

Failure: shaped as {"error": "...", "hint": "...", "field": "..."}.

When an error comes back, the card was not created. Do not tell the user it was.

Don't

• Don't fill a card with content that isn't in memory. A card is a summary of memories.
• Don't create a card the user didn't ask for. "Write this up" is not a card request.
• Don't omit or abbreviate the notice.
• Don't say "I deleted it" or "I turned search off". You can't. Point at the console.
• Don't post a created card's address anywhere on the user's behalf. Posting is the person's decision.
• Don't silently remake when something was truncated without telling the user. Remaking makes two cards.

Worth telling the user alongside

• Paste the address into messengers, Slack, or Notion and it unfolds; for Instagram, upload the image
• To make it findable, turn on search visibility in the console (off by default)
• To remove it, Console → Data → Cards. Even after deleting, preview caches elsewhere can linger for a while

Related

• Searching to gather content → Searching
• What you may store and publish → What to store
• AI rules generally → For AI agents
• Console menus → Console
• The public search page → card.aiakiv.com